diff --git a/.circleci/config.yml b/.circleci/config.yml index 0966da461ec..1462891fa7f 100644 --- a/.circleci/config.yml +++ b/.circleci/config.yml @@ -158,6 +158,8 @@ jobs: CHOCOLATEY_CONFIRM_ALL: "true" - run: name: Install Dependencies + environment: + UV_HTTP_TIMEOUT: "300" command: | $installer = Join-Path $env:TEMP "uv-install.ps1" Invoke-WebRequest -Uri https://astral.sh/uv/0.10.9/install.ps1 -OutFile $installer @@ -228,7 +230,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv \ - --cov=litellm \ + --cov=./litellm \ --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=20 \ @@ -293,7 +295,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv \ - --cov=litellm \ + --cov=./litellm \ --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=20 \ @@ -409,14 +411,25 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -v -x \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5 \ -n 2" no_output_timeout: 15m + - run: + name: Rename the coverage files + command: | + mv coverage.xml auth_ui_unit_tests_coverage.xml + mv .coverage auth_ui_unit_tests_coverage # Store test results - store_test_results: path: test-results + - persist_to_workspace: + root: . + paths: + - auth_ui_unit_tests_coverage.xml + - auth_ui_unit_tests_coverage litellm_router_testing: # Runs all tests with the "router" keyword docker: @@ -493,13 +506,24 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -v -x \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5 \ -n 4" no_output_timeout: 15m + - run: + name: Rename the coverage files + command: | + mv coverage.xml router_unit_tests_coverage.xml + mv .coverage router_unit_tests_coverage # Store test results - store_test_results: path: test-results + - persist_to_workspace: + root: . + paths: + - router_unit_tests_coverage.xml + - router_unit_tests_coverage litellm_assistants_api_testing: # Runs all tests with the "assistants" keyword docker: - *python312_image @@ -603,7 +627,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=20 \ -n 4 \ @@ -646,7 +670,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv -x -s \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5" no_output_timeout: 15m @@ -688,7 +712,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5 \ -n 2 \ @@ -732,7 +756,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv -x -s \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5 \ --retries 3 --retry-delay 5" @@ -814,7 +838,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv -x \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5 \ -n 4" @@ -856,7 +880,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv -x \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5 \ -n 4" @@ -930,7 +954,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv -x -s \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5 \ -n 2" @@ -972,7 +996,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv -x -s \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5 \ -n 2" @@ -1015,7 +1039,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv -x \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5 \ -n 4" @@ -1090,7 +1114,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ -n 4 \ --junitxml=test-results/junit.xml \ --durations=5 \ @@ -1133,7 +1157,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv -x -s \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5" no_output_timeout: 15m @@ -1184,7 +1208,7 @@ jobs: --verbose \ --command="awk '/\\.py/ {print; next} {sub(/\\.[A-Z][^.]*$/, \"\"); gsub(/\\./, \"/\"); print \$0 \".py\"}' | xargs uv run --no-sync python -m pytest \ -vv -x -s \ - --cov=litellm --cov-report=xml \ + --cov=./litellm --cov-report=xml \ --junitxml=test-results/junit.xml \ --durations=5 -n 2 \ --reruns 2 --reruns-delay 1" @@ -2280,10 +2304,11 @@ jobs: - run: name: Combine Coverage command: | - uv tool run --from 'coverage[toml]==7.10.6' coverage combine realtime_translation_coverage ocr_coverage search_coverage logging_coverage audio_coverage local_testing_part1_coverage local_testing_part2_coverage pass_through_unit_tests_coverage batches_coverage guardrails_coverage redis_caching_coverage + uv tool run --from 'coverage[toml]==7.10.6' coverage combine realtime_translation_coverage ocr_coverage search_coverage logging_coverage audio_coverage local_testing_part1_coverage local_testing_part2_coverage pass_through_unit_tests_coverage batches_coverage guardrails_coverage redis_caching_coverage agent_coverage google_generate_content_endpoint_coverage litellm_utils_coverage router_unit_tests_coverage auth_ui_unit_tests_coverage uv tool run --from 'coverage[toml]==7.10.6' coverage xml - codecov/upload: file: ./coverage.xml + flags: circleci ui_build: docker: @@ -2375,6 +2400,11 @@ jobs: environment: DATABASE_URL: "postgresql://e2euser:e2epassword@localhost:5432/litellm_e2e" CI: "true" + # Boot the proxy with an external logout URL so proxyLogoutUrl.spec.ts can + # assert the redirect. Set at job level so both the proxy boot step and the + # Playwright step (whose skip guard reads this) see the same value. Safe for + # the rest of the suite: nothing else performs a logout. + PROXY_LOGOUT_URL: "https://www.example.com" steps: - checkout - setup_google_dns @@ -2451,11 +2481,17 @@ jobs: MOCK_LLM_URL: "http://127.0.0.1:8090/v1" DISABLE_SCHEMA_UPDATE: "true" SERVER_ROOT_PATH: "" - PROXY_LOGOUT_URL: "" + # PROXY_LOGOUT_URL is inherited from the job-level environment so the + # proxy and proxyLogoutUrl.spec.ts agree on the logout target. + # LITELLM_LICENSE is forwarded from the project env so premium-gated + # UI flows can be exercised. license.spec.ts asserts the resulting + # JWT carries premium_user=true; if it ever stops being passed, that + # test fails loudly rather than silently regressing premium coverage. command: | - uv run --no-sync python -m litellm.proxy.proxy_cli \ - --config ui/litellm-dashboard/e2e_tests/fixtures/config.yml \ - --port 4000 + LITELLM_LICENSE="$LITELLM_LICENSE" \ + uv run --no-sync python -m litellm.proxy.proxy_cli \ + --config ui/litellm-dashboard/e2e_tests/fixtures/config.yml \ + --port 4000 background: true - run: name: Wait for proxy to be ready @@ -2472,9 +2508,12 @@ jobs: exit 1 - run: name: Run Playwright E2E tests + # Forward LITELLM_LICENSE so license.spec.ts can detect that the + # proxy was launched with a license and assert premium_user=true. command: | cd ui/litellm-dashboard - npx playwright test --config e2e_tests/playwright.config.ts + LITELLM_LICENSE="$LITELLM_LICENSE" \ + npx playwright test --config e2e_tests/playwright.config.ts no_output_timeout: 10m - store_artifacts: path: ui/litellm-dashboard/test-results @@ -2508,7 +2547,6 @@ jobs: paths: - litellm-docker-database.tar.zst - test_bad_database_url: machine: image: ubuntu-2204:2024.04.1 @@ -2669,6 +2707,8 @@ workflows: - local_testing_part1 - local_testing_part2 - litellm_assistants_api_testing + - litellm_router_unit_testing + - auth_ui_unit_tests - db_migration_disable_update_check: requires: - build_docker_database_image diff --git a/.github/actions/helm-oci-chart-releaser/action.yml b/.github/actions/helm-oci-chart-releaser/action.yml deleted file mode 100644 index 454c591d436..00000000000 --- a/.github/actions/helm-oci-chart-releaser/action.yml +++ /dev/null @@ -1,94 +0,0 @@ -name: Helm OCI Chart Releaser -description: Push Helm charts to OCI-based (Docker) registries -author: sergeyshaykhullin -branding: - color: yellow - icon: upload-cloud -inputs: - name: - required: true - description: Chart name - repository: - required: true - description: Chart repository name - tag: - required: true - description: Chart version - app_version: - required: true - description: App version - path: - required: false - description: Chart path (Default 'charts/{name}') - registry: - required: true - description: OCI registry - registry_username: - required: true - description: OCI registry username - registry_password: - required: true - description: OCI registry password - update_dependencies: - required: false - default: 'false' - description: Update chart dependencies before packaging (Default 'false') -outputs: - image: - value: ${{ steps.output.outputs.image }} - description: Chart image (Default '{registry}/{repository}/{image}:{tag}') -runs: - using: composite - steps: - - name: Helm | Setup - uses: azure/setup-helm@1a275c3b69536ee54be43f2070a358922e12c8d4 # v4.3.1 - with: - version: v3.20.0 - - - name: Helm | Login - shell: bash - env: - REGISTRY_PASSWORD: ${{ inputs.registry_password }} - REGISTRY_USERNAME: ${{ inputs.registry_username }} - REGISTRY: ${{ inputs.registry }} - run: echo "$REGISTRY_PASSWORD" | helm registry login -u "$REGISTRY_USERNAME" --password-stdin "$REGISTRY" - - - name: Helm | Dependency - if: inputs.update_dependencies == 'true' - shell: bash - env: - CHART_PATH: ${{ inputs.path == null && format('{0}/{1}', 'charts', inputs.name) || inputs.path }} - run: helm dependency update "$CHART_PATH" - - - name: Helm | Package - shell: bash - env: - CHART_PATH: ${{ inputs.path == null && format('{0}/{1}', 'charts', inputs.name) || inputs.path }} - TAG: ${{ inputs.tag }} - APP_VERSION: ${{ inputs.app_version }} - run: helm package "$CHART_PATH" --version "$TAG" --app-version "$APP_VERSION" - - - name: Helm | Push - shell: bash - env: - NAME: ${{ inputs.name }} - TAG: ${{ inputs.tag }} - REGISTRY: ${{ inputs.registry }} - REPOSITORY: ${{ inputs.repository }} - run: helm push "${NAME}-${TAG}.tgz" "oci://${REGISTRY}/${REPOSITORY}" - - - name: Helm | Logout - shell: bash - env: - REGISTRY: ${{ inputs.registry }} - run: helm registry logout "$REGISTRY" - - - name: Helm | Output - id: output - shell: bash - env: - REGISTRY: ${{ inputs.registry }} - REPOSITORY: ${{ inputs.repository }} - NAME: ${{ inputs.name }} - TAG: ${{ inputs.tag }} - run: echo "image=${REGISTRY}/${REPOSITORY}/${NAME}:${TAG}" >> $GITHUB_OUTPUT diff --git a/.github/pull_request_template.md b/.github/pull_request_template.md index f9ce9e5dcb8..99f79c0b272 100644 --- a/.github/pull_request_template.md +++ b/.github/pull_request_template.md @@ -10,9 +10,9 @@ **Please complete all items before asking a LiteLLM maintainer to review your PR** -- [ ] I have Added testing in the [`tests/test_litellm/`](https://github.com/BerriAI/litellm/tree/main/tests/test_litellm) directory, **Adding at least 1 test is a hard requirement** - [see details](https://docs.litellm.ai/docs/extras/contributing_code) +- [ ] I have added meaningful tests - [ ] My PR passes all unit tests on [`make test-unit`](https://docs.litellm.ai/docs/extras/contributing_code) -- [ ] My PR's scope is as isolated as possible, it only solves 1 specific problem +- [ ] My PR's scope is as isolated as possible; it only solves 1 specific problem - [ ] I have requested a Greptile review by commenting `@greptileai` and received a **Confidence Score of at least 4/5** before requesting a maintainer review ## Delays in PR merge? diff --git a/.github/workflows/README.md b/.github/workflows/README.md deleted file mode 100644 index b4e777969d9..00000000000 --- a/.github/workflows/README.md +++ /dev/null @@ -1,35 +0,0 @@ -# Simple PyPI Publishing - -A GitHub workflow to manually publish LiteLLM packages to PyPI with a specified version. - -## How to Use - -1. Go to the **Actions** tab in the GitHub repository -2. Select **Simple PyPI Publish** from the workflow list -3. Click **Run workflow** -4. Enter the version to publish (e.g., `1.74.10`) - -## What the Workflow Does - -1. **Updates** the version in `pyproject.toml` -2. **Copies** the model prices backup file -3. **Builds** the Python package -4. **Publishes** to PyPI - -## Prerequisites - -Make sure the following secret is configured in the repository: -- `PYPI_PUBLISH_PASSWORD`: PyPI API token for authentication - -## Example Usage - -- Version: `1.74.11` → Publishes as v1.74.11 -- Version: `1.74.10-hotfix1` → Publishes as v1.74.10-hotfix1 - -## Features - -- ✅ Manual trigger with version input -- ✅ Automatic version updates in `pyproject.toml` -- ✅ Repository safety check (only runs on official repo) -- ✅ Clean package building and publishing -- ✅ Success confirmation with PyPI package link \ No newline at end of file diff --git a/.github/workflows/_test-unit-base.yml b/.github/workflows/_test-unit-base.yml index 9377cbeb0ca..7e91341ac77 100644 --- a/.github/workflows/_test-unit-base.yml +++ b/.github/workflows/_test-unit-base.yml @@ -91,7 +91,7 @@ jobs: --reruns-delay 1 \ --dist=loadscope \ --durations=20 \ - --cov=litellm \ + --cov=./litellm \ --cov-report=xml:coverage.xml \ --cov-config=pyproject.toml @@ -132,4 +132,5 @@ jobs: use_oidc: true directory: coverage-reports root_dir: ${{ github.workspace }} + flags: ${{ inputs.artifact-name }} fail_ci_if_error: false diff --git a/.github/workflows/_test-unit-services-base.yml b/.github/workflows/_test-unit-services-base.yml index 8c47b6d7666..7f973d8cafa 100644 --- a/.github/workflows/_test-unit-services-base.yml +++ b/.github/workflows/_test-unit-services-base.yml @@ -132,7 +132,7 @@ jobs: --reruns "${RERUNS}" \ --reruns-delay 1 \ --durations=20 \ - --cov=litellm \ + --cov=./litellm \ --cov-report=xml:coverage.xml \ --cov-config=pyproject.toml else @@ -144,7 +144,7 @@ jobs: --reruns-delay 1 \ --dist="${DIST}" \ --durations=20 \ - --cov=litellm \ + --cov=./litellm \ --cov-report=xml:coverage.xml \ --cov-config=pyproject.toml fi @@ -186,4 +186,5 @@ jobs: use_oidc: true directory: coverage-reports root_dir: ${{ github.workspace }} + flags: ${{ inputs.artifact-name }} fail_ci_if_error: false diff --git a/.github/workflows/codeql.yml b/.github/workflows/codeql.yml index e86fca17c7a..babe3b62933 100644 --- a/.github/workflows/codeql.yml +++ b/.github/workflows/codeql.yml @@ -53,3 +53,31 @@ jobs: uses: github/codeql-action/analyze@ebcb5b36ded6beda4ceefea6a8bc4cc885255bb3 # v3 with: category: "/language:${{ matrix.language }}" + output: sarif-results + upload: failure-only + + # py/weak-sensitive-data-hashing (CWE-328) fires on the OCI signing call at + # litellm/llms/oci/common_utils.py, which hashes the HTTP request body to + # produce the x-content-sha256 header required by the OCI HTTP signing spec — + # a content-integrity hash, not a password or secret hash. SHA-256 is mandated + # by Oracle for this header; see + # https://docs.oracle.com/en-us/iaas/Content/API/Concepts/signingrequests.htm + # The `usedforsecurity=False` flag on the hashlib.sha256 call already declares + # non-security intent, but CodeQL's taint flow still re-fires when callers + # further up the stack are modified. The suppression is scoped to this one + # file/rule pair via SARIF post-filtering so every other callsite of + # py/weak-sensitive-data-hashing in the repository continues to be analyzed. + - name: Filter SARIF (OCI sha256) + if: matrix.language == 'python' + uses: advanced-security/filter-sarif@2da736ff05ef065cb2894ac6892e47b5eac2c3c0 # v1.1 + with: + patterns: | + -litellm/llms/oci/common_utils.py:py/weak-sensitive-data-hashing + input: sarif-results/python.sarif + output: sarif-results/python.sarif + + - name: Upload SARIF + uses: github/codeql-action/upload-sarif@ebcb5b36ded6beda4ceefea6a8bc4cc885255bb3 # v3 + with: + sarif_file: sarif-results + category: "/language:${{ matrix.language }}" diff --git a/.github/workflows/create_daily_oss_agent_shin_branch.yml b/.github/workflows/create_daily_oss_agent_shin_branch.yml new file mode 100644 index 00000000000..d6118f3b53c --- /dev/null +++ b/.github/workflows/create_daily_oss_agent_shin_branch.yml @@ -0,0 +1,47 @@ +name: Create Daily oss-agent-shin Branch + +on: + schedule: + - cron: "0 0 * * *" # Runs every day at midnight UTC + workflow_dispatch: # Allow manual trigger + +jobs: + create-oss-agent-shin-branch: + if: github.repository == 'BerriAI/litellm' + runs-on: ubuntu-latest + permissions: + contents: write + + steps: + - name: Checkout repository + uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 + with: + fetch-depth: 0 + persist-credentials: false + + - name: Create daily oss-agent-shin branch + env: + GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }} + run: | + # Configure Git user + git config user.name "github-actions[bot]" + git config user.email "github-actions[bot]@users.noreply.github.com" + + # Generate branch name with MM_DD_YYYY format + BRANCH_NAME="litellm_oss_agent_shin_$(date +'%m_%d_%Y')" + echo "Creating branch: $BRANCH_NAME" + + # Fetch all branches + git fetch --all + + # Check if the branch already exists + if git show-ref --verify --quiet refs/remotes/origin/$BRANCH_NAME; then + echo "Branch $BRANCH_NAME already exists. Skipping creation." + else + echo "Creating new branch: $BRANCH_NAME" + # Create the new branch from main + git checkout -b $BRANCH_NAME origin/main + # Push the new branch + git push origin $BRANCH_NAME + echo "Successfully created and pushed branch: $BRANCH_NAME" + fi diff --git a/.github/workflows/llm-translation-testing.yml b/.github/workflows/llm-translation-testing.yml deleted file mode 100644 index 8d9d52f4e58..00000000000 --- a/.github/workflows/llm-translation-testing.yml +++ /dev/null @@ -1,92 +0,0 @@ -name: LLM Translation Tests - -on: - workflow_dispatch: - inputs: - release_candidate_tag: - description: "Release candidate tag/version" - required: true - type: string - push: - tags: - - "v*-rc*" # Triggers on release candidate tags like v1.0.0-rc1 - -permissions: - contents: read - -jobs: - run-llm-translation-tests: - runs-on: ubuntu-latest - timeout-minutes: 90 - - steps: - - name: Checkout code - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 - with: - persist-credentials: false - ref: ${{ github.event.inputs.release_candidate_tag || github.ref }} - - - name: Set up Python - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0 - with: - python-version: "3.12" - - - name: Set up uv - uses: astral-sh/setup-uv@37802adc94f370d6bfd71619e3f0bf239e1f3b78 # v7 - with: - version: "0.10.9" - enable-cache: false - - - name: Restore uv dependencies cache - uses: actions/cache/restore@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0 - with: - path: | - ~/.cache/uv - .venv - key: ${{ runner.os }}-uv-${{ hashFiles('uv.lock') }} - restore-keys: | - ${{ runner.os }}-uv- - - - name: Install dependencies - run: | - uv sync --frozen - - - name: Create test results directory - run: mkdir -p test-results - - - name: Run LLM Translation Tests - env: - OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }} - ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }} - COHERE_API_KEY: ${{ secrets.COHERE_API_KEY }} - GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }} - AZURE_API_KEY: ${{ secrets.AZURE_API_KEY }} - AZURE_API_BASE: ${{ secrets.AZURE_API_BASE }} - AZURE_API_VERSION: ${{ secrets.AZURE_API_VERSION }} - RC_TAG: ${{ github.event.inputs.release_candidate_tag || github.ref_name }} - COMMIT_SHA: ${{ github.sha }} - run: | - python .github/workflows/run_llm_translation_tests.py \ - --tag "$RC_TAG" \ - --commit "$COMMIT_SHA" \ - || true # Continue even if tests fail - - - name: Display test summary - if: always() - run: | - if [ -f "test-results/llm_translation_report.md" ]; then - echo "Test report generated successfully!" - echo "Artifact will contain:" - echo "- test-results/junit.xml (JUnit XML results)" - echo "- test-results/llm_translation_report.md (Beautiful markdown report)" - else - echo "Warning: Test report was not generated" - fi - - - name: Upload test artifacts - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2 - if: always() - with: - name: LLM-Translation-Artifact-${{ github.event.inputs.release_candidate_tag || github.ref_name }} - path: test-results/ - retention-days: 30 diff --git a/.github/workflows/publish_to_pypi.yml b/.github/workflows/publish_to_pypi.yml deleted file mode 100644 index d60254a0ac5..00000000000 --- a/.github/workflows/publish_to_pypi.yml +++ /dev/null @@ -1,153 +0,0 @@ -name: Publish to PyPI - -on: - workflow_dispatch: - -jobs: - preflight-checks: - name: Preflight Checks - runs-on: ubuntu-latest - timeout-minutes: 10 - permissions: - contents: read - # No environment — read-only checks, no approval needed - outputs: - needs_publish: ${{ steps.check-litellm.outputs.needs_publish }} - version: ${{ steps.check-litellm.outputs.version }} - - steps: - - name: Checkout repo - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 - - - name: Set up Python - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0 - with: - python-version: "3.12" - - - name: Set up uv - uses: astral-sh/setup-uv@37802adc94f370d6bfd71619e3f0bf239e1f3b78 # v7 - with: - version: "0.10.9" - enable-cache: false - - - name: Check litellm version on PyPI - id: check-litellm - run: | - VERSION=$(python - <<'PY' - import tomllib - - with open("pyproject.toml", "rb") as f: - print(tomllib.load(f)["project"]["version"]) - PY - ) - echo "version=$VERSION" >> "$GITHUB_OUTPUT" - echo "Checking if litellm $VERSION exists on PyPI..." - - HTTP_STATUS=$(curl -s -o /dev/null -w "%{http_code}" "https://pypi.org/pypi/litellm/$VERSION/json") - if [ "$HTTP_STATUS" = "200" ]; then - echo "litellm $VERSION already exists on PyPI. Skipping publish." - echo "needs_publish=false" >> "$GITHUB_OUTPUT" - else - echo "litellm $VERSION not found on PyPI. Publish needed." - echo "needs_publish=true" >> "$GITHUB_OUTPUT" - fi - - - name: Sanity check proxy-extras version - run: | - # Read pinned version from project optional dependencies - PYPROJECT_VERSION=$(python3 - <<'PY' - import sys - import tomllib - - with open("pyproject.toml", "rb") as f: - proxy_requirements = tomllib.load(f)["project"]["optional-dependencies"]["proxy"] - - version = None - for requirement in proxy_requirements: - normalized = requirement.split(";", 1)[0].strip() - if not normalized.startswith("litellm-proxy-extras"): - continue - parts = normalized.split("==", 1) - if len(parts) == 2 and parts[0].strip() == "litellm-proxy-extras": - candidate = parts[1].strip() - if candidate: - version = candidate - break - - if version is None: - print( - "::error::Could not find an exact litellm-proxy-extras pin in project.optional-dependencies.proxy", - file=sys.stderr, - ) - sys.exit(1) - - print(version) - PY - ) - echo "pyproject.toml pins litellm-proxy-extras version: $PYPROJECT_VERSION" - - # Check that the pinned version exists on PyPI - echo "Checking if litellm-proxy-extras $PYPROJECT_VERSION exists on PyPI..." - HTTP_STATUS=$(curl -s -o /dev/null -w "%{http_code}" "https://pypi.org/pypi/litellm-proxy-extras/$PYPROJECT_VERSION/json") - if [ "$HTTP_STATUS" != "200" ]; then - echo "::error::litellm-proxy-extras $PYPROJECT_VERSION is not published on PyPI yet. Publish it before releasing litellm." - exit 1 - fi - echo "litellm-proxy-extras $PYPROJECT_VERSION exists on PyPI. Sanity check passed." - - publish-litellm: - name: Publish litellm to PyPI - needs: preflight-checks - if: needs.preflight-checks.outputs.needs_publish == 'true' - runs-on: ubuntu-latest - timeout-minutes: 10 - permissions: - id-token: write - contents: read - environment: pypi-publish - - steps: - - name: Checkout repo - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 - - - name: Set up Python - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0 - with: - python-version: "3.12" - - - name: Set up uv - uses: astral-sh/setup-uv@37802adc94f370d6bfd71619e3f0bf239e1f3b78 # v7 - with: - version: "0.10.9" - enable-cache: false - - - name: Copy model prices backup - run: cp model_prices_and_context_window.json litellm/model_prices_and_context_window_backup.json - - - name: Build package - run: | - rm -rf build dist - uv build - - - name: Verify build artifacts - env: - EXPECTED_VERSION: ${{ needs.preflight-checks.outputs.version }} - run: | - echo "Contents of dist/:" - ls -la dist/ - # Ensure we have both sdist and wheel - ls dist/*.tar.gz - ls dist/*.whl - # Verify built version matches expected - ls dist/ | grep -q "litellm-${EXPECTED_VERSION}" || { - echo "::error::Built artifacts do not match expected version $EXPECTED_VERSION" - ls dist/ - exit 1 - } - - - name: Validate package metadata - run: | - uv tool run --from 'twine==6.2.0' twine check dist/* - - - name: Publish to PyPI - uses: pypa/gh-action-pypi-publish@ed0c53931b1dc9bd32cbe73a98c7f6766f8a527e # v1.13.0 diff --git a/.github/workflows/read_pyproject_version.yml b/.github/workflows/read_pyproject_version.yml deleted file mode 100644 index 04b4a38ce19..00000000000 --- a/.github/workflows/read_pyproject_version.yml +++ /dev/null @@ -1,28 +0,0 @@ -name: Read Version from pyproject.toml - -on: - push: - branches: - - main # Change this to the default branch of your repository - -permissions: - contents: read - -jobs: - read-version: - runs-on: ubuntu-latest - - steps: - - name: Checkout code - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 - with: - persist-credentials: false - - - name: Read version from pyproject.toml - id: read-version - run: | - version=$(grep -m1 '^version' pyproject.toml | sed 's/version = "\(.*\)"/\1/') - printf "LITELLM_VERSION=%s" "$version" >> $GITHUB_ENV - - - name: Display version - run: echo "Current version is $LITELLM_VERSION" diff --git a/.github/workflows/results_stats.csv b/.github/workflows/results_stats.csv deleted file mode 100644 index bcef047b0fb..00000000000 --- a/.github/workflows/results_stats.csv +++ /dev/null @@ -1,27 +0,0 @@ -Date,"Ben -Ashley",Tom Brooks,Jimmy Cooney,"Sue -Daniels",Berlinda Fong,Terry Jones,Angelina Little,Linda Smith -10/1,FALSE,TRUE,TRUE,TRUE,TRUE,TRUE,FALSE,TRUE -10/2,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/3,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/4,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/5,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/6,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/7,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/8,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/9,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/10,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/11,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/12,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/13,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/14,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/15,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/16,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/17,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/18,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/19,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/20,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/21,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/22,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -10/23,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE,FALSE -Total,0,1,1,1,1,1,0,1 \ No newline at end of file diff --git a/.github/workflows/run_observatory_tests.yml b/.github/workflows/run_observatory_tests.yml deleted file mode 100644 index a25b96766d7..00000000000 --- a/.github/workflows/run_observatory_tests.yml +++ /dev/null @@ -1,229 +0,0 @@ -name: Run Observatory Tests -on: - workflow_dispatch: - inputs: - tag: - description: "Docker image tag to test (e.g. v1.61.0.rc1)" - required: true - type: string - commit_hash: - description: "Commit hash (defaults to HEAD of current branch)" - required: false - type: string - workflow_call: - inputs: - tag: - description: "Docker image tag to test" - required: true - type: string - commit_hash: - description: "Commit hash of the release" - required: true - type: string - -permissions: - contents: read - -env: - LITELLM_MASTER_KEY: ${{ secrets.LITELLM_MASTER_KEY_STAGING }} - -jobs: - observatory-tests: - runs-on: ubuntu-latest - timeout-minutes: 30 - steps: - - name: Checkout repository - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 - with: - persist-credentials: false - - - name: Validate tag input - env: - TAG: ${{ inputs.tag }} - run: | - if [[ ! "$TAG" =~ ^v[0-9]+\.[0-9]+\.[0-9]+ ]]; then - echo "Invalid tag format: $TAG (expected vX.Y.Z...)" - exit 1 - fi - - - name: Start LiteLLM container - env: - TAG: ${{ inputs.tag }} - AZURE_API_KEY: ${{ secrets.AZURE_API_KEY }} - AZURE_API_BASE: ${{ secrets.AZURE_API_BASE }} - WORKSPACE: ${{ github.workspace }} - run: | - docker run -d \ - --name litellm-rc \ - -p 4000:4000 \ - -v "${WORKSPACE}/.github/observatory/litellm_config.yaml:/app/config.yaml" \ - -e LITELLM_MASTER_KEY="${LITELLM_MASTER_KEY}" \ - -e AZURE_API_KEY="${AZURE_API_KEY}" \ - -e AZURE_API_BASE="${AZURE_API_BASE}" \ - "litellm/litellm:${TAG}" \ - --config /app/config.yaml --port 4000 - - - name: Wait for LiteLLM health check - run: | - echo "Waiting for LiteLLM to be ready..." - for i in $(seq 1 30); do - if curl -s -f http://localhost:4000/health/liveliness > /dev/null 2>&1; then - echo "LiteLLM is healthy" - exit 0 - fi - echo "Attempt $i/30 - not ready yet, waiting 10s..." - sleep 10 - done - echo "LiteLLM failed to start within 5 minutes" - docker logs litellm-rc - exit 1 - - - name: Start cloudflared tunnel - run: | - # Install cloudflared (pinned version + checksum) - curl -sL https://github.com/cloudflare/cloudflared/releases/download/2025.2.1/cloudflared-linux-amd64 -o /usr/local/bin/cloudflared - echo "afdfadd1ef552e66bffc35246fe30a9bd578356d2d386de95585ccfc432472b8 /usr/local/bin/cloudflared" | sha256sum -c - - chmod +x /usr/local/bin/cloudflared - - # Start a quick tunnel (no account needed) and capture the URL - cloudflared tunnel --url http://localhost:4000 --no-autoupdate > /tmp/cloudflared.log 2>&1 & - CLOUDFLARED_PID=$! - echo "CLOUDFLARED_PID=$CLOUDFLARED_PID" >> $GITHUB_ENV - - # Wait for tunnel URL to appear in logs - echo "Waiting for tunnel URL..." - for i in $(seq 1 30); do - TUNNEL_URL=$(grep -oP 'https://[a-z0-9-]+\.trycloudflare\.com' /tmp/cloudflared.log | head -1 || true) - if [ -n "$TUNNEL_URL" ]; then - echo "Tunnel URL: $TUNNEL_URL" - echo "TUNNEL_URL=$TUNNEL_URL" >> $GITHUB_ENV - exit 0 - fi - sleep 2 - done - echo "Failed to get tunnel URL" - cat /tmp/cloudflared.log - exit 1 - - - name: Verify tunnel connectivity - run: | - echo "Testing tunnel at ${TUNNEL_URL}..." - # Quick tunnels need time for DNS propagation; retry to avoid - # transient NXDOMAIN (curl exit code 6) on first attempt. - for i in $(seq 1 10); do - if curl -sf "${TUNNEL_URL}/health/liveliness" > /dev/null 2>&1; then - echo "Tunnel is working (attempt $i)" - exit 0 - fi - echo "Attempt $i/10 - tunnel not routable yet, waiting 5s..." - sleep 5 - done - echo "Tunnel failed to become reachable after 50s" - cat /tmp/cloudflared.log - exit 1 - - - name: Trigger observatory test run - id: trigger - env: - OBSERVATORY_URL: ${{ secrets.OBSERVATORY_URL }} - OBSERVATORY_API_KEY: ${{ secrets.OBSERVATORY_API_KEY }} - run: | - PAYLOAD=$(jq -n \ - --arg url "${TUNNEL_URL}" \ - --arg key "${LITELLM_MASTER_KEY}" \ - '{ - deployment_url: $url, - api_key: $key, - test_suite: "TestOAIAzureRelease", - models: ["gpt-4o-mini", "gpt-4o"] - }') - RESPONSE=$(curl -s -w "\n%{http_code}" -X POST "${OBSERVATORY_URL}/run-test" \ - -H "Content-Type: application/json" \ - -H "X-LiteLLM-Observatory-API-Key: ${OBSERVATORY_API_KEY}" \ - -d "$PAYLOAD") - HTTP_CODE=$(echo "$RESPONSE" | tail -1) - BODY=$(echo "$RESPONSE" | head -n -1) - echo "Response ($HTTP_CODE): $BODY" - if [ "$HTTP_CODE" -ge 400 ]; then - echo "Failed to trigger test run" - exit 1 - fi - - # Extract request_id for polling this specific run - REQUEST_ID=$(echo "$BODY" | jq -r '.results.request_id') - if [ -z "$REQUEST_ID" ] || [ "$REQUEST_ID" = "null" ]; then - echo "Failed to extract request_id from response" - exit 1 - fi - echo "Request ID: $REQUEST_ID" - echo "request_id=$REQUEST_ID" >> $GITHUB_OUTPUT - - - name: Poll for test completion - id: poll - env: - OBSERVATORY_URL: ${{ secrets.OBSERVATORY_URL }} - OBSERVATORY_API_KEY: ${{ secrets.OBSERVATORY_API_KEY }} - REQUEST_ID: ${{ steps.trigger.outputs.request_id }} - run: | - TIMEOUT=900 # 15 minutes - INTERVAL=30 - ELAPSED=0 - while [ $ELAPSED -lt $TIMEOUT ]; do - STATUS=$(curl -s "${OBSERVATORY_URL}/run-status/${REQUEST_ID}" \ - -H "X-LiteLLM-Observatory-API-Key: ${OBSERVATORY_API_KEY}") - RUN_STATUS=$(echo "$STATUS" | jq -r '.status') - echo "Run status (${ELAPSED}s elapsed): $RUN_STATUS" - - if [ "$RUN_STATUS" = "completed" ] || [ "$RUN_STATUS" = "failed" ]; then - echo "Test finished with status: $RUN_STATUS" - echo "$STATUS" > /tmp/observatory_result.json - exit 0 - fi - - sleep $INTERVAL - ELAPSED=$((ELAPSED + INTERVAL)) - done - echo "Timed out waiting for test to complete after ${TIMEOUT}s" - exit 1 - - - name: Verify test results - run: | - RESULT=$(cat /tmp/observatory_result.json) - echo "Full result: $RESULT" - - STATUS=$(echo "$RESULT" | jq -r '.status') - TEST_PASSED=$(echo "$RESULT" | jq -r '.result.test_passed // false') - FAILURE_RATE=$(echo "$RESULT" | jq -r '.result.failure_rate // "N/A"') - ERROR=$(echo "$RESULT" | jq -r '.error // empty') - - echo "Status: $STATUS" - echo "Test passed: $TEST_PASSED" - echo "Failure rate: $FAILURE_RATE" - - if [ -n "$ERROR" ]; then - echo "Error: $ERROR" - fi - - if [ "$STATUS" = "failed" ]; then - echo "Test run failed" - exit 1 - fi - - if [ "$TEST_PASSED" != "true" ]; then - echo "Tests did not pass (failure rate: $FAILURE_RATE)" - exit 1 - fi - - echo "All tests passed!" - - - name: Print LiteLLM logs on failure - if: failure() - run: | - docker logs litellm-rc 2>/dev/null || true - cat /tmp/cloudflared.log 2>/dev/null || true - - - name: Cleanup - if: always() - run: | - kill "$CLOUDFLARED_PID" 2>/dev/null || true - docker rm -f litellm-rc 2>/dev/null || true diff --git a/.github/workflows/scan_duplicate_issues.yml b/.github/workflows/scan_duplicate_issues.yml deleted file mode 100644 index ab0ac2aa3ac..00000000000 --- a/.github/workflows/scan_duplicate_issues.yml +++ /dev/null @@ -1,48 +0,0 @@ -name: Scan Duplicate Issues (One-Time) - -on: - workflow_dispatch: - inputs: - threshold: - description: "Similarity threshold (0-1)" - required: false - default: "0.85" - close: - description: "Actually close duplicates (false = dry run)" - required: false - type: boolean - default: false - -jobs: - scan: - runs-on: ubuntu-latest - permissions: - issues: write - contents: read - steps: - - name: Checkout scripts - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 - with: - sparse-checkout: .github/scripts - persist-credentials: false - - - name: Set up Python - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0 - with: - python-version: "3.12" - - - name: Scan for duplicate issues - env: - GH_TOKEN: ${{ secrets.GITHUB_TOKEN }} - INPUT_THRESHOLD: ${{ inputs.threshold }} - INPUT_CLOSE: ${{ inputs.close }} - run: | - CLOSE_FLAG="" - if [ "$INPUT_CLOSE" = "true" ]; then - CLOSE_FLAG="--close" - fi - python3 .github/scripts/close_duplicate_issues.py \ - --scan \ - --repo ${{ github.repository }} \ - --threshold "$INPUT_THRESHOLD" \ - $CLOSE_FLAG diff --git a/.github/workflows/test-litellm.yml b/.github/workflows/test-litellm.yml deleted file mode 100644 index 938647f5d0c..00000000000 --- a/.github/workflows/test-litellm.yml +++ /dev/null @@ -1,45 +0,0 @@ -name: LiteLLM Mock Tests (folder - tests/test_litellm) - -# DEPRECATED: This workflow is replaced by test-litellm-matrix.yml which runs -# the same tests in parallel across 10 jobs for faster CI times. -# Kept for manual debugging only. -on: - workflow_dispatch: # Manual trigger only - # pull_request: - # branches: [ main ] - -permissions: - contents: read - -jobs: - test: - runs-on: ubuntu-latest - timeout-minutes: 25 - - steps: - - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0 - with: - persist-credentials: false - - - name: Thank You Message - run: | - echo "### 🙏 Thank you for contributing to LiteLLM!" >> $GITHUB_STEP_SUMMARY - echo "Your PR is being tested now. We appreciate your help in making LiteLLM better!" >> $GITHUB_STEP_SUMMARY - - - name: Set up Python - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0 - with: - python-version: "3.12" - - - name: Set up uv - uses: astral-sh/setup-uv@37802adc94f370d6bfd71619e3f0bf239e1f3b78 # v7 - with: - version: "0.10.9" - - - name: Install dependencies - run: | - uv lock --check - uv sync --frozen --group ci --group proxy-dev --extra google --extra proxy --extra semantic-router - - name: Run tests - run: | - uv run --no-sync pytest tests/test_litellm --tb=short -vv --maxfail=10 -n 4 --durations=50 diff --git a/.github/workflows/test-mcp.yml b/.github/workflows/test-mcp.yml index 313043e12fe..2ae60951afc 100644 --- a/.github/workflows/test-mcp.yml +++ b/.github/workflows/test-mcp.yml @@ -43,4 +43,4 @@ jobs: - name: Run MCP tests run: | - uv run --no-sync pytest tests/mcp_tests -x -vv -n 4 --cov=litellm --cov-report=xml --durations=5 + uv run --no-sync pytest tests/mcp_tests -x -vv -n 4 --cov=./litellm --cov-report=xml --durations=5 diff --git a/.github/workflows/test-unit-caching-redis.yml b/.github/workflows/test-unit-caching-redis.yml deleted file mode 100644 index ca274324f2f..00000000000 --- a/.github/workflows/test-unit-caching-redis.yml +++ /dev/null @@ -1,38 +0,0 @@ -name: "Unit Tests: Caching (Redis)" - -# Uses cloud Redis credentials — only runs on trusted branches, not PRs. -# This prevents external PRs from accessing Redis credentials. -on: - push: - branches: [main, "litellm_*"] - -permissions: - contents: read - -concurrency: - group: ${{ github.workflow }}-${{ github.ref }} - cancel-in-progress: true - -jobs: - caching-redis: - uses: ./.github/workflows/_test-unit-services-base.yml - with: - # Redis-only tests that do NOT require provider API keys. - # Tests needing API keys (test_caching.py, test_caching_ssl.py, test_prometheus_service.py, - # test_router_caching.py) are in Phase 3 integration workflows. - test-path: >- - tests/local_testing/test_dual_cache.py - tests/local_testing/test_redis_batch_optimizations.py - tests/local_testing/test_router_utils.py - workers: 2 - reruns: 2 - timeout-minutes: 20 - enable-redis: true - enable-postgres: false - secrets: - REDIS_HOST: ${{ secrets.REDIS_HOST }} - REDIS_PORT: ${{ secrets.REDIS_PORT }} - REDIS_PASSWORD: ${{ secrets.REDIS_PASSWORD }} - DATABASE_URL: ${{ secrets.DATABASE_URL }} - POSTGRES_USER: ${{ secrets.POSTGRES_USER }} - POSTGRES_PASSWORD: ${{ secrets.POSTGRES_PASSWORD }} diff --git a/.github/workflows/test-unit-proxy-db.yml b/.github/workflows/test-unit-proxy-db.yml index bb44b001947..2d4e85630dc 100644 --- a/.github/workflows/test-unit-proxy-db.yml +++ b/.github/workflows/test-unit-proxy-db.yml @@ -143,6 +143,7 @@ jobs: tests/proxy_unit_tests/test_proxy_pass_user_config.py tests/proxy_unit_tests/test_proxy_token_counter.py tests/proxy_unit_tests/test_request_size_limit_middleware.py + tests/proxy_unit_tests/test_multipart_bypass_repro.py workers: 4 dist: loadscope timeout: 15 @@ -214,8 +215,10 @@ jobs: tests/proxy_unit_tests/test_models_fallback_endpoint.py tests/proxy_unit_tests/test_google_endpoint_routing.py tests/proxy_unit_tests/test_google_gemini_proxy_request.py + tests/proxy_unit_tests/test_gemini_agents_endpoints.py tests/proxy_unit_tests/test_get_favicon.py tests/proxy_unit_tests/test_get_image.py + tests/proxy_unit_tests/test_reducto_ocr_route.py tests/proxy_unit_tests/test_ui_path_detection.py tests/proxy_unit_tests/test_prompt_test_endpoint.py tests/proxy_unit_tests/test_check_batch_cost.py diff --git a/.github/workflows/test-unit-proxy-endpoints.yml b/.github/workflows/test-unit-proxy-endpoints.yml index 1439b2c07f7..6b34a08a8e0 100644 --- a/.github/workflows/test-unit-proxy-endpoints.yml +++ b/.github/workflows/test-unit-proxy-endpoints.yml @@ -7,6 +7,7 @@ on: - litellm_internal_staging - litellm_oss_branch - "litellm_**" + workflow_dispatch: permissions: contents: read @@ -32,6 +33,7 @@ jobs: tests/test_litellm/proxy/image_endpoints tests/test_litellm/proxy/vector_store_endpoints tests/test_litellm/proxy/agent_endpoints + tests/test_litellm/proxy/a2a tests/test_litellm/proxy/discovery_endpoints tests/test_litellm/proxy/health_endpoints tests/test_litellm/proxy/public_endpoints @@ -42,3 +44,16 @@ jobs: workers: 2 reruns: 2 artifact-name: proxy-endpoints + + # Behavior-pinning tests for litellm/proxy/proxy_server.py. Owns its + # own job (not a path on the proxy-endpoints job above) so its budget + # is independent and its coverage artifact is uploaded separately. + # See: https://www.notion.so/36c43b8acdab81ee845fd5365128a2fc + proxy-server: + uses: ./.github/workflows/_test-unit-base.yml + with: + test-path: tests/test_litellm/proxy/proxy_server + workers: 4 + reruns: 2 + timeout-minutes: 60 + artifact-name: proxy-server diff --git a/.github/workflows/test-unit-proxy-mgmt-behavior.yml b/.github/workflows/test-unit-proxy-mgmt-behavior.yml new file mode 100644 index 00000000000..e73997323a4 --- /dev/null +++ b/.github/workflows/test-unit-proxy-mgmt-behavior.yml @@ -0,0 +1,34 @@ +name: "Unit Tests: Proxy Management-Endpoint Behavior Pinning" + +on: + pull_request: + branches: + - main + - litellm_internal_staging + - litellm_oss_branch + - "litellm_**" + +permissions: + contents: read + id-token: write + pull-requests: write + +concurrency: + group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }} + cancel-in-progress: true + +jobs: + proxy-mgmt-behavior: + uses: ./.github/workflows/_test-unit-services-base.yml + with: + test-path: tests/proxy_behavior + # workers=0 (no xdist): the world seed is a single shared Postgres + # state — two xdist workers both call seed_world() and race on the + # ``behavior-pin-budget`` row, producing UniqueViolation + cascading + # missing-membership FK failures. The whole suite is ~7s sequentially, + # so the cost of disabling parallelism here is negligible. + workers: 0 + reruns: 0 + enable-postgres: true + artifact-name: proxy-mgmt-behavior + timeout-minutes: 15 diff --git a/.github/workflows/test_server_root_path.yml b/.github/workflows/test_server_root_path.yml index 155445acdf6..57ff746c9c8 100644 --- a/.github/workflows/test_server_root_path.yml +++ b/.github/workflows/test_server_root_path.yml @@ -101,6 +101,31 @@ jobs: docker logs litellm-test exit 1 + - name: Setup Node for Playwright + uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0 + with: + node-version: "20" + + - name: Install UI deps and Chromium + working-directory: ui/litellm-dashboard + run: | + npm ci + npx playwright install --with-deps chromium + + - name: Run SERVER_ROOT_PATH redirect e2e + working-directory: ui/litellm-dashboard + env: + SERVER_ROOT_PATH: ${{ matrix.root_path }} + run: npx playwright test --config=e2e_tests/serverRootPath.config.ts + + - name: Upload Playwright artifacts on failure + if: failure() + uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2 + with: + name: playwright-trace-${{ strategy.job-index }} + path: ui/litellm-dashboard/test-results/ + retention-days: 7 + - name: Cleanup if: always() run: | diff --git a/.github/workflows/update_release.py b/.github/workflows/update_release.py deleted file mode 100644 index f70509e8e75..00000000000 --- a/.github/workflows/update_release.py +++ /dev/null @@ -1,54 +0,0 @@ -import os -import requests -from datetime import datetime - -# GitHub API endpoints -GITHUB_API_URL = "https://api.github.com" -REPO_OWNER = "BerriAI" -REPO_NAME = "litellm" - -# GitHub personal access token (required for uploading release assets) -GITHUB_ACCESS_TOKEN = os.environ.get("GITHUB_ACCESS_TOKEN") - -# Headers for GitHub API requests -headers = { - "Accept": "application/vnd.github+json", - "Authorization": f"Bearer {GITHUB_ACCESS_TOKEN}", - "X-GitHub-Api-Version": "2022-11-28", -} - -# Get the latest release -releases_url = f"{GITHUB_API_URL}/repos/{REPO_OWNER}/{REPO_NAME}/releases/latest" -response = requests.get(releases_url, headers=headers) -latest_release = response.json() -print("Latest release:", latest_release) - -# Upload an asset to the latest release -upload_url = latest_release["upload_url"].split("{?")[0] -asset_name = "results_stats.csv" -asset_path = os.path.join(os.getcwd(), asset_name) -print("upload_url:", upload_url) - -with open(asset_path, "rb") as asset_file: - asset_data = asset_file.read() - -upload_payload = { - "name": asset_name, - "label": "Load test results", - "created_at": datetime.utcnow().isoformat() + "Z", -} - -upload_headers = headers.copy() -upload_headers["Content-Type"] = "application/octet-stream" - -upload_response = requests.post( - upload_url, - headers=upload_headers, - data=asset_data, - params=upload_payload, -) - -if upload_response.status_code == 201: - print(f"Asset '{asset_name}' uploaded successfully to the latest release.") -else: - print(f"Failed to upload asset. Response: {upload_response.text}") diff --git a/.gitignore b/.gitignore index 20355a8e4ef..572830d35f6 100644 --- a/.gitignore +++ b/.gitignore @@ -28,6 +28,8 @@ litellm/tests/config_*.yaml litellm/tests/langfuse.log langfuse.log .langfuse.log +.pin_list.txt +.cov_new.xml litellm/tests/test_custom_logger.py litellm/tests/langfuse.log litellm/tests/dynamo*.log @@ -101,4 +103,24 @@ STABILIZATION_TODO.md **/*.storageState.json **/coverage test-config -.vscode \ No newline at end of file + +# ---------- Terraform ---------- +# Provider binaries + module cache — regenerated by `terraform init`. +**/.terraform/ +# State files often contain secrets (DB passwords, API keys snapshotted from +# data sources). Keep state in a remote backend, never in git. +*.tfstate +*.tfstate.* +*.tfstate.backup +# Plan files can also contain sensitive values (variables in plaintext). +*.tfplan +# User-specific variable inputs — example files (terraform.tfvars.example) are +# tracked because they end in .example, which doesn't match the glob below. +*.tfvars +*.auto.tfvars +crash.log +crash.*.log +# .terraform.lock.hcl is intentionally NOT ignored — it pins provider versions +# and should be committed. +.vscode +.pin_list.txt diff --git a/AGENTS.md b/AGENTS.md index e99bf79d783..41921fdff4d 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -1,293 +1 @@ -# INSTRUCTIONS FOR LITELLM - -This document provides comprehensive instructions for AI agents working in the LiteLLM repository. - -## OVERVIEW - -LiteLLM is a unified interface for 100+ LLMs that: -- Translates inputs to provider-specific completion, embedding, and image generation endpoints -- Provides consistent OpenAI-format output across all providers -- Includes retry/fallback logic across multiple deployments (Router) -- Offers a proxy server (LLM Gateway) with budgets, rate limits, and authentication -- Supports advanced features like function calling, streaming, caching, and observability - -## REPOSITORY STRUCTURE - -### Core Components -- `litellm/` - Main library code - - `llms/` - Provider-specific implementations (OpenAI, Anthropic, Azure, etc.) - - `proxy/` - Proxy server implementation (LLM Gateway) - - `router_utils/` - Load balancing and fallback logic - - `types/` - Type definitions and schemas - - `integrations/` - Third-party integrations (observability, caching, etc.) - -### Key Directories -- `tests/` - Comprehensive test suites -- `ui/litellm-dashboard/` - Admin dashboard UI -- `enterprise/` - Enterprise-specific features - -Documentation lives in the separate [BerriAI/litellm-docs](https://github.com/BerriAI/litellm-docs) repository and is served at [docs.litellm.ai](https://docs.litellm.ai). - -## DEVELOPMENT GUIDELINES - -### MAKING CODE CHANGES - -1. **Provider Implementations**: When adding/modifying LLM providers: - - Follow existing patterns in `litellm/llms/{provider}/` - - Implement proper transformation classes that inherit from `BaseConfig` - - Support both sync and async operations - - Handle streaming responses appropriately - - Include proper error handling with provider-specific exceptions - -2. **Type Safety**: - - Use proper type hints throughout - - Update type definitions in `litellm/types/` - - Ensure compatibility with both Pydantic v1 and v2 - -3. **Testing**: - - Add tests in appropriate `tests/` subdirectories - - Include both unit tests and integration tests - - Test provider-specific functionality thoroughly - - Consider adding load tests for performance-critical changes - -### MAKING CODE CHANGES FOR THE UI (IGNORE FOR BACKEND) - -1. **Always use `antd` for new UI components — Tremor is DEPRECATED** - - We are migrating off of `@tremor/react`. Do not introduce new `Badge`, `Text`, `Card`, `Grid`, `Title`, or other imports from `@tremor/react` in any new or modified file. - - Use `antd` equivalents: `Tag` for labels, plain ``/`
` with Tailwind classes (or `Typography.Text`) for text, `Card` from `antd`, etc. Note that `antd` has no `"yellow"` Tag color — use `"gold"` for amber/yellow. - - The only exception is the Tremor Table component and its required Tremor Table sub components. - -2. **Use Common Components as much as possible**: - - These are usually defined in the `common_components` directory - - Use these components as much as possible and avoid building new components unless needed - -3. **Testing**: - - The codebase uses **Vitest** and **React Testing Library** - - **Query Priority Order**: Use query methods in this order: `getByRole`, `getByLabelText`, `getByPlaceholderText`, `getByText`, `getByTestId` - - **Always use `screen`** instead of destructuring from `render()` (e.g., use `screen.getByText()` not `getByText`) - - **Wrap user interactions in `act()`**: Always wrap `fireEvent` calls with `act()` to ensure React state updates are properly handled - - **Use `query` methods for absence checks**: Use `queryBy*` methods (not `getBy*`) when expecting an element to NOT be present - - **Test names must start with "should"**: All test names should follow the pattern `it("should ...")` - - **Mock external dependencies**: Check `setupTests.ts` for global mocks and mock child components/networking calls as needed - - **Structure tests properly**: - - First test should verify the component renders successfully - - Subsequent tests should focus on functionality and user interactions - - Use `waitFor` for async operations that aren't already awaited - - **Avoid using `querySelector`**: Prefer React Testing Library queries over direct DOM manipulation - -### IMPORTANT PATTERNS - -1. **Function/Tool Calling**: - - LiteLLM standardizes tool calling across providers - - OpenAI format is the standard, with transformations for other providers - - See `litellm/llms/anthropic/chat/transformation.py` for complex tool handling - -2. **Streaming**: - - All providers should support streaming where possible - - Use consistent chunk formatting across providers - - Handle both sync and async streaming - -3. **Error Handling**: - - Use provider-specific exception classes - - Maintain consistent error formats across providers - - Include proper retry logic and fallback mechanisms - -4. **Configuration**: - - Support both environment variables and programmatic configuration - - Use `BaseConfig` classes for provider configurations - - Allow dynamic parameter passing - -## PROXY SERVER (LLM GATEWAY) - -The proxy server is a critical component that provides: -- Authentication and authorization -- Rate limiting and budget management -- Load balancing across multiple models/deployments -- Observability and logging -- Admin dashboard UI -- Enterprise features - -Key files: -- `litellm/proxy/proxy_server.py` - Main server implementation -- `litellm/proxy/auth/` - Authentication logic -- `litellm/proxy/management_endpoints/` - Admin API endpoints - -**Database (proxy)**: Use Prisma model methods (`prisma_client.db..upsert`, `.find_many`, `.find_unique`, etc.), not raw SQL (`execute_raw`/`query_raw`). See COMMON PITFALLS for details. - -## MCP (MODEL CONTEXT PROTOCOL) SUPPORT - -LiteLLM supports MCP for agent workflows: -- MCP server integration for tool calling -- Transformation between OpenAI and MCP tool formats -- Support for external MCP servers (Zapier, Jira, Linear, etc.) -- See `litellm/experimental_mcp_client/` and `litellm/proxy/_experimental/mcp_server/` - -## RUNNING SCRIPTS - -Use `uv run python script.py` to run Python scripts in the project environment (for non-test files). - -## GITHUB TEMPLATES - -When opening issues or pull requests, follow these templates: - -### Bug Reports (`.github/ISSUE_TEMPLATE/bug_report.yml`) -- Describe what happened vs. expected behavior -- Include relevant log output -- Specify LiteLLM version -- Indicate if you're part of an ML Ops team (helps with prioritization) - -### Feature Requests (`.github/ISSUE_TEMPLATE/feature_request.yml`) -- Clearly describe the feature -- Explain motivation and use case with concrete examples - -### Pull Requests (`.github/pull_request_template.md`) -- Add at least 1 test in `tests/litellm/` -- Ensure `make test-unit` passes - - -## TESTING CONSIDERATIONS - -1. **Provider Tests**: Test against real provider APIs when possible -2. **Proxy Tests**: Include authentication, rate limiting, and routing tests -3. **Performance Tests**: Load testing for high-throughput scenarios -4. **Integration Tests**: End-to-end workflows including tool calling - -## DOCUMENTATION - -- Keep documentation in sync with code changes -- Update provider documentation when adding new providers -- Include code examples for new features -- Update changelog and release notes - -## SECURITY CONSIDERATIONS - -- Handle API keys securely -- Validate all inputs, especially for proxy endpoints -- Consider rate limiting and abuse prevention -- Follow security best practices for authentication - -## ENTERPRISE FEATURES - -- Some features are enterprise-only -- Check `enterprise/` directory for enterprise-specific code -- Maintain compatibility between open-source and enterprise versions - -## COMMON PITFALLS TO AVOID - -1. **Breaking Changes**: LiteLLM has many users - avoid breaking existing APIs -2. **Provider Specifics**: Each provider has unique quirks - handle them properly -3. **Rate Limits**: Respect provider rate limits in tests -4. **Memory Usage**: Be mindful of memory usage in streaming scenarios -5. **Dependencies**: Keep dependencies minimal and well-justified -6. **UI/Backend Contract Mismatch**: When adding a new entity type to the UI, always check whether the backend endpoint accepts a single value or an array. Match the UI control accordingly (single-select vs. multi-select) to avoid silently dropping user selections -7. **Missing Tests for New Entity Types**: When adding a new entity type (e.g., in `EntityUsage`, `UsageViewSelect`), always add corresponding tests in the existing test files and update any icon/component mocks -8. **Raw SQL in proxy DB code**: Do not use `execute_raw` or `query_raw` for proxy database access. Use Prisma model methods (e.g. `prisma_client.db.litellm_tooltable.upsert()`, `.find_many()`, `.find_unique()`) so behavior stays consistent with the schema, the client stays mockable in tests, and you avoid the pitfalls of hand-written SQL (parameter ordering, type casting, schema drift) - -8. **Do not hardcode model-specific flags**: Put model-specific capability flags in `model_prices_and_context_window.json` and read them via `get_model_info` (or existing helpers like `supports_reasoning`). This prevents users from needing to upgrade LiteLLM each time a new model supports a feature. - - **Example of BAD** (hardcoded model checks): - - ```python - @staticmethod - def _is_effort_supported_model(model: str) -> bool: - """Check if the model supports the output_config.effort parameter...""" - model_lower = model.lower() - if AnthropicConfig._is_claude_4_6_model(model): - return True - return any( - v in model_lower for v in ("opus-4-5", "opus_4_5", "opus-4.5", "opus_4.5") - ) - ``` - - **Example of GOOD** (config-driven or helper that reads from config): - - ```python - if ( - "claude-3-7-sonnet" in model - or AnthropicConfig._is_claude_4_6_model(model) - or supports_reasoning( - model=model, - custom_llm_provider=self.custom_llm_provider, - ) - ): - ... - ``` - - Using helpers like `supports_reasoning` (which read from `model_prices_and_context_window.json` / `get_model_info`) allows future model updates to "just work" without code changes. - -9. **Never close HTTP/SDK clients on cache eviction**: Do not add `close()`, `aclose()`, or `create_task(close_fn())` inside `LLMClientCache._remove_key()` or any cache eviction path. Evicted clients may still be held by in-flight requests; closing them causes `RuntimeError: Cannot send a request, as the client has been closed.` in production after the cache TTL (1 hour) expires. Connection cleanup is handled at shutdown by `close_litellm_async_clients()`. See PR #22247 for the full incident history. - -## HELPFUL RESOURCES - -- Main documentation: https://docs.litellm.ai/ (source: [BerriAI/litellm-docs](https://github.com/BerriAI/litellm-docs)) -- Provider-specific docs: https://docs.litellm.ai/docs/providers/ -- Admin UI for testing proxy features - -## WHEN IN DOUBT - -- Follow existing patterns in the codebase -- Check similar provider implementations -- Ensure comprehensive test coverage -- Update documentation appropriately -- Consider backward compatibility impact - -## Cursor Cloud specific instructions - -### Environment - -- uv is installed in `~/.local/bin`; the update script ensures it is on `PATH`. -- Python 3.12, Node 22 are pre-installed. -- The project virtual environment lives under `.venv/`. - -### Running the proxy server - -Create a minimal config file and start the proxy: - -```yaml -# config.yaml -model_list: - - model_name: fake-openai-endpoint - litellm_params: - model: openai/fake-model - api_key: fake-key - api_base: https://fake-api.example.com - -general_settings: - master_key: sk-1234 - -litellm_settings: - drop_params: True - telemetry: False -``` - -```bash -uv run litellm --config config.yaml --port 4000 -``` - -The proxy takes ~15-20 seconds to fully start (it runs Prisma migrations on boot). Wait for `/health` to return before sending requests. Without a PostgreSQL `DATABASE_URL`, the proxy connects to a default Neon dev database embedded in the `litellm-proxy-extras` package. - -### Running tests - -See `CLAUDE.md` and the `Makefile` for standard commands. Key notes: - -- `uv sync --group proxy-dev --extra proxy` installs the Prisma and proxy-side test dependencies used by the standard local workflow. -- The `--timeout` pytest flag is NOT available; don't pass it. -- Unit tests: `uv run pytest tests/test_litellm/ -x -vv -n 4` -- **Before committing, always run `uv run black .` to format your code.** Black formatting is enforced in CI. -- If `uv sync` fails because the lockfile is outdated, run `uv lock` and retry. - -### Lint - -```bash -cd litellm && uv run ruff check . -``` - -Ruff is the primary fast linter. For the full lint suite (including mypy, black, circular imports), run `make lint` per `CLAUDE.md`. - -### UI Dashboard development - -- The UI is at `ui/litellm-dashboard/`. Run `npm run dev` from that directory for the Next.js dev server on port 3000. -- The proxy at port 4000 serves a **pre-built** static UI from `litellm/proxy/_experimental/out/`. After making UI code changes, you must run `npm run build` in the dashboard directory and copy the output: `cp -r ui/litellm-dashboard/out/* litellm/proxy/_experimental/out/` for the proxy to serve the updated UI. -- SVGs used as provider logos (loaded via `` tags) must NOT use `fill="currentColor"` — replace with an explicit color like `#000000` or use the `-color` variant from lobehub icons, since CSS color inheritance does not work inside `` elements. -- Provider logos live in `ui/litellm-dashboard/public/assets/logos/` (source) and `litellm/proxy/_experimental/out/assets/logos/` (pre-built). Both locations must have the file for it to work in dev and proxy-served modes. -- UI Vitest tests: `cd ui/litellm-dashboard && npx vitest run` +Read @CLAUDE.md for coding guidelines diff --git a/CLAUDE.md b/CLAUDE.md index 938801df7c7..3477b71a621 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -1,180 +1,70 @@ -# CLAUDE.md +Do not write comments unless they are absolutely necessary to explain some very complex business logic. Please clean up if there are comments that are not absolutely necessary. Do not remove comments that are unrelated to the addition of the code of this PR -This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository. +Explanation: code comments are, in a way, a violation of DRY code. You must update logic in two locations to change the code and "hard to change" is literally the definition of tech debt. We should instead aim to write code that is intuitive to the reader, while being both easy to maintain and high performance -## Documentation +Don't assume that the existing code is correct or the right way of doing things / good coding patterns. In fact, there are a lot of bad coding practices, overly complex code, code smells, etc. If something doesn't look right, speak up. Feel free to break existing patterns or question weird existing code to make new code high quality, as in: +- correct +- secure +- performant +- readable +- easy to maintain/change +- modern +In that order of importance -Documentation lives in a separate repository: [BerriAI/litellm-docs](https://github.com/BerriAI/litellm-docs). It is served at [docs.litellm.ai](https://docs.litellm.ai). Do not create or edit documentation files in this repository — open doc PRs against `BerriAI/litellm-docs` instead. +When adding new features, add meaningful tests. Don't add tests that don't check anything substantial and is there just to make the code coverage pass. Yes, code coverage is important, but I'd rather have no signal whether the code is working than tests that don't fail when code is broken. The goal is to have tests that would fail before the feature was added/if the code was mutated in a way that breaks the feature and succeed only when the feature is fully working. I should run mutation testing and see > 90% kill rate -## Development Commands +Same thing for bug fixes. The tests should make it so that this specific bug can never happen again without failing tests (i.e., regression) -### Installation -- `make install-dev` - Install core development dependencies -- `make install-proxy-dev` - Install proxy development dependencies with full feature set -- `make install-test-deps` - Install the full local test environment and generate the Prisma client +When creating PRs, don't set base to `main`. `litellm_internal_staging` serves that purpose -### Testing -- `make test` - Run all tests -- `make test-unit` - Run unit tests (tests/test_litellm) with 4 parallel workers -- `make test-integration` - Run integration tests (excludes unit tests) -- `pytest tests/` - Direct pytest execution +Always use @.github/pull_request_template.md as a guide for your PR body -### Code Quality -- `make lint` - Run all linting (Ruff, MyPy, Black, circular imports, import safety) -- `make format` - Apply Black code formatting -- `make lint-ruff` - Run Ruff linting only -- `make lint-mypy` - Run MyPy type checking only -- **Before committing, always run `uv run black .` to format your code.** Black formatting is enforced in CI. +Never use `pytest` commands or the like as "Screenshots / Proof of Fix". We prefer curl'ing a live proxy instance running on localhost:4000 (I like to run it with `python litellm/proxy/proxy_cli.py --config litellm/proxy/dev_config.yaml --detailed_debug --reload --use_v2_migration_resolver 2>&1 | tee litellm.log`) and showing both the command run and the output. Also, it should hit real LLM provider APIs, not mocks, and cost real $$$ because that is the most realistic test. The proof of fix should be exactly what the end user / customer would see / do. The run logs in PR #27703 is a prime example of how to do it (not a huge fan of using a python test script that future me and the team will have no visibility into; I prefer just curl commands or a short list of bash commands (e.g., using `for`)). If it's a UI thing, just tell me which URLs to go to (e.g., http://localhost:4000/ui/?page=logs), where to click, what fields to fill out, etc. along with the other commands to run in an ordered list, and I'll do it myself and post the screenshots after you make the PR -### Single Test Files -- `uv run pytest tests/path/to/test_file.py -v` - Run specific test file -- `uv run pytest tests/path/to/test_file.py::test_function -v` - Run specific test +If you ever make public-facing PR descriptions, comments, issues, commit messages, etc., always follow these guidelines to sound less AI-y: +- don't use emojis +- don't use "—". Instead, reach for ";", ".", etc. +- don't use the pattern "It's not X, it's Y", "You're not X, you're Y", etc. +- don't use bulleted or numbered lists unless it would be nonsensical not to. Instead, prefer prose -### Running Scripts -- `uv run python script.py` - Run Python scripts (use for non-test files) +Don't hesitate to use values in .env to get needed API keys and other secrets, as long as you never add them to conversation history, commit them, or include them in GitHub issues / PRs -### GitHub Issue & PR Templates -When contributing to the project, use the appropriate templates: +Run tests, format your code, and lint your code before each commit -**Bug Reports** (`.github/ISSUE_TEMPLATE/bug_report.yml`): -- Describe what happened vs. what you expected -- Include relevant log output -- Specify your LiteLLM version +Ask to commit and push your work when you're done (or if you're confident that your code is good and works, just do it) -**Feature Requests** (`.github/ISSUE_TEMPLATE/feature_request.yml`): -- Describe the feature clearly -- Explain the motivation and use case +When you must use real LLM models to, for example, write e2e tests, write a QA runbook, etc., make sure to use the latest models (doesn't have to be smartest, can also be a modern small, fast one. No strong preference for smart vs fast here, just use something modern) as of the year and month of the current date. Do a web search as necessary to figure that out -**Pull Requests** (`.github/pull_request_template.md`): -- Add at least 1 test in `tests/litellm/` -- Ensure `make test-unit` passes +If you're an internal contributor, when creating a new PR, the typical flow is to branch off litellm_internal_staging and create a branch prefixed with litellm_. Do not create a branch prefixed with claude/ and generally do not have / in your branch names -## Architecture Overview +Do not add `Co-Authored-By: Claude` or any Claude attribution to commit messages. Never use a `claude/` prefix or put a `/` in a branch name. Do not add "Generated with Claude Code" (or any similar attribution) to PR descriptions. Do not create a new PR/branch off the existing PR to fix/add something that is related and could've just been committed directly to the existing PR's branch -LiteLLM is a unified interface for 100+ LLM providers with two main components: +When working on a PR, keep the PR description in sync with new commits being made -### Core Library (`litellm/`) -- **Main entry point**: `litellm/main.py` - Contains core completion() function -- **Provider implementations**: `litellm/llms/` - Each provider has its own subdirectory -- **Router system**: `litellm/router.py` + `litellm/router_utils/` - Load balancing and fallback logic -- **Type definitions**: `litellm/types/` - Pydantic models and type hints -- **Integrations**: `litellm/integrations/` - Third-party observability, caching, logging -- **Caching**: `litellm/caching/` - Multiple cache backends (Redis, in-memory, S3, etc.) +Monkeypatching attributes of a class to do testing is an anti-pattern. Prefer dependency-injecting things into classes. That way, at unit test time, you can pass a mocked dependency in -### Proxy Server (`litellm/proxy/`) -- **Main server**: `proxy_server.py` - FastAPI application -- **Authentication**: `auth/` - API key management, JWT, OAuth2 -- **Database**: `db/` - Prisma ORM with PostgreSQL/SQLite support -- **Management endpoints**: `management_endpoints/` - Admin APIs for keys, teams, models -- **Pass-through endpoints**: `pass_through_endpoints/` - Provider-specific API forwarding -- **Guardrails**: `guardrails/` - Safety and content filtering hooks -- **UI Dashboard**: Served from `_experimental/out/` (Next.js build) +Do not put names of customers or customer company names in code, PRs, and issues. The codebase is public -## Key Patterns +CI supply-chain safety: Never pipe a remote script into a shell (`curl ... | bash`, `wget ... | sh`); download the artifact to a file, verify its SHA-256 checksum, then install. Pin every external tool to a specific version with a full URL (not `latest` or `stable`). Verify checksums for all downloaded binaries, using the provider's official `.sha256` / `.sha256sum` sidecar when available. These rules apply to every download in CI -### Provider Implementation -- Providers inherit from base classes in `litellm/llms/base.py` -- Each provider has transformation functions for input/output formatting -- Support both sync and async operations -- Handle streaming responses and function calling +## Think Before Coding -### Error Handling -- Provider-specific exceptions mapped to OpenAI-compatible errors -- Fallback logic handled by Router system -- Comprehensive logging through `litellm/_logging.py` +**Don't assume. Don't hide confusion. Surface tradeoffs.** -### Configuration -- YAML config files for proxy server (see `proxy/example_config_yaml/`) -- Environment variables for API keys and settings -- Database schema managed via Prisma (`proxy/schema.prisma`) +Before implementing: +- State your assumptions explicitly. If uncertain, ask. +- If multiple interpretations exist, present them. Don't pick silently. +- If a simpler approach exists, say so. Push back when warranted. +- If something is unclear, stop. Name what's confusing. Ask. -## Development Notes +## Simplicity First -### Code Style -- Uses Black formatter, Ruff linter, MyPy type checker -- Pydantic v2 for data validation -- Async/await patterns throughout -- Type hints required for all public APIs -- **Avoid imports within methods** — place all imports at the top of the file (module-level). Inline imports inside functions/methods make dependencies harder to trace and hurt readability. The only exception is avoiding circular imports where absolutely necessary. -- **Use dict spread for immutable copies** — prefer `{**original, "key": new_value}` over `dict(obj)` + mutation. The spread produces the final dict in one step and makes intent clear. -- **Guard at resolution time** — when resolving an optional value through a fallback chain (`a or b or ""`), raise immediately if the resolved result being empty is an error. Don't pass empty strings or sentinel values downstream for the callee to deal with. -- **Extract complex comprehensions to named helpers** — a set/dict comprehension that calls into the DB or manager (e.g. "which of these server IDs are OAuth2?") belongs in a named helper function, not inline in the caller. -- **FastAPI parameter declarations** — mark required query/form params with `= Query(...)` / `= Form(...)` explicitly when other params in the same handler are optional. Mixing `str` (required) with `Optional[str] = None` in the same signature causes silent 422s when the required param is missing. +**Minimum code that solves the problem. Nothing speculative.** -### Testing Strategy -- Unit tests in `tests/test_litellm/` -- Integration tests for each provider in `tests/llm_translation/` -- Proxy tests in `tests/proxy_unit_tests/` -- Load tests in `tests/load_tests/` -- **Always add tests when adding new entity types or features** — if the existing test file covers other entity types, add corresponding tests for the new one -- **Keep monkeypatch stubs in sync with real signatures** — when a function gains a new optional parameter, update every `fake_*` / `stub_*` in tests that patch it to also accept that kwarg (even as `**kwargs`). Stale stubs fail with `unexpected keyword argument` and mask real bugs. -- **Test all branches of name→ID resolution** — when adding server/resource lookup that resolves names to UUIDs, test: (1) name resolves and UUID is allowed, (2) name resolves but UUID is not allowed, (3) name does not resolve at all. The silent-fallback path is where access-control bugs hide. +- No features beyond what was asked. +- No abstractions for single-use code. +- No "flexibility" or "configurability" that wasn't requested. +- No error handling for impossible scenarios. +- If you write 200 lines and it could be 50, rewrite it. -### UI / Backend Consistency -- When wiring a new UI entity type to an existing backend endpoint, verify the backend API contract (single value vs. array, required vs. optional params) and ensure the UI controls match — e.g., use a single-select dropdown when the backend accepts a single value, not a multi-select - -### UI Component Library -- **Always use `antd` for new UI components** — we are migrating off of `@tremor/react`. Do not introduce new `Badge`, `Text`, `Card`, `Grid`, `Title`, or other imports from `@tremor/react` in any new or modified file. Use `antd` equivalents: `Tag` for labels, `Typography.Text` / `Typography.Title` / `Typography.Paragraph` for textual content (avoid plain text-only ``, `

`, `` when Typography fits), and `Card` from `antd`. Note that `antd` has no `"yellow"` Tag color — use `"gold"` for amber/yellow. - -### MCP OAuth / OpenAPI Transport Mapping -- `TRANSPORT.OPENAPI` is a UI-only concept. The backend only accepts `"http"`, `"sse"`, or `"stdio"`. Always map it to `"http"` before any API call (including pre-OAuth temp-session calls). -- FastAPI validation errors return `detail` as an array of `{loc, msg, type}` objects. Error extractors must handle: array (map `.msg`), string, nested `{error: string}`, and fallback. -- When an MCP server already has `authorization_url` stored, skip OAuth discovery (`_discovery_metadata`) — the server URL for OpenAPI MCPs is the spec file, not the API base, and fetching it causes timeouts. -- `client_id` should be optional in the `/authorize` endpoint — if the server has a stored `client_id` in credentials, use that. Never require callers to re-supply it. - -### MCP Credential Storage -- OAuth credentials and BYOK credentials share the `litellm_mcpusercredentials` table, distinguished by a `"type"` field in the JSON payload (`"oauth2"` vs plain string). -- When deleting OAuth credentials, check type before deleting to avoid accidentally deleting a BYOK credential for the same `(user_id, server_id)` pair. -- Always pass the raw `expires_at` timestamp to the client — never set it to `None` for expired credentials. Let the frontend compute the "Expired" display state from the timestamp. -- Use `RecordNotFoundError` (not bare `except Exception`) when catching "already deleted" in credential delete endpoints. - -### Browser Storage Safety (UI) -- Never write LiteLLM access tokens or API keys to `localStorage` — use `sessionStorage` only. `localStorage` survives browser close and is readable by any injected script (XSS). -- Shared utility functions (e.g. `extractErrorMessage`) belong in `src/utils/` — never define them inline in hooks or duplicate them across files. - -### Database Migrations -- Prisma handles schema migrations -- Migration files auto-generated with `prisma migrate dev` -- Always test migrations against both PostgreSQL and SQLite - -### Proxy database access -- **Do not write raw SQL** for proxy DB operations. Use Prisma model methods instead of `execute_raw` / `query_raw`. -- Use the generated client: `prisma_client.db.` (e.g. `litellm_tooltable`, `litellm_usertable`) with `.upsert()`, `.find_many()`, `.find_unique()`, `.update()`, `.update_many()` as appropriate. This avoids schema/client drift, keeps code testable with simple mocks, and matches patterns used in spend logs and other proxy code. -- **No N+1 queries.** Never query the DB inside a loop. Batch-fetch with `{"in": ids}` and distribute in-memory. -- **Batch writes.** Use `create_many`/`update_many`/`delete_many` instead of individual calls (these return counts only; `update_many`/`delete_many` no-op silently on missing rows). When multiple separate writes target the same table (e.g. in `batch_()`), order by primary key to avoid deadlocks. -- **Push work to the DB.** Filter, sort, group, and aggregate in SQL, not Python. Verify Prisma generates the expected SQL — e.g. prefer `group_by` over `find_many(distinct=...)` which does client-side processing. -- **Bound large result sets.** Prisma materializes full results in memory. For results over ~10 MB, paginate with `take`/`skip` or `cursor`/`take`, always with an explicit `order`. Prefer cursor-based pagination (`skip` is O(n)). Don't paginate naturally small result sets. -- **Limit fetched columns on wide tables.** Use `select` to fetch only needed fields — returns a partial object, so downstream code must not access unselected fields. -- **Check index coverage.** For new or modified queries, check `schema.prisma` for a supporting index. Prefer extending an existing index (e.g. `@@index([a])` → `@@index([a, b])`) over adding a new one, unless it's a `@@unique`. Only add indexes for large/frequent queries. -- **Keep schema files in sync.** Apply schema changes to all `schema.prisma` copies (`schema.prisma`, `litellm/proxy/`, `litellm-proxy-extras/`) with a migration under `litellm-proxy-extras/litellm_proxy_extras/migrations/`. - -### Setup Wizard (`litellm/setup_wizard.py`) -- The wizard is implemented as a single `SetupWizard` class with `@staticmethod` methods — keep it that way. No module-level functions except `run_setup_wizard()` (the public entrypoint) and pure helpers (color, ANSI). -- Use `litellm.utils.check_valid_key(model, api_key)` for credential validation — never roll a custom completion call. -- Do not hardcode provider env-key names or model lists that already exist in the codebase. Add a `test_model` field to each provider entry to drive `check_valid_key`; set it to `None` for providers that can't be validated with a single API key (Azure, Bedrock, Ollama). - -### Enterprise Features -- Enterprise-specific code in `enterprise/` directory -- Optional features enabled via environment variables -- Separate licensing and authentication for enterprise features - -### CI Supply-Chain Safety -- **Never pipe a remote script into a shell** (`curl ... | bash`, `wget ... | sh`). Download the artifact to a file, verify its SHA-256 checksum, then install. -- **Pin every external tool to a specific version** with a full URL (not `latest` or `stable`). Unversioned downloads silently change under you. -- **Verify checksums for all downloaded binaries.** Use the provider's official `.sha256` / `.sha256sum` sidecar file when available; otherwise compute and hardcode the digest. -- **Prefer reusable CircleCI commands** (`commands:` section) so a tool is installed and verified in exactly one place, then referenced everywhere with `- install_` or `- wait_for_service`. -- **Don't add tools just because they were there before.** Audit whether an external dependency is still needed. If it can be replaced with a shell one-liner or a tool already in the image, remove it. -- These rules apply to every download in CI: binaries, install scripts, language version managers, package repos. No exceptions. - -### HTTP Client Cache Safety -- **Never close HTTP/SDK clients on cache eviction.** `LLMClientCache._remove_key()` must not call `close()`/`aclose()` on evicted clients — they may still be used by in-flight requests. Doing so causes `RuntimeError: Cannot send a request, as the client has been closed.` after the 1-hour TTL expires. Cleanup happens at shutdown via `close_litellm_async_clients()`. - -### Troubleshooting: DB schema out of sync after proxy restart -`litellm-proxy-extras` runs `prisma migrate deploy` on startup using **its own** bundled migration files, which may lag behind schema changes in the current worktree. Symptoms: `Unknown column`, `Invalid prisma invocation`, or missing data on new fields. - -**Diagnose:** Run `\d "TableName"` in psql and compare against `schema.prisma` — missing columns confirm the issue. - -**Fix options:** -1. **Create a Prisma migration** (permanent) — run `prisma migrate dev --name ` in the worktree. The generated file will be picked up by `prisma migrate deploy` on next startup. -2. **Apply manually for local dev** — `psql -d litellm -c "ALTER TABLE ... ADD COLUMN IF NOT EXISTS ..."` after each proxy start. Fine for dev, not for production. -3. **Update litellm-proxy-extras** — if the package is installed from PyPI, its migration directory must include the new file. Either update the package or run the migration manually until the next release ships it. +Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify. diff --git a/GEMINI.md b/GEMINI.md index 9e950d89b33..41921fdff4d 100644 --- a/GEMINI.md +++ b/GEMINI.md @@ -1,108 +1 @@ -# GEMINI.md - -This file provides guidance to Gemini when working with code in this repository. - -## Development Commands - -### Installation -- `make install-dev` - Install core development dependencies -- `make install-proxy-dev` - Install proxy development dependencies with full feature set -- `make install-test-deps` - Install all test dependencies - -### Testing -- `make test` - Run all tests -- `make test-unit` - Run unit tests (tests/test_litellm) with 4 parallel workers -- `make test-integration` - Run integration tests (excludes unit tests) -- `pytest tests/` - Direct pytest execution - -### Code Quality -- `make lint` - Run all linting (Ruff, MyPy, Black, circular imports, import safety) -- `make format` - Apply Black code formatting -- `make lint-ruff` - Run Ruff linting only -- `make lint-mypy` - Run MyPy type checking only - -### Single Test Files -- `uv run pytest tests/path/to/test_file.py -v` - Run specific test file -- `uv run pytest tests/path/to/test_file.py::test_function -v` - Run specific test - -### Running Scripts -- `uv run python script.py` - Run Python scripts (use for non-test files) - -### GitHub Issue & PR Templates -When contributing to the project, use the appropriate templates: - -**Bug Reports** (`.github/ISSUE_TEMPLATE/bug_report.yml`): -- Describe what happened vs. what you expected -- Include relevant log output -- Specify your LiteLLM version - -**Feature Requests** (`.github/ISSUE_TEMPLATE/feature_request.yml`): -- Describe the feature clearly -- Explain the motivation and use case - -**Pull Requests** (`.github/pull_request_template.md`): -- Add at least 1 test in `tests/litellm/` -- Ensure `make test-unit` passes - -## Architecture Overview - -LiteLLM is a unified interface for 100+ LLM providers with two main components: - -### Core Library (`litellm/`) -- **Main entry point**: `litellm/main.py` - Contains core completion() function -- **Provider implementations**: `litellm/llms/` - Each provider has its own subdirectory -- **Router system**: `litellm/router.py` + `litellm/router_utils/` - Load balancing and fallback logic -- **Type definitions**: `litellm/types/` - Pydantic models and type hints -- **Integrations**: `litellm/integrations/` - Third-party observability, caching, logging -- **Caching**: `litellm/caching/` - Multiple cache backends (Redis, in-memory, S3, etc.) - -### Proxy Server (`litellm/proxy/`) -- **Main server**: `proxy_server.py` - FastAPI application -- **Authentication**: `auth/` - API key management, JWT, OAuth2 -- **Database**: `db/` - Prisma ORM with PostgreSQL/SQLite support -- **Management endpoints**: `management_endpoints/` - Admin APIs for keys, teams, models -- **Pass-through endpoints**: `pass_through_endpoints/` - Provider-specific API forwarding -- **Guardrails**: `guardrails/` - Safety and content filtering hooks -- **UI Dashboard**: Served from `_experimental/out/` (Next.js build) - -## Key Patterns - -### Provider Implementation -- Providers inherit from base classes in `litellm/llms/base.py` -- Each provider has transformation functions for input/output formatting -- Support both sync and async operations -- Handle streaming responses and function calling - -### Error Handling -- Provider-specific exceptions mapped to OpenAI-compatible errors -- Fallback logic handled by Router system -- Comprehensive logging through `litellm/_logging.py` - -### Configuration -- YAML config files for proxy server (see `proxy/example_config_yaml/`) -- Environment variables for API keys and settings -- Database schema managed via Prisma (`proxy/schema.prisma`) - -## Development Notes - -### Code Style -- Uses Black formatter, Ruff linter, MyPy type checker -- Pydantic v2 for data validation -- Async/await patterns throughout -- Type hints required for all public APIs - -### Testing Strategy -- Unit tests in `tests/test_litellm/` -- Integration tests for each provider in `tests/llm_translation/` -- Proxy tests in `tests/proxy_unit_tests/` -- Load tests in `tests/load_tests/` - -### Database Migrations -- Prisma handles schema migrations -- Migration files auto-generated with `prisma migrate dev` -- Always test migrations against both PostgreSQL and SQLite - -### Enterprise Features -- Enterprise-specific code in `enterprise/` directory -- Optional features enabled via environment variables -- Separate licensing and authentication for enterprise features +Read @CLAUDE.md for coding guidelines diff --git a/README.md b/README.md index 72fd43925c9..8df351e9303 100644 --- a/README.md +++ b/README.md @@ -292,7 +292,7 @@ curl -X POST 'http://0.0.0.0:4000/v1/chat/completions' \ | [CompactifAI (`compactifai`)](https://docs.litellm.ai/docs/providers/compactifai) | ✅ | ✅ | ✅ | | | | | | | | | [Custom (`custom`)](https://docs.litellm.ai/docs/providers/custom_llm_server) | ✅ | ✅ | ✅ | | | | | | | | | [Custom OpenAI (`custom_openai`)](https://docs.litellm.ai/docs/providers/openai_compatible) | ✅ | ✅ | ✅ | | | ✅ | ✅ | ✅ | ✅ | | -| [Dashscope (`dashscope`)](https://docs.litellm.ai/docs/providers/dashscope) | ✅ | ✅ | ✅ | | | | | | | | +| [Dashscope (`dashscope`)](https://docs.litellm.ai/docs/providers/dashscope) | ✅ | ✅ | ✅ | ✅ | | | | | | ✅ | | [Databricks (`databricks`)](https://docs.litellm.ai/docs/providers/databricks) | ✅ | ✅ | ✅ | | | | | | | | | [DataRobot (`datarobot`)](https://docs.litellm.ai/docs/providers/datarobot) | ✅ | ✅ | ✅ | | | | | | | | | [Deepgram (`deepgram`)](https://docs.litellm.ai/docs/providers/deepgram) | ✅ | ✅ | ✅ | | | ✅ | | | | | diff --git a/backend/Dockerfile b/backend/Dockerfile new file mode 100644 index 00000000000..2cfdde8a517 --- /dev/null +++ b/backend/Dockerfile @@ -0,0 +1,97 @@ +ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:31da6565f35af6401031c1d7aa91dc84ac76c5c48edd17fb90f0ed9e3173c7a9 +ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:31da6565f35af6401031c1d7aa91dc84ac76c5c48edd17fb90f0ed9e3173c7a9 +ARG UV_IMAGE=ghcr.io/astral-sh/uv:0.11.7@sha256:240fb85ab0f263ef12f492d8476aa3a2e4e1e333f7d67fbdd923d00a506a516a + +FROM $UV_IMAGE AS uvbin + +# ---------- Builder ---------- +FROM $LITELLM_BUILD_IMAGE AS builder + +WORKDIR /app +USER root + +COPY --from=uvbin /uv /uvx /usr/local/bin/ + +# nodejs/npm so `prisma generate` uses Wolfi's Node via PRISMA_USE_GLOBAL_NODE +# instead of nodeenv downloading one whose dynamic deps may not be in Wolfi +# (e.g. Node 26.2.0 needs libatomic). Retry for transient apk.cgr.dev flakes. +RUN for i in 1 2 3; do \ + apk add --no-cache bash gcc python3 python3-dev openssl openssl-dev libsndfile nodejs npm && break; \ + [ $i = 3 ] && { echo "apk add failed after 3 retries" >&2; exit 1; }; \ + sleep 5; \ + done + +# UV_COMPILE_BYTECODE=1 precompiles .pyc at install time → faster cold start. +# UV_LINK_MODE=copy avoids hardlink warnings when uv installs from a +# BuildKit cache mount (different filesystem). +# UV_PYTHON_DOWNLOADS=0 force uv to use the apk-installed CPython instead of +# silently pulling a managed interpreter. +# PRISMA_USE_GLOBAL_NODE explicit (matches default) so an env override can't +# silently re-enable nodeenv's Node download. +ENV UV_PROJECT_ENVIRONMENT=/app/.venv \ + UV_LINK_MODE=copy \ + UV_COMPILE_BYTECODE=1 \ + UV_PYTHON_DOWNLOADS=0 \ + PRISMA_USE_GLOBAL_NODE=true \ + PATH="/app/.venv/bin:${PATH}" + +# Stage 1 — install dependencies only. +RUN --mount=type=cache,target=/root/.cache/uv \ + --mount=type=bind,source=pyproject.toml,target=pyproject.toml \ + --mount=type=bind,source=uv.lock,target=uv.lock \ + --mount=type=bind,source=enterprise/pyproject.toml,target=enterprise/pyproject.toml \ + --mount=type=bind,source=litellm-proxy-extras/pyproject.toml,target=litellm-proxy-extras/pyproject.toml \ + uv sync --frozen --no-install-project --no-install-workspace --no-default-groups --no-editable \ + --extra proxy \ + --extra proxy-runtime \ + --extra extra_proxy \ + --extra semantic-router \ + --python python3 + +# Stage 2 — copy source and install the project + workspace members. +COPY . . + +RUN --mount=type=cache,target=/root/.cache/uv \ + uv sync --frozen --no-default-groups --no-editable \ + --extra proxy \ + --extra proxy-runtime \ + --extra extra_proxy \ + --extra semantic-router \ + --python python3 + +RUN mkdir -p /home/nonroot && \ + HOME=/home/nonroot prisma generate --schema=./schema.prisma && \ + chown -R nonroot:nonroot /home/nonroot/.cache + +# ---------- Runtime ---------- +FROM $LITELLM_RUNTIME_IMAGE AS runtime + +USER root + +RUN for i in 1 2 3; do \ + apk add --no-cache bash openssl tzdata python3 libsndfile libatomic && break; \ + [ $i = 3 ] && { echo "apk add failed after 3 retries" >&2; exit 1; }; \ + sleep 5; \ + done + +# wolfi-base ships an unprivileged `nonroot` account (UID/GID 65532) with +# /home/nonroot. We run the backend as that user +WORKDIR /app +ENV HOME=/home/nonroot \ + PATH="/app/.venv/bin:${PATH}" \ + PYTHONPATH="/app" \ + PYTHONDONTWRITEBYTECODE=1 \ + PYTHONUNBUFFERED=1 + +COPY --from=builder --chown=nonroot:nonroot /app /app +COPY --from=builder --chown=nonroot:nonroot /home/nonroot/.cache /home/nonroot/.cache + +RUN find /app/.venv -type f -path "*/tornado/test/*" -delete && \ + find /app/.venv -type d -path "*/tornado/test" -delete + +USER nonroot + +EXPOSE 4001/tcp + +ENTRYPOINT ["uvicorn", "backend.main:app"] +CMD ["--host", "0.0.0.0", "--port", "4001"] diff --git a/backend/main.py b/backend/main.py new file mode 100644 index 00000000000..4092cd63f69 --- /dev/null +++ b/backend/main.py @@ -0,0 +1,51 @@ +"""UI backend entrypoint. + +Reuses the existing FastAPI app from `litellm.proxy.proxy_server` and trims its +route table to just the management/admin surface used by the dashboard. Purely +additive — no existing module is modified. + +Run with: + uvicorn backend.main:app --host 0.0.0.0 --port 4001 +""" + +from contextlib import asynccontextmanager + +from fastapi.routing import Mount + +# See gateway/main.py for why we assemble DATABASE_URL(s) here before +# importing proxy_server. +from litellm.proxy.db.db_url_settings import DatabaseURLSettings + +DatabaseURLSettings.from_env().apply_to_env() + +from litellm.proxy.proxy_server import app + +from backend.routes.allowlist import BACKEND_EXACT_PATHS, BACKEND_PATH_PREFIXES + + +def _is_backend_route(route) -> bool: + """Keep the route on the backend if its path is in the management surface.""" + path = getattr(route, "path", None) + if path is None: + return False + if isinstance(route, Mount): + # Static UI mounts are served by the dedicated UI container, not here. + return False + if path in BACKEND_EXACT_PATHS: + return True + return any(path.startswith(prefix) for prefix in BACKEND_PATH_PREFIXES) + + +# See gateway/main.py for why the trim runs inside the lifespan instead of at +# module scope. +_proxy_lifespan = app.router.lifespan_context + + +@asynccontextmanager +async def _backend_lifespan(app_): + async with _proxy_lifespan(app_): + app_.router.routes = [r for r in app_.router.routes if _is_backend_route(r)] + yield + + +app.router.lifespan_context = _backend_lifespan diff --git a/backend/routes/__init__.py b/backend/routes/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/backend/routes/allowlist.py b/backend/routes/allowlist.py new file mode 100644 index 00000000000..610ba3dbd69 --- /dev/null +++ b/backend/routes/allowlist.py @@ -0,0 +1,135 @@ +"""Path allowlist for the UI backend (control plane) component. + +The backend exposes management/admin endpoints consumed by the UI: keys, users, +teams, orgs, customers, budgets, tags, workflows, model management, spend & +analytics, settings (router/cache/cost-tracking/fallbacks), SSO/onboarding, +audit logs, debug, enterprise admin, and UI bootstrap helpers (logo, favicon, +.well-known config). + +Anything LLM data-plane is dropped — those run on the gateway component. +""" + +BACKEND_PATH_PREFIXES: tuple[str, ...] = ( + # Identity / access + "/key/", + "/v2/key/", + "/user/", + "/v2/user/", + "/team/", + "/v2/team/", + "/organization/", + "/customer/", + "/end_user/", + "/sso/", + "/login", + "/v2/login", + "/v3/login", + "/logout", + "/token", + "/onboarding/", + "/audit", + "/oauth/", + "/invitation/", + "/jwt/", + # Models & routing config + "/model/", + "/v1/model/info", + "/v2/model/", + "/model_group", + "/model_access_group/", + "/model_hub/", + "/v1/access_group", + "/access_group/", + "/router/", + "/router_settings", + "/adaptive_router/", + "/fallback", + "/fallbacks", + "/cache_settings", + "/cost_tracking", + "/cost/", + "/credentials", + "/credential", + "/provider/budgets", + # Tools / agents (registry & policy admin) + "/v1/tool/", + "/v1/agents", + # Guardrails admin + "/v2/guardrails/", + # MCP server admin + BYOK OAuth flow (UI-initiated) + dynamic per-server endpoints + "/v1/mcp/", + "/test/", + "/{mcp_server_name}/", + # Budgets / tags / workflows / memory mgmt + "/budget/", + "/tag/", + "/workflow/", + "/v1/workflows/", + "/project/", + "/memory/", + "/mcp/", + # Spend / analytics + "/spend/", + "/analytics/", + "/global/", + "/user_agent", + "/usage/", + "/daily/", + # CloudZero cost-export admin (init / settings / export / dry-run / delete) + "/cloudzero/", + # Caching admin + "/cache/", + "/caching/", + # Callbacks / hooks + "/active/callbacks", + "/callbacks", + "/team_callback", + # Alerting / email / IP allowlist + "/alerting/", + "/email/", + "/add/allowed_ip", + "/delete/allowed_ip", + "/get/", + # Enterprise admin + "/enterprise/", + # Debug / config / profiling + "/debug/", + "/config/", + "/memory-usage-in-mem-cache", + "/otel-spans", + "/lazy/", + "/in_product_nudges", + # Admin reload / schedule + "/reload/", + "/schedule/", + "/settings", + "/update/", + "/upload/", + # Dev / admin utilities + "/utils/", + # UI bootstrap helpers (assets the dashboard fetches) + "/get_logo_url", + "/get_image", + "/get_favicon", + "/.well-known/", + "/litellm/.well-known/", + "/ui_discovery/", + "/ui-config", + "/sso_settings", + "/public/", + "/robots.txt", + # Health (k8s probes) + "/health", +) + +BACKEND_EXACT_PATHS: frozenset[str] = frozenset( + { + "/", + "/routes", + "/openapi.json", + "/docs", + "/docs/oauth2-redirect", + "/redoc", + "/fallback/login", + } +) diff --git a/codecov.yaml b/codecov.yaml index 8609d3143d6..58681b884d0 100644 --- a/codecov.yaml +++ b/codecov.yaml @@ -3,6 +3,16 @@ codecov: notify: wait_for_ci: false # post as soon as expected uploads arrive, don't wait on CI +# Uploads are flagged per workflow/shard (GHA) or "circleci". carryforward makes +# a re-upload of a flag replace its prior session instead of accumulating a +# conflicting one, and lets a commit reuse a flag from its parent when that flag +# was not re-uploaded. Required because the same commit can receive the +# push-triggered workflows more than once (re-runs / branches cut at the same +# SHA); flagless overlapping sessions made Codecov drop the largest files. +flag_management: + default_rules: + carryforward: true + component_management: individual_components: - component_id: "Router" diff --git a/cookbook/gollem_go_agent_framework/go.mod b/cookbook/gollem_go_agent_framework/go.mod index 89d9033aa22..a8dc9365d7f 100644 --- a/cookbook/gollem_go_agent_framework/go.mod +++ b/cookbook/gollem_go_agent_framework/go.mod @@ -1,5 +1,5 @@ module github.com/BerriAI/litellm/cookbook/gollem_go_agent_framework -go 1.25.1 +go 1.26.3 require github.com/fugue-labs/gollem v0.1.0 diff --git a/deploy/charts/litellm-helm/Chart.yaml b/deploy/charts/litellm-helm/Chart.yaml index 0f6db331e50..0aef2442bfe 100644 --- a/deploy/charts/litellm-helm/Chart.yaml +++ b/deploy/charts/litellm-helm/Chart.yaml @@ -24,7 +24,7 @@ version: 1.1.0 # incremented each time you make changes to the application. Versions are not expected to # follow Semantic Versioning. They should reflect the version the application is using. # It is recommended to use it with quotes. -appVersion: v1.80.12 +appVersion: v1.85.1 annotations: org.opencontainers.image.source: "https://github.com/BerriAI/litellm" diff --git a/deploy/charts/litellm-helm/templates/deployment.yaml b/deploy/charts/litellm-helm/templates/deployment.yaml index 25f69080878..b9cd1be06ec 100644 --- a/deploy/charts/litellm-helm/templates/deployment.yaml +++ b/deploy/charts/litellm-helm/templates/deployment.yaml @@ -30,7 +30,7 @@ spec: checksum/config: {{ include (print $.Template.BasePath "/configmap-litellm.yaml") . | sha256sum }} {{- end }} {{- with .Values.podAnnotations }} - {{- toYaml . | nindent 8 }} + {{- tpl (toYaml .) $ | nindent 8 }} {{- end }} labels: {{- include "litellm.labels" . | nindent 8 }} @@ -53,7 +53,7 @@ spec: - name: {{ include "litellm.name" . }} securityContext: {{- toYaml .Values.securityContext | nindent 12 }} - image: "{{ .Values.image.repository }}:{{ .Values.image.tag | default (printf "main-%s" .Chart.AppVersion) }}" + image: "{{ .Values.image.repository }}:{{ .Values.image.tag | default .Chart.AppVersion }}" imagePullPolicy: {{ .Values.image.pullPolicy }} env: - name: HOST diff --git a/deploy/charts/litellm-helm/templates/hpa.yaml b/deploy/charts/litellm-helm/templates/hpa.yaml index 71e199c5aeb..fec4d1f5c5e 100644 --- a/deploy/charts/litellm-helm/templates/hpa.yaml +++ b/deploy/charts/litellm-helm/templates/hpa.yaml @@ -12,6 +12,10 @@ spec: name: {{ include "litellm.fullname" . }} minReplicas: {{ .Values.autoscaling.minReplicas }} maxReplicas: {{ .Values.autoscaling.maxReplicas }} + {{- if .Values.autoscaling.behavior }} + behavior: + {{- toYaml .Values.autoscaling.behavior | nindent 4 }} + {{- end }} metrics: {{- if .Values.autoscaling.targetCPUUtilizationPercentage }} - type: Resource diff --git a/deploy/charts/litellm-helm/templates/migrations-job.yaml b/deploy/charts/litellm-helm/templates/migrations-job.yaml index c3f32fe32f3..5ec7f5b7f3e 100644 --- a/deploy/charts/litellm-helm/templates/migrations-job.yaml +++ b/deploy/charts/litellm-helm/templates/migrations-job.yaml @@ -41,7 +41,7 @@ spec: {{- end }} containers: - name: prisma-migrations - image: "{{ .Values.image.repository }}:{{ .Values.image.tag | default (printf "main-%s" .Chart.AppVersion) }}" + image: "{{ .Values.image.repository }}:{{ .Values.image.tag | default .Chart.AppVersion }}" imagePullPolicy: {{ .Values.image.pullPolicy }} securityContext: {{- toYaml .Values.securityContext | nindent 12 }} diff --git a/deploy/charts/litellm-helm/tests/deployment_tests.yaml b/deploy/charts/litellm-helm/tests/deployment_tests.yaml index df6d1345644..f3d62651d8f 100644 --- a/deploy/charts/litellm-helm/tests/deployment_tests.yaml +++ b/deploy/charts/litellm-helm/tests/deployment_tests.yaml @@ -377,3 +377,28 @@ tests: content: name: sidecar-tpl image: "ghcr.io/berriai/litellm-database:test" + - it: should support tpl in podAnnotations + template: deployment.yaml + set: + image: + repository: ghcr.io/berriai/litellm-database + tag: test + # Mirrors the real-world scenario this feature unblocks: + # user disables the built-in ConfigMap (and its built-in checksum/config + # annotation) and re-implements checksum/config themselves via tpl. + proxyConfigMap: + create: false + podAnnotations: + checksum/config: "{{ .Values.image.tag }}" + example.com/some-key: "{{ .Values.image.repository }}" + example.com/literal: "plain-string-value" + asserts: + - equal: + path: spec.template.metadata.annotations["checksum/config"] + value: "test" + - equal: + path: spec.template.metadata.annotations["example.com/some-key"] + value: "ghcr.io/berriai/litellm-database" + - equal: + path: spec.template.metadata.annotations["example.com/literal"] + value: "plain-string-value" diff --git a/deploy/charts/litellm-helm/tests/hpa_tests.yaml b/deploy/charts/litellm-helm/tests/hpa_tests.yaml new file mode 100644 index 00000000000..ec18c3591d3 --- /dev/null +++ b/deploy/charts/litellm-helm/tests/hpa_tests.yaml @@ -0,0 +1,36 @@ +suite: "hpa with behavior" +templates: + - hpa.yaml +tests: + - it: "renders behavior when set" + set: + autoscaling.enabled: true + autoscaling.behavior: + scaleUp: + stabilizationWindowSeconds: 60 + policies: + - type: Pods + value: 2 + periodSeconds: 60 + scaleDown: + stabilizationWindowSeconds: 90 + policies: + - type: Pods + value: 1 + periodSeconds: 60 + asserts: + - isKind: { of: HorizontalPodAutoscaler } + - equal: { path: spec.behavior.scaleUp.stabilizationWindowSeconds, value: 60 } + - equal: { path: spec.behavior.scaleDown.stabilizationWindowSeconds, value: 90 } + +--- +suite: "hpa without behavior" +templates: + - hpa.yaml +tests: + - it: "does not render behavior when not set" + set: + autoscaling.enabled: true + asserts: + - isKind: { of: HorizontalPodAutoscaler } + - isNull: { path: spec.behavior } diff --git a/deploy/charts/litellm-helm/values.yaml b/deploy/charts/litellm-helm/values.yaml index 9c7c013341b..a9cdf28f0e7 100644 --- a/deploy/charts/litellm-helm/values.yaml +++ b/deploy/charts/litellm-helm/values.yaml @@ -10,7 +10,7 @@ image: repository: ghcr.io/berriai/litellm-database pullPolicy: Always # Overrides the image tag whose default is the chart appVersion. - # tag: "main-latest" + # tag: "latest" tag: "" imagePullSecrets: [] @@ -184,6 +184,7 @@ autoscaling: maxReplicas: 100 targetCPUUtilizationPercentage: 80 # targetMemoryUtilizationPercentage: 80 + # behavior: {} # Autoscaling with keda is mutually exclusive with hpa keda: diff --git a/docker/Dockerfile.non_root b/docker/Dockerfile.non_root index 4de4a55981d..8717e5b3fcd 100644 --- a/docker/Dockerfile.non_root +++ b/docker/Dockerfile.non_root @@ -24,7 +24,8 @@ RUN for i in 1 2 3; do \ curl \ openssl \ libsndfile \ - nodejs && break || sleep 5; \ + nodejs \ + npm && break || sleep 5; \ done ENV UV_PROJECT_ENVIRONMENT=/app/.venv \ @@ -54,22 +55,10 @@ COPY . . # Set non-root flag for build time consistency ENV LITELLM_NON_ROOT=true -# Stage the pre-built Admin UI from the checked-in Next.js static export. -# _experimental/out/ is regenerated as part of the release runbook. -# Restructure extensionless routes (foo.html -> foo/index.html) to match the layout -# proxy_server.py expects, and drop a readiness marker. RUN mkdir -p /var/lib/litellm/ui /var/lib/litellm/assets && \ cp -r /app/litellm/proxy/_experimental/out/. /var/lib/litellm/ui/ && \ cp /app/litellm/proxy/logo.jpg /var/lib/litellm/assets/logo.jpg && \ - ( cd /var/lib/litellm/ui && \ - for html_file in *.html; do \ - if [ "$html_file" != "index.html" ] && [ -f "$html_file" ]; then \ - folder_name="${html_file%.html}" && \ - mkdir -p "$folder_name" && \ - mv "$html_file" "$folder_name/index.html"; \ - fi; \ - done && \ - touch .litellm_ui_ready ) + touch /var/lib/litellm/ui/.litellm_ui_ready RUN --mount=type=cache,target=/app/.cache/uv,id=litellm-uv-cache \ if [ "$PROXY_EXTRAS_SOURCE" = "published" ]; then \ diff --git a/enterprise/litellm_enterprise/enterprise_callbacks/send_emails/resend_email.py b/enterprise/litellm_enterprise/enterprise_callbacks/send_emails/resend_email.py index 7593e66aa47..3fad5601f52 100644 --- a/enterprise/litellm_enterprise/enterprise_callbacks/send_emails/resend_email.py +++ b/enterprise/litellm_enterprise/enterprise_callbacks/send_emails/resend_email.py @@ -19,12 +19,26 @@ RESEND_API_ENDPOINT = "https://api.resend.com/emails" class ResendEmailLogger(BaseEmailLogger): + """ + Send emails using Resend's API. + + Required env vars: + - RESEND_API_KEY + + Optional env vars: + - RESEND_FROM_EMAIL: Override the default sender address. Must be on a + domain verified in your Resend account. When unset, falls back to the + `from_email` argument passed by the caller (which defaults to + `notifications@alerts.litellm.ai` and only works on LiteLLM Cloud). + """ + def __init__(self, internal_usage_cache=None, **kwargs): super().__init__(internal_usage_cache=internal_usage_cache, **kwargs) self.async_httpx_client = get_async_httpx_client( llm_provider=httpxSpecialProvider.LoggingCallback ) self.resend_api_key = os.getenv("RESEND_API_KEY") + self.resend_from_email = os.getenv("RESEND_FROM_EMAIL") async def send_email( self, @@ -33,13 +47,14 @@ class ResendEmailLogger(BaseEmailLogger): subject: str, html_body: str, ): + sender_email = self.resend_from_email or from_email verbose_logger.debug( - f"Sending email from {from_email} to {to_email} with subject {subject}" + f"Sending email from {sender_email} to {to_email} with subject {subject}" ) response = await self.async_httpx_client.post( url=RESEND_API_ENDPOINT, json={ - "from": from_email, + "from": sender_email, "to": to_email, "subject": subject, "html": html_body, diff --git a/enterprise/litellm_enterprise/proxy/common_utils/check_batch_cost.py b/enterprise/litellm_enterprise/proxy/common_utils/check_batch_cost.py index 356f6ecd4b5..ee7745d0add 100644 --- a/enterprise/litellm_enterprise/proxy/common_utils/check_batch_cost.py +++ b/enterprise/litellm_enterprise/proxy/common_utils/check_batch_cost.py @@ -300,6 +300,42 @@ class CheckBatchCost: custom_llm_provider=custom_llm_provider, ) + # CheckBatchCost bypasses async_post_call_success_hook, so convert raw + # output/error file IDs to managed base64 IDs before the DB write here. + managed_files_hook = self.proxy_logging_obj.get_proxy_hook("managed_files") + if managed_files_hook is not None: + from litellm.proxy._types import UserAPIKeyAuth + _minimal_auth = UserAPIKeyAuth( + user_id=job.created_by or "default-user-id", + team_id=getattr(job, "team_id", None), + ) + for _file_attr in ["output_file_id", "error_file_id"]: + _raw_file_id = getattr(response, _file_attr, None) + if _raw_file_id and not _is_base64_encoded_unified_file_id(_raw_file_id): + try: + _unified_file_id = managed_files_hook.get_unified_output_file_id( + output_file_id=_raw_file_id, + model_id=model_id, + model_name=str(model_name) if model_name else deployment_info.model_name or None, + ) + await managed_files_hook.store_unified_file_id( + file_id=_unified_file_id, + file_object=None, + litellm_parent_otel_span=None, + model_mappings={model_id: _raw_file_id}, + user_api_key_dict=_minimal_auth, + ) + setattr(response, _file_attr, _unified_file_id) + verbose_proxy_logger.info( + f"CheckBatchCost: converted {_file_attr} " + f"{_raw_file_id!r} -> managed ID for batch {batch_id}" + ) + except Exception as _e: + verbose_proxy_logger.warning( + f"CheckBatchCost: failed to create managed file ID for " + f"{_file_attr}={_raw_file_id!r}: {_e}" + ) + # Pass deployment model_info so custom batch pricing # (input_cost_per_token_batches etc.) is used for cost calc deployment_model_info = deployment_info.model_info.model_dump() if deployment_info.model_info else {} diff --git a/enterprise/litellm_enterprise/proxy/hooks/managed_files.py b/enterprise/litellm_enterprise/proxy/hooks/managed_files.py index 5ed49070347..ae5905f9cdf 100644 --- a/enterprise/litellm_enterprise/proxy/hooks/managed_files.py +++ b/enterprise/litellm_enterprise/proxy/hooks/managed_files.py @@ -658,7 +658,7 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints): if isinstance(content, str): continue for c in content: - if c["type"] == "file": + if c.get("type") == "file": file_object = cast(ChatCompletionFileObject, c) file_object_file_field = file_object["file"] file_id = file_object_file_field.get("file_id") diff --git a/enterprise/pyproject.toml b/enterprise/pyproject.toml index 9698e7912d4..d0432448433 100644 --- a/enterprise/pyproject.toml +++ b/enterprise/pyproject.toml @@ -1,6 +1,6 @@ [project] name = "litellm-enterprise" -version = "0.1.40" +version = "0.1.42" description = "Package for LiteLLM Enterprise features" readme = "README.md" requires-python = ">=3.9" @@ -26,7 +26,7 @@ required-version = ">=0.10.9" module-root = "" [tool.commitizen] -version = "0.1.40" +version = "0.1.42" version_files = [ "pyproject.toml:^version", "../pyproject.toml:litellm-enterprise==", diff --git a/gateway/Dockerfile b/gateway/Dockerfile new file mode 100644 index 00000000000..19c8a10fdfe --- /dev/null +++ b/gateway/Dockerfile @@ -0,0 +1,97 @@ +ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:31da6565f35af6401031c1d7aa91dc84ac76c5c48edd17fb90f0ed9e3173c7a9 +ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:31da6565f35af6401031c1d7aa91dc84ac76c5c48edd17fb90f0ed9e3173c7a9 +ARG UV_IMAGE=ghcr.io/astral-sh/uv:0.11.7@sha256:240fb85ab0f263ef12f492d8476aa3a2e4e1e333f7d67fbdd923d00a506a516a + +FROM $UV_IMAGE AS uvbin + +# ---------- Builder ---------- +FROM $LITELLM_BUILD_IMAGE AS builder + +WORKDIR /app +USER root + +COPY --from=uvbin /uv /uvx /usr/local/bin/ + +# nodejs/npm so `prisma generate` uses Wolfi's Node via PRISMA_USE_GLOBAL_NODE +# instead of nodeenv downloading one whose dynamic deps may not be in Wolfi +# (e.g. Node 26.2.0 needs libatomic). Retry for transient apk.cgr.dev flakes. +RUN for i in 1 2 3; do \ + apk add --no-cache bash gcc python3 python3-dev openssl openssl-dev libsndfile nodejs npm && break; \ + [ $i = 3 ] && { echo "apk add failed after 3 retries" >&2; exit 1; }; \ + sleep 5; \ + done + +# UV_COMPILE_BYTECODE=1 precompiles .pyc at install time → faster cold start. +# UV_LINK_MODE=copy avoids hardlink warnings when uv installs from a +# BuildKit cache mount (different filesystem). +# UV_PYTHON_DOWNLOADS=0 force uv to use the apk-installed CPython instead of +# silently pulling a managed interpreter. +# PRISMA_USE_GLOBAL_NODE explicit (matches default) so an env override can't +# silently re-enable nodeenv's Node download. +ENV UV_PROJECT_ENVIRONMENT=/app/.venv \ + UV_LINK_MODE=copy \ + UV_COMPILE_BYTECODE=1 \ + UV_PYTHON_DOWNLOADS=0 \ + PRISMA_USE_GLOBAL_NODE=true \ + PATH="/app/.venv/bin:${PATH}" + +# Stage 1 — install dependencies only. +RUN --mount=type=cache,target=/root/.cache/uv \ + --mount=type=bind,source=pyproject.toml,target=pyproject.toml \ + --mount=type=bind,source=uv.lock,target=uv.lock \ + --mount=type=bind,source=enterprise/pyproject.toml,target=enterprise/pyproject.toml \ + --mount=type=bind,source=litellm-proxy-extras/pyproject.toml,target=litellm-proxy-extras/pyproject.toml \ + uv sync --frozen --no-install-project --no-install-workspace --no-default-groups --no-editable \ + --extra proxy \ + --extra proxy-runtime \ + --extra extra_proxy \ + --extra semantic-router \ + --python python3 + +# Stage 2 — copy source and install the project + workspace members. +COPY . . + +RUN --mount=type=cache,target=/root/.cache/uv \ + uv sync --frozen --no-default-groups --no-editable \ + --extra proxy \ + --extra proxy-runtime \ + --extra extra_proxy \ + --extra semantic-router \ + --python python3 + +RUN mkdir -p /home/nonroot && \ + HOME=/home/nonroot prisma generate --schema=./schema.prisma && \ + chown -R nonroot:nonroot /home/nonroot/.cache + +# ---------- Runtime ---------- +FROM $LITELLM_RUNTIME_IMAGE AS runtime + +USER root + +RUN for i in 1 2 3; do \ + apk add --no-cache bash openssl tzdata python3 libsndfile libatomic && break; \ + [ $i = 3 ] && { echo "apk add failed after 3 retries" >&2; exit 1; }; \ + sleep 5; \ + done + +# wolfi-base ships an unprivileged `nonroot` account (UID/GID 65532) with +# /home/nonroot. We run the proxy as that user. +WORKDIR /app +ENV HOME=/home/nonroot \ + PATH="/app/.venv/bin:${PATH}" \ + PYTHONPATH="/app" \ + PYTHONDONTWRITEBYTECODE=1 \ + PYTHONUNBUFFERED=1 + +COPY --from=builder --chown=nonroot:nonroot /app /app +COPY --from=builder --chown=nonroot:nonroot /home/nonroot/.cache /home/nonroot/.cache + +RUN find /app/.venv -type f -path "*/tornado/test/*" -delete && \ + find /app/.venv -type d -path "*/tornado/test" -delete + +USER nonroot + +EXPOSE 4000/tcp + +ENTRYPOINT ["sh", "-c", "exec uvicorn gateway.main:app --workers \"${NUM_WORKERS:-1}\" \"$@\"", "--"] +CMD ["--host", "0.0.0.0", "--port", "4000"] diff --git a/gateway/main.py b/gateway/main.py new file mode 100644 index 00000000000..09d30f5da3f --- /dev/null +++ b/gateway/main.py @@ -0,0 +1,59 @@ +"""Gateway entrypoint. + +Reuses the existing FastAPI app from `litellm.proxy.proxy_server` and trims its +route table to just the LLM data-plane surface. The trim is purely additive — +no existing module is modified, the full app continues to work via the legacy +entrypoint (`litellm.proxy.proxy_server:app`). + +Run with: + uvicorn gateway.main:app --host 0.0.0.0 --port 4000 +""" + +from contextlib import asynccontextmanager + +from fastapi.routing import Mount + +# Assemble DATABASE_URL (+ DATABASE_URL_READ_REPLICA) from the discrete +# DATABASE_* env vars before proxy_server imports spin up Prisma. Handles +# both IAM (mint a token) and password auth, writer and reader. The standard +# CLI flow does this in proxy_cli.py; we bypass proxy_cli by uvicorn'ing the +# app directly, so without this Prisma initializes with the placeholder URL +# and every DB-needing endpoint returns "Database not connected". +from litellm.proxy.db.db_url_settings import DatabaseURLSettings + +DatabaseURLSettings.from_env().apply_to_env() + +from litellm.proxy.proxy_server import app + +from gateway.routes.allowlist import GATEWAY_EXACT_PATHS, GATEWAY_PATH_PREFIXES + + +def _is_gateway_route(route) -> bool: + """Keep the route on the gateway if its path is in the LLM data-plane surface.""" + path = getattr(route, "path", None) + if path is None: + return False + if isinstance(route, Mount): + # Gateway never serves the static UI or its asset bundles. + return False + if path in GATEWAY_EXACT_PATHS: + return True + return any(path.startswith(prefix) for prefix in GATEWAY_PATH_PREFIXES) + + +# Wrap proxy_server's existing lifespan so the route trim runs *after* its +# startup hooks (and any plugin code those hooks load) have had a chance to +# register routes. A module-load filter would miss routes added during +# startup; running inside the lifespan, after the inner __aenter__, catches +# them while still completing before uvicorn opens the listener. +_proxy_lifespan = app.router.lifespan_context + + +@asynccontextmanager +async def _gateway_lifespan(app_): + async with _proxy_lifespan(app_): + app_.router.routes = [r for r in app_.router.routes if _is_gateway_route(r)] + yield + + +app.router.lifespan_context = _gateway_lifespan diff --git a/gateway/routes/__init__.py b/gateway/routes/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/gateway/routes/allowlist.py b/gateway/routes/allowlist.py new file mode 100644 index 00000000000..cbbf55c9873 --- /dev/null +++ b/gateway/routes/allowlist.py @@ -0,0 +1,121 @@ +"""Path allowlist for the gateway component. + +The gateway exposes the LLM data-plane surface: chat/completions, embeddings, +audio, batches, files, fine-tuning, rerank, ocr, rag, video, search, image, +responses, vector stores, passthrough providers, realtime websockets, MCP +tool-call endpoints, and operational endpoints (/health, /metrics). + +Any path not listed here is dropped from the gateway process so management/UI +endpoints don't ride on the same pods. + +Versioned data-plane paths are enumerated explicitly rather than allowing a +blanket `/v1/` or `/v2/` prefix — those broad prefixes would otherwise also +match management routes like `/v1/access_group`, `/v1/tool/{tool_name}/logs`, +`/v2/key/info`, etc. +""" + +GATEWAY_PATH_PREFIXES: tuple[str, ...] = ( + # OpenAI-compatible data-plane surface (versioned + unversioned) + "/v1/chat/", + "/chat/", + "/v1/completions", + "/completions", + "/v1/embeddings", + "/embeddings", + "/v1/moderations", + "/moderations", + "/v1/audio/", + "/audio/", + "/v1/images/", + "/images/", + "/v1/files", + "/files", + "/v1/batches", + "/batches", + "/v1/fine_tuning/", + "/fine_tuning/", + "/v1/fine-tuning/", + "/fine-tuning/", + "/v1/responses", + "/responses", + "/v1/threads", + "/threads", + "/v1/assistants", + "/assistants", + "/v1/vector_stores", + "/vector_stores", + "/v1/indexes", + "/v1/models", + "/models", + "/openai/", + "/engines/", + # Anthropic / agentic data-plane surface + "/v1/messages", + "/messages", + "/v1/skills", + "/v1/a2a/", + # LiteLLM-native LLM surface + "/v1/rerank", + "/v2/rerank", + "/rerank", + "/v1/ocr", + "/ocr", + "/v1/rag/", + "/rag/", + "/v1/video", + "/v1/videos", + "/video/", + "/videos", + "/v1/search", + "/search", + "/v1/containers", + "/containers", + "/v1/evals", + "/v1/memory", + "/queue/chat/", + # Google data plane (v1beta is the Google AI Studio version) + "/v1beta/", + "/interactions", + # Provider passthrough + "/anthropic/", + "/azure/", + "/azure_ai/", + "/aws/", + "/bedrock/", + "/cohere/", + "/gemini/", + "/google/", + "/vertex_ai/", + "/vertex-ai/", + "/assemblyai/", + "/eu.assemblyai/", + "/langfuse/", + "/vllm/", + "/mistral/", + "/groq/", + "/voyage/", + "/cursor/", + "/milvus/", + "/openai_passthrough/", + # Dynamic provider / toolset passthrough (path templates) + "/{provider}/", + "/toolset/", + # Realtime / streaming + "/v1/realtime", + "/realtime", + # Health & ops + "/health", + "/metrics", +) + +GATEWAY_EXACT_PATHS: frozenset[str] = frozenset( + { + "/", + "/routes", + "/openapi.json", + "/docs", + "/docs/oauth2-redirect", + "/redoc", + "/test", + } +) diff --git a/helm/litellm/Chart.yaml b/helm/litellm/Chart.yaml new file mode 100644 index 00000000000..e67f5790c7e --- /dev/null +++ b/helm/litellm/Chart.yaml @@ -0,0 +1,8 @@ +apiVersion: v2 +name: litellm +description: LiteLLM componentized — gateway, UI backend, and UI as separate services +type: application +version: 0.1.0 +appVersion: "0.1.0" +annotations: + org.opencontainers.image.source: "https://github.com/BerriAI/litellm" diff --git a/helm/litellm/templates/NOTES.txt b/helm/litellm/templates/NOTES.txt new file mode 100644 index 00000000000..5b939fe480a --- /dev/null +++ b/helm/litellm/templates/NOTES.txt @@ -0,0 +1,49 @@ +LiteLLM componentized — release {{ .Release.Name }} in namespace {{ .Release.Namespace }}. + +Components: +{{- if .Values.gateway.enabled }} + - gateway : Service {{ include "litellm.gateway.fullname" . }} on port {{ .Values.gateway.service.port }} +{{- end }} +{{- if .Values.backend.enabled }} + - backend : Service {{ include "litellm.backend.fullname" . }} on port {{ .Values.backend.service.port }} +{{- end }} +{{- if .Values.ui.enabled }} + - ui : Service {{ include "litellm.ui.fullname" . }} on port {{ .Values.ui.service.port }} +{{- end }} + +Port-forward examples: + kubectl -n {{ .Release.Namespace }} port-forward svc/{{ include "litellm.gateway.fullname" . }} {{ .Values.gateway.service.port }} + kubectl -n {{ .Release.Namespace }} port-forward svc/{{ include "litellm.backend.fullname" . }} {{ .Values.backend.service.port }} + kubectl -n {{ .Release.Namespace }} port-forward svc/{{ include "litellm.ui.fullname" . }} {{ .Values.ui.service.port }} + +Reminders: + - Sensitive values come from Secret references only. Before installing, set: + - masterKey.secretName (Secret with the proxy master key) + - database.writer.{host,port,dbname} (writer connection pieces) + - database.writer.passwordSecret.{name,usernameKey,passwordKey} + (Secret holding the writer DB username + password) + - database.writer.useIAMAuth: true (optional — chart sets IAM_TOKEN_DB_AUTH=true and + omits DATABASE_PASSWORD / DATABASE_URL so the proxy + mints the URL from an IAM token at startup) + - database.reader.host (optional — enables read-replica routing; reader + .passwordSecret.name is required when set, unless + .useIAMAuth is true) + - database.reader.useIAMAuth: true (optional, requires database.writer.useIAMAuth: true — + chart emits DATABASE_*_READ_REPLICA env vars and + omits DATABASE_PASSWORD_READ_REPLICA / + DATABASE_URL_READ_REPLICA so the proxy mints the + reader URL from an IAM token at startup) + - redis.passwordSecret.name (optional — set when redis.host is provided and the + cache requires auth) + - redis.cluster: true (optional — chart sets REDIS_CLUSTER_NODES from + redis.host / redis.port so the proxy's Cache() + constructs a RedisClusterCache; the cluster client + discovers remaining nodes from CLUSTER SLOTS) + - Per-component extras (gateway / backend / ui): + - {component}.extraEnv / envConfigMaps / envSecrets (the latter two are lists of resource names → + envFrom configMapRef / secretRef) + - {component}.logLevel (renders as LITELLM_LOG) + - gateway.config.proxy_config (rendered into a ConfigMap and mounted at + /app/config/config.yaml; gateway reads it via + CONFIG_FILE_PATH) + - Enable ingress.enabled=true to dispatch / → ui, gateway data-plane prefixes → gateway, and the catch-all → backend. diff --git a/helm/litellm/templates/_helpers.tpl b/helm/litellm/templates/_helpers.tpl new file mode 100644 index 00000000000..4319907883e --- /dev/null +++ b/helm/litellm/templates/_helpers.tpl @@ -0,0 +1,263 @@ +{{/* +Common naming + label helpers shared by gateway, backend, and ui templates. +*/}} + +{{- define "litellm.name" -}} +{{- default .Chart.Name .Values.nameOverride | trunc 63 | trimSuffix "-" -}} +{{- end -}} + +{{- define "litellm.fullname" -}} +{{- if .Values.fullnameOverride -}} +{{- .Values.fullnameOverride | trunc 63 | trimSuffix "-" -}} +{{- else -}} +{{- $name := default .Chart.Name .Values.nameOverride -}} +{{- printf "%s-%s" .Release.Name $name | trunc 63 | trimSuffix "-" -}} +{{- end -}} +{{- end -}} + +{{- define "litellm.gateway.fullname" -}} +{{- printf "%s-gateway" (include "litellm.fullname" .) | trunc 63 | trimSuffix "-" -}} +{{- end -}} + +{{- define "litellm.backend.fullname" -}} +{{- printf "%s-backend" (include "litellm.fullname" .) | trunc 63 | trimSuffix "-" -}} +{{- end -}} + +{{- define "litellm.ui.fullname" -}} +{{- printf "%s-ui" (include "litellm.fullname" .) | trunc 63 | trimSuffix "-" -}} +{{- end -}} + +{{- define "litellm.commonLabels" -}} +app.kubernetes.io/name: {{ include "litellm.name" . }} +app.kubernetes.io/instance: {{ .Release.Name }} +app.kubernetes.io/managed-by: {{ .Release.Service }} +helm.sh/chart: {{ printf "%s-%s" .Chart.Name .Chart.Version | replace "+" "_" }} +{{- end -}} + +{{/* +Per-component selector labels — used in both Service selectors and Deployment matchLabels. +*/}} +{{- define "litellm.gateway.selectorLabels" -}} +app.kubernetes.io/name: {{ include "litellm.name" . }} +app.kubernetes.io/instance: {{ .Release.Name }} +app.kubernetes.io/component: gateway +{{- end -}} + +{{- define "litellm.backend.selectorLabels" -}} +app.kubernetes.io/name: {{ include "litellm.name" . }} +app.kubernetes.io/instance: {{ .Release.Name }} +app.kubernetes.io/component: backend +{{- end -}} + +{{- define "litellm.ui.selectorLabels" -}} +app.kubernetes.io/name: {{ include "litellm.name" . }} +app.kubernetes.io/instance: {{ .Release.Name }} +app.kubernetes.io/component: ui +{{- end -}} + +{{/* +Per-component ServiceAccount name helpers. + +Each component (gateway, backend, ui) has its own SA config under +.Values.serviceAccounts.. When `create` is true and `name` is +empty the chart defaults to "-litellm-". When `create` +is false the chart uses the provided name, or the namespace `default` SA. +*/}} +{{- define "litellm.gateway.serviceAccountName" -}} +{{- if .Values.serviceAccounts.gateway.create -}} +{{ default (include "litellm.gateway.fullname" .) .Values.serviceAccounts.gateway.name }} +{{- else -}} +{{ default "default" .Values.serviceAccounts.gateway.name }} +{{- end -}} +{{- end -}} + +{{- define "litellm.backend.serviceAccountName" -}} +{{- if .Values.serviceAccounts.backend.create -}} +{{ default (include "litellm.backend.fullname" .) .Values.serviceAccounts.backend.name }} +{{- else -}} +{{ default "default" .Values.serviceAccounts.backend.name }} +{{- end -}} +{{- end -}} + +{{- define "litellm.ui.serviceAccountName" -}} +{{- if .Values.serviceAccounts.ui.create -}} +{{ default (include "litellm.ui.fullname" .) .Values.serviceAccounts.ui.name }} +{{- else -}} +{{ default "default" .Values.serviceAccounts.ui.name }} +{{- end -}} +{{- end -}} + +{{/* +Master-key + database + redis env block — shared by gateway, backend, and the +migrations Job. + +Invoke with a dict: `(dict "root" $ "component" .Values.gateway)`. `root` is +the chart context (needed for .Values), `component` selects which component's +`extraEnv` / `logLevel` to render. + +Sensitive values (master key, DB username + password, Redis password) come +only from referenced Secrets; the chart never accepts inline values for them. + +The chart never assembles DATABASE_URL itself. It emits only the discrete +DATABASE_HOST/PORT/USER/NAME/SCHEMA (+ DATABASE_PASSWORD for password auth) +vars; the proxy's entrypoint (DatabaseURLSettings in +litellm/proxy/db/db_url_settings.py) builds the URL from them and +percent-encodes the credentials. Assembling the URL here via Kubernetes +`$(VAR)` substitution would embed the raw secret value, corrupting the URL +whenever the password contains a URL-reserved character (@, /, ?, %, +, +...) — as AWS RDS auto-generated passwords routinely do. + +When `database.writer.useIAMAuth: true`, the chart injects +IAM_TOKEN_DB_AUTH=true and omits DATABASE_PASSWORD — the entrypoint mints +the URL from DATABASE_HOST/PORT/USER/NAME plus a short-lived IAM token +instead of a static password. + +The read replica is opt-in via `database.reader.host`. The chart emits +DATABASE_HOST_READ_REPLICA / DATABASE_PORT_READ_REPLICA / +DATABASE_NAME_READ_REPLICA (+ DATABASE_SCHEMA_READ_REPLICA) for both auth +modes, plus DATABASE_USER_READ_REPLICA / DATABASE_PASSWORD_READ_REPLICA for +password auth. When `database.reader.useIAMAuth: true` it omits +DATABASE_PASSWORD_READ_REPLICA and the entrypoint mints the reader URL the +same way. Reader IAM only takes effect when the writer also uses IAM auth +(the proxy gates URL minting on IAM_TOKEN_DB_AUTH, which only the writer +sets). +*/}} +{{- define "litellm.serverEnv" -}} +{{- $root := .root -}} +{{- $component := .component -}} +- name: LITELLM_MASTER_KEY + valueFrom: + secretKeyRef: + name: {{ required "masterKey.secretName is required (the chart no longer accepts an inline master key)" $root.Values.masterKey.secretName }} + key: {{ $root.Values.masterKey.secretKey | default "master-key" }} +{{- if $component.logLevel }} +- name: LITELLM_LOG + value: {{ $component.logLevel | quote }} +{{- end }} +{{- with $root.Values.database.writer }} +- name: DATABASE_HOST + value: {{ required "database.writer.host is required" .host | quote }} +- name: DATABASE_PORT + value: {{ .port | default 5432 | quote }} +- name: DATABASE_USER + valueFrom: + secretKeyRef: + name: {{ required "database.writer.passwordSecret.name is required" .passwordSecret.name }} + key: {{ .passwordSecret.usernameKey | default "username" }} +- name: DATABASE_NAME + value: {{ required "database.writer.dbname is required" .dbname | quote }} +{{- if .schema }} +- name: DATABASE_SCHEMA + value: {{ .schema | quote }} +{{- end }} +{{- if .useIAMAuth }} +- name: IAM_TOKEN_DB_AUTH + value: "true" +{{- else }} +- name: DATABASE_PASSWORD + valueFrom: + secretKeyRef: + name: {{ .passwordSecret.name }} + key: {{ .passwordSecret.passwordKey | default "password" }} +{{- end }} +{{- end }} +{{- with $root.Values.database.reader }} +{{- if .host }} +{{- if and .useIAMAuth (not $root.Values.database.writer.useIAMAuth) }} +{{- fail "database.reader.useIAMAuth requires database.writer.useIAMAuth: true (the proxy gates IAM URL minting on IAM_TOKEN_DB_AUTH, which is only set by the writer)" }} +{{- end }} +- name: DATABASE_HOST_READ_REPLICA + value: {{ .host | quote }} +- name: DATABASE_PORT_READ_REPLICA + value: {{ .port | default 5432 | quote }} +- name: DATABASE_NAME_READ_REPLICA + value: {{ required "database.reader.dbname is required when database.reader.host is set" .dbname | quote }} +{{- if .schema }} +- name: DATABASE_SCHEMA_READ_REPLICA + value: {{ .schema | quote }} +{{- end }} +{{- if .useIAMAuth }} +{{- if .passwordSecret.name }} +- name: DATABASE_USER_READ_REPLICA + valueFrom: + secretKeyRef: + name: {{ .passwordSecret.name }} + key: {{ .passwordSecret.usernameKey | default "username" }} +{{- end }} +{{- else }} +{{- if not .passwordSecret.name }} +{{- fail "database.reader.passwordSecret.name is required when database.reader.host is set" }} +{{- end }} +- name: DATABASE_USER_READ_REPLICA + valueFrom: + secretKeyRef: + name: {{ .passwordSecret.name }} + key: {{ .passwordSecret.usernameKey | default "username" }} +- name: DATABASE_PASSWORD_READ_REPLICA + valueFrom: + secretKeyRef: + name: {{ .passwordSecret.name }} + key: {{ .passwordSecret.passwordKey | default "password" }} +{{- end }} +{{- end }} +{{- end }} +{{/* +The migrations Job (helm.sh/hook: pre-upgrade) is the single owner of +`prisma migrate deploy`. Without this, every gateway/backend pod also runs +Prisma schema-update on startup and contends with the Job — and with each +other — for Prisma's Postgres advisory lock on the writer, which makes the +Job's `migrate deploy` intermittently block until its per-attempt timeout +and retry-exhaust. The Job's entrypoint (migrations/run.py) does not import +proxy_server and never reads DISABLE_SCHEMA_UPDATE, so emitting it here is a +harmless no-op for the Job and authoritative for the app pods. +*/}} +- name: DISABLE_SCHEMA_UPDATE + value: "true" +{{- if $root.Values.redis.host }} +- name: REDIS_HOST + value: {{ $root.Values.redis.host | quote }} +- name: REDIS_PORT + value: {{ $root.Values.redis.port | quote }} +{{- if $root.Values.redis.passwordSecret.name }} +- name: REDIS_PASSWORD + valueFrom: + secretKeyRef: + name: {{ $root.Values.redis.passwordSecret.name }} + key: {{ $root.Values.redis.passwordSecret.passwordKey | default "password" }} +{{- end }} +{{- if $root.Values.redis.cluster }} +{{/* The proxy's Cache() reads REDIS_CLUSTER_NODES as JSON and constructs a + RedisClusterCache when it's set (litellm/caching/caching.py:169-192). + We seed with the single configured endpoint — the cluster client + discovers the remaining nodes from CLUSTER SLOTS at startup. */}} +- name: REDIS_CLUSTER_NODES + value: {{ printf "[{\"host\":%q,\"port\":%v}]" $root.Values.redis.host (int $root.Values.redis.port) | quote }} +{{- end }} +{{- end }} +{{- with $component.extraEnv }} +{{ toYaml . }} +{{- end }} +{{- end -}} + +{{/* +Renders `envFrom:` block for a component's `envConfigMaps` / `envSecrets` +lists. Each entry is a resource name; the chart wires the whole ConfigMap / +Secret into the container's env via configMapRef / secretRef. + +Invoke with just the component dict, e.g. `.Values.gateway`. Emits nothing +when both lists are empty so the container spec stays clean. +*/}} +{{- define "litellm.envFrom" -}} +{{- $component := . -}} +{{- if or $component.envConfigMaps $component.envSecrets }} +envFrom: +{{- range $component.envConfigMaps }} + - configMapRef: + name: {{ . }} +{{- end }} +{{- range $component.envSecrets }} + - secretRef: + name: {{ . }} +{{- end }} +{{- end }} +{{- end -}} diff --git a/helm/litellm/templates/backend/deployment.yaml b/helm/litellm/templates/backend/deployment.yaml new file mode 100644 index 00000000000..3b59c58c8bf --- /dev/null +++ b/helm/litellm/templates/backend/deployment.yaml @@ -0,0 +1,61 @@ +{{- if .Values.backend.enabled }} +apiVersion: apps/v1 +kind: Deployment +metadata: + name: {{ include "litellm.backend.fullname" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: backend +spec: + selector: + matchLabels: + {{- include "litellm.backend.selectorLabels" . | nindent 6 }} + template: + metadata: + {{- with .Values.backend.podAnnotations }} + annotations: + {{- toYaml . | nindent 8 }} + {{- end }} + labels: + {{- include "litellm.backend.selectorLabels" . | nindent 8 }} + spec: + serviceAccountName: {{ include "litellm.backend.serviceAccountName" . }} + automountServiceAccountToken: {{ .Values.serviceAccounts.backend.automount }} + {{- with .Values.imagePullSecrets }} + imagePullSecrets: + {{- toYaml . | nindent 8 }} + {{- end }} + containers: + - name: backend + image: "{{ .Values.backend.image.repository }}:{{ .Values.backend.image.tag | default .Chart.AppVersion }}" + imagePullPolicy: {{ .Values.backend.image.pullPolicy }} + ports: + - name: http + containerPort: 4001 + protocol: TCP + env: + {{- include "litellm.serverEnv" (dict "root" $ "component" .Values.backend) | nindent 12 }} + {{- include "litellm.envFrom" .Values.backend | nindent 10 }} + {{- with .Values.backend.livenessProbe }} + livenessProbe: + {{- toYaml . | nindent 12 }} + {{- end }} + {{- with .Values.backend.readinessProbe }} + readinessProbe: + {{- toYaml . | nindent 12 }} + {{- end }} + resources: + {{- toYaml .Values.backend.resources | nindent 12 }} + {{- with .Values.backend.nodeSelector }} + nodeSelector: + {{- toYaml . | nindent 8 }} + {{- end }} + {{- with .Values.backend.affinity }} + affinity: + {{- toYaml . | nindent 8 }} + {{- end }} + {{- with .Values.backend.tolerations }} + tolerations: + {{- toYaml . | nindent 8 }} + {{- end }} +{{- end }} diff --git a/helm/litellm/templates/backend/hpa.yaml b/helm/litellm/templates/backend/hpa.yaml new file mode 100644 index 00000000000..d02f011d0bb --- /dev/null +++ b/helm/litellm/templates/backend/hpa.yaml @@ -0,0 +1,33 @@ +{{- if and .Values.backend.enabled .Values.backend.hpa.enabled }} +apiVersion: autoscaling/v2 +kind: HorizontalPodAutoscaler +metadata: + name: {{ include "litellm.backend.fullname" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: backend +spec: + scaleTargetRef: + apiVersion: apps/v1 + kind: Deployment + name: {{ include "litellm.backend.fullname" . }} + minReplicas: {{ .Values.backend.hpa.minReplicas }} + maxReplicas: {{ .Values.backend.hpa.maxReplicas }} + metrics: + {{- if .Values.backend.hpa.targetCPUUtilizationPercentage }} + - type: Resource + resource: + name: cpu + target: + type: Utilization + averageUtilization: {{ .Values.backend.hpa.targetCPUUtilizationPercentage }} + {{- end }} + {{- if .Values.backend.hpa.targetMemoryUtilizationPercentage }} + - type: Resource + resource: + name: memory + target: + type: Utilization + averageUtilization: {{ .Values.backend.hpa.targetMemoryUtilizationPercentage }} + {{- end }} +{{- end }} diff --git a/helm/litellm/templates/backend/service.yaml b/helm/litellm/templates/backend/service.yaml new file mode 100644 index 00000000000..d480c654784 --- /dev/null +++ b/helm/litellm/templates/backend/service.yaml @@ -0,0 +1,18 @@ +{{- if .Values.backend.enabled }} +apiVersion: v1 +kind: Service +metadata: + name: {{ include "litellm.backend.fullname" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: backend +spec: + type: {{ .Values.backend.service.type }} + ports: + - port: {{ .Values.backend.service.port }} + targetPort: http + protocol: TCP + name: http + selector: + {{- include "litellm.backend.selectorLabels" . | nindent 4 }} +{{- end }} diff --git a/helm/litellm/templates/gateway/configmap.yaml b/helm/litellm/templates/gateway/configmap.yaml new file mode 100644 index 00000000000..d262bf25b87 --- /dev/null +++ b/helm/litellm/templates/gateway/configmap.yaml @@ -0,0 +1,9 @@ +{{- if .Values.gateway.config.create }} +apiVersion: v1 +kind: ConfigMap +metadata: + name: {{ include "litellm.gateway.fullname" . }}-config +data: + config.yaml: | +{{ .Values.gateway.config.proxy_config | toYaml | indent 6 }} +{{- end }} diff --git a/helm/litellm/templates/gateway/deployment.yaml b/helm/litellm/templates/gateway/deployment.yaml new file mode 100644 index 00000000000..05ea4052159 --- /dev/null +++ b/helm/litellm/templates/gateway/deployment.yaml @@ -0,0 +1,84 @@ +{{- if .Values.gateway.enabled }} +apiVersion: apps/v1 +kind: Deployment +metadata: + name: {{ include "litellm.gateway.fullname" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: gateway +spec: + selector: + matchLabels: + {{- include "litellm.gateway.selectorLabels" . | nindent 6 }} + template: + metadata: + annotations: + {{- if .Values.gateway.config.create }} + checksum/config: {{ include (print $.Template.BasePath "/gateway/configmap.yaml") . | sha256sum }} + {{- end }} + {{- with .Values.gateway.podAnnotations }} + {{- toYaml . | nindent 8 }} + {{- end }} + labels: + {{- include "litellm.gateway.selectorLabels" . | nindent 8 }} + spec: + serviceAccountName: {{ include "litellm.gateway.serviceAccountName" . }} + automountServiceAccountToken: {{ .Values.serviceAccounts.gateway.automount }} + {{- with .Values.imagePullSecrets }} + imagePullSecrets: + {{- toYaml . | nindent 8 }} + {{- end }} + containers: + - name: gateway + image: "{{ .Values.gateway.image.repository }}:{{ .Values.gateway.image.tag | default .Chart.AppVersion }}" + imagePullPolicy: {{ .Values.gateway.image.pullPolicy }} + ports: + - name: http + containerPort: 4000 + protocol: TCP + env: + {{- include "litellm.serverEnv" (dict "root" $ "component" .Values.gateway) | nindent 12 }} + {{- if .Values.gateway.config.create }} + - name: CONFIG_FILE_PATH + value: /app/config/config.yaml + {{- end }} + {{- if .Values.gateway.numWorkers }} + - name: NUM_WORKERS + value: {{ .Values.gateway.numWorkers | quote }} + {{- end }} + {{- include "litellm.envFrom" .Values.gateway | nindent 10 }} + {{- if .Values.gateway.config.create }} + volumeMounts: + - name: gateway-config + mountPath: /app/config/config.yaml + subPath: config.yaml + {{- end }} + {{- with .Values.gateway.livenessProbe }} + livenessProbe: + {{- toYaml . | nindent 12 }} + {{- end }} + {{- with .Values.gateway.readinessProbe }} + readinessProbe: + {{- toYaml . | nindent 12 }} + {{- end }} + resources: + {{- toYaml .Values.gateway.resources | nindent 12 }} + {{- if .Values.gateway.config.create }} + volumes: + - name: gateway-config + configMap: + name: {{ include "litellm.gateway.fullname" . }}-config + {{- end }} + {{- with .Values.gateway.nodeSelector }} + nodeSelector: + {{- toYaml . | nindent 8 }} + {{- end }} + {{- with .Values.gateway.affinity }} + affinity: + {{- toYaml . | nindent 8 }} + {{- end }} + {{- with .Values.gateway.tolerations }} + tolerations: + {{- toYaml . | nindent 8 }} + {{- end }} +{{- end }} diff --git a/helm/litellm/templates/gateway/hpa.yaml b/helm/litellm/templates/gateway/hpa.yaml new file mode 100644 index 00000000000..27c4f05ba59 --- /dev/null +++ b/helm/litellm/templates/gateway/hpa.yaml @@ -0,0 +1,33 @@ +{{- if and .Values.gateway.enabled .Values.gateway.hpa.enabled }} +apiVersion: autoscaling/v2 +kind: HorizontalPodAutoscaler +metadata: + name: {{ include "litellm.gateway.fullname" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: gateway +spec: + scaleTargetRef: + apiVersion: apps/v1 + kind: Deployment + name: {{ include "litellm.gateway.fullname" . }} + minReplicas: {{ .Values.gateway.hpa.minReplicas }} + maxReplicas: {{ .Values.gateway.hpa.maxReplicas }} + metrics: + {{- if .Values.gateway.hpa.targetCPUUtilizationPercentage }} + - type: Resource + resource: + name: cpu + target: + type: Utilization + averageUtilization: {{ .Values.gateway.hpa.targetCPUUtilizationPercentage }} + {{- end }} + {{- if .Values.gateway.hpa.targetMemoryUtilizationPercentage }} + - type: Resource + resource: + name: memory + target: + type: Utilization + averageUtilization: {{ .Values.gateway.hpa.targetMemoryUtilizationPercentage }} + {{- end }} +{{- end }} diff --git a/helm/litellm/templates/gateway/service.yaml b/helm/litellm/templates/gateway/service.yaml new file mode 100644 index 00000000000..03a4167a0ab --- /dev/null +++ b/helm/litellm/templates/gateway/service.yaml @@ -0,0 +1,18 @@ +{{- if .Values.gateway.enabled }} +apiVersion: v1 +kind: Service +metadata: + name: {{ include "litellm.gateway.fullname" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: gateway +spec: + type: {{ .Values.gateway.service.type }} + ports: + - port: {{ .Values.gateway.service.port }} + targetPort: http + protocol: TCP + name: http + selector: + {{- include "litellm.gateway.selectorLabels" . | nindent 4 }} +{{- end }} diff --git a/helm/litellm/templates/ingress.yaml b/helm/litellm/templates/ingress.yaml new file mode 100644 index 00000000000..30a8e7c974b --- /dev/null +++ b/helm/litellm/templates/ingress.yaml @@ -0,0 +1,153 @@ +{{- if .Values.ingress.enabled -}} +{{- $gatewayName := include "litellm.gateway.fullname" . -}} +{{- $backendName := include "litellm.backend.fullname" . -}} +{{- $uiName := include "litellm.ui.fullname" . -}} +{{- $gatewayPort := .Values.gateway.service.port -}} +{{- $backendPort := .Values.backend.service.port -}} +{{- $uiPort := .Values.ui.service.port -}} +{{/* + Gateway data-plane prefixes — must mirror gateway/routes/allowlist.py. + Versioned paths are listed explicitly to avoid routing management routes + (e.g. /v1/access_group, /v2/key/info, /v1/tool/*, /v1/agents, /v1/workflows, + /v2/user/info, /v2/team/list, /v2/model/info, /v2/login, /v2/guardrails/*, + /v1/mcp/*) onto the gateway via a broad /v1 or /v2 prefix. +*/}} +{{- $gatewayPrefixes := list + "/v1/chat" "/chat" "/v1/completions" "/completions" "/v1/embeddings" "/embeddings" + "/v1/moderations" "/moderations" "/v1/audio" "/audio" "/v1/images" "/images" + "/v1/files" "/files" "/v1/batches" "/batches" "/v1/fine_tuning" "/fine_tuning" + "/v1/fine-tuning" "/fine-tuning" "/v1/responses" "/responses" "/v1/threads" "/threads" + "/v1/assistants" "/assistants" "/v1/vector_stores" "/vector_stores" "/v1/indexes" + "/v1/models" "/models" "/openai" "/engines" + "/v1/messages" "/messages" "/v1/skills" "/v1/a2a" + "/v1/rerank" "/v2/rerank" "/rerank" "/v1/ocr" "/ocr" "/v1/rag" "/rag" + "/v1/video" "/v1/videos" "/video" "/videos" "/v1/search" "/search" + "/v1/containers" "/containers" "/v1/evals" "/v1/memory" "/queue/chat" + "/v1beta" "/interactions" + "/anthropic" "/azure" "/azure_ai" "/aws" "/bedrock" "/cohere" "/gemini" "/google" + "/vertex_ai" "/vertex-ai" "/assemblyai" "/eu.assemblyai" "/langfuse" "/vllm" + "/mistral" "/groq" "/voyage" "/cursor" "/milvus" "/openai_passthrough" + "/toolset" + "/v1/realtime" "/realtime" + "/health" "/metrics" +-}} +{{/* + /test is gateway-only as an EXACT path (GATEWAY_EXACT_PATHS), but its + children /test/connection and /test/tools/list are MCP-server management + endpoints kept only on the backend ("/test/" in BACKEND_PATH_PREFIXES). + A Prefix match here would route /test/* to the gateway, which trims those + routes at startup -> 404. So /test is rendered as a standalone Exact path + and /test/* falls through to the backend catch-all. +*/}} +apiVersion: networking.k8s.io/v1 +kind: Ingress +metadata: + name: {{ include "litellm.fullname" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + {{- with .Values.ingress.annotations }} + annotations: + {{- toYaml . | nindent 4 }} + {{- end }} +spec: + {{- with .Values.ingress.className }} + ingressClassName: {{ . | quote }} + {{- end }} + {{- with .Values.ingress.tls }} + tls: + {{- toYaml . | nindent 4 }} + {{- end }} + rules: + - {{- with .Values.ingress.host }} + host: {{ . | quote }} + {{- end }} + http: + paths: + # --- UI (Next.js static export) --- + - path: / + pathType: Exact + backend: + service: + name: {{ $uiName }} + port: + number: {{ $uiPort }} + - path: /favicon.ico + pathType: Exact + backend: + service: + name: {{ $uiName }} + port: + number: {{ $uiPort }} + - path: /litellm-asset-prefix + pathType: Prefix + backend: + service: + name: {{ $uiName }} + port: + number: {{ $uiPort }} + - path: /_next + pathType: Prefix + backend: + service: + name: {{ $uiName }} + port: + number: {{ $uiPort }} + # /ui/* is where the Next.js SPA serves its login + dashboard + # routes (e.g. /ui/login). Without this, /ui/* falls into the + # catch-all → backend → 404. + - path: /ui + pathType: Prefix + backend: + service: + name: {{ $uiName }} + port: + number: {{ $uiPort }} + # Next.js App Router (output: "export", basePath: "") emits the + # RSC/flight payload for every route as a ROOT-level .txt + # (/index.txt, /teams.txt, /__next._tree.txt, ...). The client + # router fetches these on every soft navigation / prefetch as + # .txt?_rsc= (the query string is irrelevant to path + # matching). They are not under /ui, /_next, or + # /litellm-asset-prefix, so without this rule they fall to the + # backend catch-all → 404 → client-side navigation never settles + # and the login flow spins in an infinite redirect loop + # (/ ⇄ /ui/login). ui/nginx.conf already serves *.txt from the + # export; this rule only routes the request to it. Needs an + # ingress controller whose ImplementationSpecific path is a + # wildcard pattern (AWS ALB: `*` = 0+ chars); this chart targets + # the AWS Load Balancer Controller. + - path: /*.txt + pathType: ImplementationSpecific + backend: + service: + name: {{ $uiName }} + port: + number: {{ $uiPort }} + # --- Gateway data plane --- + # Exact /test only (see the $gatewayPrefixes comment above); + # /test/* MCP management endpoints fall to the backend catch-all. + - path: /test + pathType: Exact + backend: + service: + name: {{ $gatewayName }} + port: + number: {{ $gatewayPort }} + {{- range $gatewayPrefixes }} + - path: {{ . }} + pathType: Prefix + backend: + service: + name: {{ $gatewayName }} + port: + number: {{ $gatewayPort }} + {{- end }} + # --- Catch-all → backend (management API: /key/*, /user/*, /team/*, ...) --- + - path: / + pathType: Prefix + backend: + service: + name: {{ $backendName }} + port: + number: {{ $backendPort }} +{{- end }} diff --git a/helm/litellm/templates/migrations-job.yaml b/helm/litellm/templates/migrations-job.yaml new file mode 100644 index 00000000000..92671388546 --- /dev/null +++ b/helm/litellm/templates/migrations-job.yaml @@ -0,0 +1,46 @@ +{{- if .Values.migrationJob.enabled -}} +# Pre-install / pre-upgrade hook that runs `prisma migrate deploy` against +# the writer database before the gateway and backend Deployments are rolled +# out. Required because the gateway and backend both spin up Prisma at +# startup and assume the LiteLLM schema (LiteLLM_Config, +# LiteLLM_VerificationToken, LiteLLM_SpendLogs, ...) already exists. +# +# Running this pre-upgrade closes the window where new application pods would +# otherwise serve traffic against the previous release's unmigrated schema. +apiVersion: batch/v1 +kind: Job +metadata: + name: {{ include "litellm.fullname" . }}-migrations + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: migrations + annotations: + helm.sh/hook: pre-install,pre-upgrade + helm.sh/hook-delete-policy: before-hook-creation + helm.sh/hook-weight: "0" +spec: + backoffLimit: {{ .Values.migrationJob.backoffLimit }} + ttlSecondsAfterFinished: {{ .Values.migrationJob.ttlSecondsAfterFinished }} + template: + metadata: + labels: + {{- include "litellm.commonLabels" . | nindent 8 }} + app.kubernetes.io/component: migrations + spec: + restartPolicy: Never + serviceAccountName: {{ include "litellm.backend.serviceAccountName" . }} + {{- with .Values.imagePullSecrets }} + imagePullSecrets: + {{- toYaml . | nindent 8 }} + {{- end }} + containers: + - name: prisma-migrations + image: "{{ .Values.migrationJob.image.repository }}:{{ .Values.migrationJob.image.tag | default .Chart.AppVersion }}" + imagePullPolicy: {{ .Values.migrationJob.image.pullPolicy }} + env: + {{- include "litellm.serverEnv" (dict "root" $ "component" .Values.migrationJob) | nindent 12 }} + {{- with .Values.migrationJob.resources }} + resources: + {{- toYaml . | nindent 12 }} + {{- end }} +{{- end }} diff --git a/helm/litellm/templates/serviceaccount.yaml b/helm/litellm/templates/serviceaccount.yaml new file mode 100644 index 00000000000..a2fc52f47c0 --- /dev/null +++ b/helm/litellm/templates/serviceaccount.yaml @@ -0,0 +1,51 @@ +{{- $prev := false -}} +{{- if .Values.serviceAccounts.gateway.create -}} +{{- $prev = true }} +apiVersion: v1 +kind: ServiceAccount +metadata: + name: {{ include "litellm.gateway.serviceAccountName" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: gateway + {{- with .Values.serviceAccounts.gateway.annotations }} + annotations: + {{- toYaml . | nindent 4 }} + {{- end }} +automountServiceAccountToken: {{ .Values.serviceAccounts.gateway.automount }} +{{- end }} +{{- if .Values.serviceAccounts.backend.create }} +{{- if $prev }} +--- +{{- end }} +{{- $prev = true }} +apiVersion: v1 +kind: ServiceAccount +metadata: + name: {{ include "litellm.backend.serviceAccountName" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: backend + {{- with .Values.serviceAccounts.backend.annotations }} + annotations: + {{- toYaml . | nindent 4 }} + {{- end }} +automountServiceAccountToken: {{ .Values.serviceAccounts.backend.automount }} +{{- end }} +{{- if .Values.serviceAccounts.ui.create }} +{{- if $prev }} +--- +{{- end }} +apiVersion: v1 +kind: ServiceAccount +metadata: + name: {{ include "litellm.ui.serviceAccountName" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: ui + {{- with .Values.serviceAccounts.ui.annotations }} + annotations: + {{- toYaml . | nindent 4 }} + {{- end }} +automountServiceAccountToken: {{ .Values.serviceAccounts.ui.automount }} +{{- end }} diff --git a/helm/litellm/templates/ui/deployment.yaml b/helm/litellm/templates/ui/deployment.yaml new file mode 100644 index 00000000000..b40b44cca53 --- /dev/null +++ b/helm/litellm/templates/ui/deployment.yaml @@ -0,0 +1,71 @@ +{{- if .Values.ui.enabled }} +apiVersion: apps/v1 +kind: Deployment +metadata: + name: {{ include "litellm.ui.fullname" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: ui +spec: + selector: + matchLabels: + {{- include "litellm.ui.selectorLabels" . | nindent 6 }} + template: + metadata: + {{- with .Values.ui.podAnnotations }} + annotations: + {{- toYaml . | nindent 8 }} + {{- end }} + labels: + {{- include "litellm.ui.selectorLabels" . | nindent 8 }} + spec: + serviceAccountName: {{ include "litellm.ui.serviceAccountName" . }} + automountServiceAccountToken: {{ .Values.serviceAccounts.ui.automount }} + {{- with .Values.imagePullSecrets }} + imagePullSecrets: + {{- toYaml . | nindent 8 }} + {{- end }} + containers: + - name: ui + image: "{{ .Values.ui.image.repository }}:{{ .Values.ui.image.tag | default .Chart.AppVersion }}" + imagePullPolicy: {{ .Values.ui.image.pullPolicy }} + ports: + - name: http + containerPort: 3000 + protocol: TCP + env: + {{- if .Values.ui.logLevel }} + - name: LITELLM_LOG + value: {{ .Values.ui.logLevel | quote }} + {{- end }} + {{- if .Values.ui.backendUrl }} + - name: LITELLM_BACKEND_URL + value: {{ .Values.ui.backendUrl | quote }} + {{- end }} + {{- with .Values.ui.extraEnv }} + {{- toYaml . | nindent 12 }} + {{- end }} + {{- include "litellm.envFrom" .Values.ui | nindent 10 }} + {{- with .Values.ui.livenessProbe }} + livenessProbe: + {{- toYaml . | nindent 12 }} + {{- end }} + {{- with .Values.ui.readinessProbe }} + readinessProbe: + {{- toYaml . | nindent 12 }} + {{- end }} + resources: + {{- toYaml .Values.ui.resources | nindent 12 }} + {{- with .Values.ui.nodeSelector }} + nodeSelector: + {{- toYaml . | nindent 8 }} + {{- end }} + {{- with .Values.ui.affinity }} + affinity: + {{- toYaml . | nindent 8 }} + {{- end }} + {{- with .Values.ui.tolerations }} + tolerations: + {{- toYaml . | nindent 8 }} + {{- end }} +{{- end }} diff --git a/helm/litellm/templates/ui/hpa.yaml b/helm/litellm/templates/ui/hpa.yaml new file mode 100644 index 00000000000..b43eda5ac4a --- /dev/null +++ b/helm/litellm/templates/ui/hpa.yaml @@ -0,0 +1,33 @@ +{{- if and .Values.ui.enabled .Values.ui.hpa.enabled }} +apiVersion: autoscaling/v2 +kind: HorizontalPodAutoscaler +metadata: + name: {{ include "litellm.ui.fullname" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: ui +spec: + scaleTargetRef: + apiVersion: apps/v1 + kind: Deployment + name: {{ include "litellm.ui.fullname" . }} + minReplicas: {{ .Values.ui.hpa.minReplicas }} + maxReplicas: {{ .Values.ui.hpa.maxReplicas }} + metrics: + {{- if .Values.ui.hpa.targetCPUUtilizationPercentage }} + - type: Resource + resource: + name: cpu + target: + type: Utilization + averageUtilization: {{ .Values.ui.hpa.targetCPUUtilizationPercentage }} + {{- end }} + {{- if .Values.ui.hpa.targetMemoryUtilizationPercentage }} + - type: Resource + resource: + name: memory + target: + type: Utilization + averageUtilization: {{ .Values.ui.hpa.targetMemoryUtilizationPercentage }} + {{- end }} +{{- end }} diff --git a/helm/litellm/templates/ui/service.yaml b/helm/litellm/templates/ui/service.yaml new file mode 100644 index 00000000000..52b539fa00c --- /dev/null +++ b/helm/litellm/templates/ui/service.yaml @@ -0,0 +1,18 @@ +{{- if .Values.ui.enabled }} +apiVersion: v1 +kind: Service +metadata: + name: {{ include "litellm.ui.fullname" . }} + labels: + {{- include "litellm.commonLabels" . | nindent 4 }} + app.kubernetes.io/component: ui +spec: + type: {{ .Values.ui.service.type }} + ports: + - port: {{ .Values.ui.service.port }} + targetPort: http + protocol: TCP + name: http + selector: + {{- include "litellm.ui.selectorLabels" . | nindent 4 }} +{{- end }} diff --git a/helm/litellm/values.yaml b/helm/litellm/values.yaml new file mode 100644 index 00000000000..934661643bd --- /dev/null +++ b/helm/litellm/values.yaml @@ -0,0 +1,242 @@ +# LiteLLM helm chart values + +nameOverride: "" +fullnameOverride: "" + +imagePullSecrets: [] + +# Optional Ingress wiring the three component Services behind a single L7 +# entrypoint. Required when serving the static UI bundle over the network. +ingress: + enabled: false + className: "" + annotations: {} + host: "" # optional; if set, becomes the rule's host + tls: [] + +# Per-component ServiceAccounts for gateway, backend, and ui. +# +# Each section mirrors the old shared serviceAccount shape. Set `create: +# true` to have the chart provision the SA (useful for EKS Pod Identity / +# GKE Workload Identity annotations). Set `name` to bind an existing SA. +# When both are unset the component pod runs with the namespace `default` SA. +# +# The UI SA deliberately defaults to `automount: false` — the static nginx +# container does not need the K8s API and should not carry a projected +# ServiceAccount token that a compromised container could use to call the +# cloud-provider metadata service or the K8s API. +serviceAccounts: + gateway: + create: false + automount: true + annotations: {} + name: "" + backend: + create: false + automount: true + annotations: {} + name: "" + ui: + create: false + automount: false + annotations: {} + name: "" + +# Pre-install / pre-upgrade Helm hook that runs `prisma migrate deploy` +# against the writer database, creating the LiteLLM schema (tables that +# gateway + backend assume exist at startup: LiteLLM_Config, +# LiteLLM_VerificationToken, LiteLLM_SpendLogs, ...). Disable if your +# pipeline runs migrations out-of-band. +# +# Uses a dedicated `litellm-migrations` image (prisma CLI + the migration +# files from `litellm-proxy-extras`) instead of the backend image, so the +# Job doesn't drag in the rest of the proxy and doesn't run `prisma +# generate` — the migration engine doesn't need the generated client. +migrationJob: + enabled: true + backoffLimit: 4 + ttlSecondsAfterFinished: 120 + resources: {} + image: + repository: ghcr.io/berriai/litellm-migrations + tag: "" # defaults to .Chart.AppVersion + pullPolicy: IfNotPresent + # Extra env appended to the migration container. The migration entrypoint + # uses the v2 resolver by default (no diff-and-force recovery — avoids the + # schema thrashing seen during rolling deploys). To opt back into the v1 + # resolver, append `- name: USE_V2_MIGRATION_RESOLVER` / `value: "false"`. + extraEnv: [] + +# Required: a master key used by gateway + backend to mint/verify proxy tokens. +# Must reference an existing Secret. +masterKey: + secretName: litellm-master-key-secret # name of a Secret containing the master key + secretKey: master-key + +# External Postgres connection. +database: + writer: + host: "" + port: 5432 + dbname: "" + schema: "" + useIAMAuth: false + passwordSecret: + name: litellm-writer-secret + usernameKey: username + passwordKey: password + + # Optional read-replica routing. When `reader.host` is set, the proxy routes + # reads (find_*, count, group_by, query_raw/_first) to this endpoint while + # writes stay on the writer. Leave `reader.host` empty to disable. + reader: + host: "" + port: 5432 + dbname: "" + schema: "" + useIAMAuth: false + passwordSecret: + name: litellm-reader-secret + usernameKey: username + passwordKey: password + +# Optional Redis (caching, rate limiting). Leave host empty to disable. +# +# Set `cluster: true` for Redis Cluster mode (e.g. AWS ElastiCache Cluster, +# self-hosted Redis Cluster). The chart emits REDIS_CLUSTER_NODES from +# `host` / `port` as the single seed; the cluster client discovers the +# remaining nodes from CLUSTER SLOTS at startup. +redis: + cluster: false + host: "" + port: 6379 + passwordSecret: + name: "" # Leave empty for auth-less Redis + passwordKey: password + +# ---------- gateway (LLM data plane) ---------- +gateway: + enabled: true + logLevel: INFO + # Number of uvicorn worker processes per gateway pod. Sets NUM_WORKERS, + # consumed by the gateway image entrypoint. Default is 1. + numWorkers: 1 + extraEnv: [] # Add extra environment variables to the gateway + envConfigMaps: [] # Add extra environment variables to the gateway from config maps + envSecrets: [] # Add extra environment variables to the gateway from secrets + config: + create: true + proxy_config: {} + image: + repository: ghcr.io/berriai/litellm-gateway + tag: "" # defaults to .Chart.AppVersion + pullPolicy: IfNotPresent + service: + type: ClusterIP + port: 4000 + resources: + requests: + cpu: "1" + memory: 4Gi + limits: + cpu: "2" + memory: 4Gi + livenessProbe: + httpGet: { path: /health/liveliness, port: http } + initialDelaySeconds: 10 + periodSeconds: 15 + readinessProbe: + httpGet: { path: /health/readiness, port: http } + initialDelaySeconds: 5 + periodSeconds: 10 + hpa: + enabled: true + minReplicas: 1 + maxReplicas: 10 + targetCPUUtilizationPercentage: 70 + targetMemoryUtilizationPercentage: 80 + podAnnotations: {} + nodeSelector: {} + tolerations: [] + affinity: {} + +# ---------- backend (UI / management API) ---------- +backend: + enabled: true + logLevel: INFO + extraEnv: [] + envConfigMaps: [] + envSecrets: [] + image: + repository: ghcr.io/berriai/litellm-backend + tag: "" + pullPolicy: IfNotPresent + service: + type: ClusterIP + port: 4001 + resources: + requests: + cpu: "1" + memory: 4Gi + limits: + cpu: "2" + memory: 4Gi + livenessProbe: + httpGet: { path: /health/liveliness, port: http } + initialDelaySeconds: 10 + periodSeconds: 15 + readinessProbe: + httpGet: { path: /health/readiness, port: http } + initialDelaySeconds: 5 + periodSeconds: 10 + hpa: + enabled: true + minReplicas: 1 + maxReplicas: 4 + targetCPUUtilizationPercentage: 70 + podAnnotations: {} + nodeSelector: {} + tolerations: [] + affinity: {} + +# ---------- ui (Next.js static dashboard) ---------- +ui: + enabled: true + logLevel: INFO + extraEnv: [] + envConfigMaps: [] + envSecrets: [] + image: + repository: ghcr.io/berriai/litellm-ui + tag: "" + pullPolicy: IfNotPresent + service: + type: ClusterIP + port: 3000 + # The dashboard expects to know where to reach the backend API. Set this to + # the externally-routable URL (typically the ingress host + /api or similar). + backendUrl: "" + resources: + requests: + cpu: 500m + memory: 500Mi + limits: + cpu: "1" + memory: 1Gi + livenessProbe: + httpGet: { path: /, port: http } + initialDelaySeconds: 5 + periodSeconds: 20 + readinessProbe: + httpGet: { path: /, port: http } + initialDelaySeconds: 2 + periodSeconds: 10 + hpa: + enabled: false + minReplicas: 1 + maxReplicas: 3 + targetCPUUtilizationPercentage: 80 + podAnnotations: {} + nodeSelector: {} + tolerations: [] + affinity: {} diff --git a/license_cache.json b/license_cache.json index 4b09afacaa3..dc061b48f4f 100644 --- a/license_cache.json +++ b/license_cache.json @@ -49,5 +49,13 @@ "grpc-google-iam-v1:0.14.3": "Apache 2.0", "h11:0.16.0": "MIT", "requests-toolbelt:1.0.0": "Apache 2.0", - "tornado:6.5.4": "Apache-2.0" + "tornado:6.5.4": "Apache-2.0", + "granian:2.5.7": "BSD-3-Clause", + "mlflow:3.11.1": "Copyright 2018 Databricks, Inc. All rights reserved.\n \n \t\t\t\tApache License\n Version 2.0, January 2004\n http://www.apache.org/licenses/\n \n TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION\n \n 1. Definitions.\n \n \"License\" shall mean the terms and conditions for use, reproduction,\n and distribution as defined by Sections 1 through 9 of this document.\n \n \"Licensor\" shall mean the copyright owner or entity authorized by\n the copyright owner that is granting the License.\n \n \"Legal Entity\" shall mean the union of the acting entity and all\n other entities that control, are controlled by, or are under common\n control with that entity. For the purposes of this definition,\n \"control\" means (i) the power, direct or indirect, to cause the\n direction or management of such entity, whether by contract or\n otherwise, or (ii) ownership of fifty percent (50%) or more of the\n outstanding shares, or (iii) beneficial ownership of such entity.\n \n \"You\" (or \"Your\") shall mean an individual or Legal Entity\n exercising permissions granted by this License.\n \n \"Source\" form shall mean the preferred form for making modifications,\n including but not limited to software source code, documentation\n source, and configuration files.\n \n \"Object\" form shall mean any form resulting from mechanical\n transformation or translation of a Source form, including but\n not limited to compiled object code, generated documentation,\n and conversions to other media types.\n \n \"Work\" shall mean the work of authorship, whether in Source or\n Object form, made available under the License, as indicated by a\n copyright notice that is included in or attached to the work\n (an example is provided in the Appendix below).\n \n \"Derivative Works\" shall mean any work, whether in Source or Object\n form, that is based on (or derived from) the Work and for which the\n editorial revisions, annotations, elaborations, or other modifications\n represent, as a whole, an original work of authorship. For the purposes\n of this License, Derivative Works shall not include works that remain\n separable from, or merely link (or bind by name) to the interfaces of,\n the Work and Derivative Works thereof.\n \n \"Contribution\" shall mean any work of authorship, including\n the original version of the Work and any modifications or additions\n to that Work or Derivative Works thereof, that is intentionally\n submitted to Licensor for inclusion in the Work by the copyright owner\n or by an individual or Legal Entity authorized to submit on behalf of\n the copyright owner. For the purposes of this definition, \"submitted\"\n means any form of electronic, verbal, or written communication sent\n to the Licensor or its representatives, including but not limited to\n communication on electronic mailing lists, source code control systems,\n and issue tracking systems that are managed by, or on behalf of, the\n Licensor for the purpose of discussing and improving the Work, but\n excluding communication that is conspicuously marked or otherwise\n designated in writing by the copyright owner as \"Not a Contribution.\"\n \n \"Contributor\" shall mean Licensor and any individual or Legal Entity\n on behalf of whom a Contribution has been received by Licensor and\n subsequently incorporated within the Work.\n \n 2. Grant of Copyright License. Subject to the terms and conditions of\n this License, each Contributor hereby grants to You a perpetual,\n worldwide, non-exclusive, no-charge, royalty-free, irrevocable\n copyright license to reproduce, prepare Derivative Works of,\n publicly display, publicly perform, sublicense, and distribute the\n Work and such Derivative Works in Source or Object form.\n \n 3. Grant of Patent License. Subject to the terms and conditions of\n this License, each Contributor hereby grants to You a perpetual,\n worldwide, non-exclusive, no-charge, royalty-free, irrevocable\n (except as stated in this section) patent license to make, have made,\n use, offer to sell, sell, import, and otherwise transfer the Work,\n where such license applies only to those patent claims licensable\n by such Contributor that are necessarily infringed by their\n Contribution(s) alone or by combination of their Contribution(s)\n with the Work to which such Contribution(s) was submitted. If You\n institute patent litigation against any entity (including a\n cross-claim or counterclaim in a lawsuit) alleging that the Work\n or a Contribution incorporated within the Work constitutes direct\n or contributory patent infringement, then any patent licenses\n granted to You under this License for that Work shall terminate\n as of the date such litigation is filed.\n \n 4. Redistribution. You may reproduce and distribute copies of the\n Work or Derivative Works thereof in any medium, with or without\n modifications, and in Source or Object form, provided that You\n meet the following conditions:\n \n (a) You must give any other recipients of the Work or\n Derivative Works a copy of this License; and\n \n (b) You must cause any modified files to carry prominent notices\n stating that You changed the files; and\n \n (c) You must retain, in the Source form of any Derivative Works\n that You distribute, all copyright, patent, trademark, and\n attribution notices from the Source form of the Work,\n excluding those notices that do not pertain to any part of\n the Derivative Works; and\n \n (d) If the Work includes a \"NOTICE\" text file as part of its\n distribution, then any Derivative Works that You distribute must\n include a readable copy of the attribution notices contained\n within such NOTICE file, excluding those notices that do not\n pertain to any part of the Derivative Works, in at least one\n of the following places: within a NOTICE text file distributed\n as part of the Derivative Works; within the Source form or\n documentation, if provided along with the Derivative Works; or,\n within a display generated by the Derivative Works, if and\n wherever such third-party notices normally appear. The contents\n of the NOTICE file are for informational purposes only and\n do not modify the License. You may add Your own attribution\n notices within Derivative Works that You distribute, alongside\n or as an addendum to the NOTICE text from the Work, provided\n that such additional attribution notices cannot be construed\n as modifying the License.\n \n You may add Your own copyright statement to Your modifications and\n may provide additional or different license terms and conditions\n for use, reproduction, or distribution of Your modifications, or\n for any such Derivative Works as a whole, provided Your use,\n reproduction, and distribution of the Work otherwise complies with\n the conditions stated in this License.\n \n 5. Submission of Contributions. Unless You explicitly state otherwise,\n any Contribution intentionally submitted for inclusion in the Work\n by You to the Licensor shall be under the terms and conditions of\n this License, without any additional terms or conditions.\n Notwithstanding the above, nothing herein shall supersede or modify\n the terms of any separate license agreement you may have executed\n with Licensor regarding such Contributions.\n \n 6. Trademarks. This License does not grant permission to use the trade\n names, trademarks, service marks, or product names of the Licensor,\n except as required for reasonable and customary use in describing the\n origin of the Work and reproducing the content of the NOTICE file.\n \n 7. Disclaimer of Warranty. Unless required by applicable law or\n agreed to in writing, Licensor provides the Work (and each\n Contributor provides its Contributions) on an \"AS IS\" BASIS,\n WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or\n implied, including, without limitation, any warranties or conditions\n of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A\n PARTICULAR PURPOSE. You are solely responsible for determining the\n appropriateness of using or redistributing the Work and assume any\n risks associated with Your exercise of permissions under this License.\n \n 8. Limitation of Liability. In no event and under no legal theory,\n whether in tort (including negligence), contract, or otherwise,\n unless required by applicable law (such as deliberate and grossly\n negligent acts) or agreed to in writing, shall any Contributor be\n liable to You for damages, including any direct, indirect, special,\n incidental, or consequential damages of any character arising as a\n result of this License or out of the use or inability to use the\n Work (including but not limited to damages for loss of goodwill,\n work stoppage, computer failure or malfunction, or any and all\n other commercial damages or losses), even if such Contributor\n has been advised of the possibility of such damages.\n \n 9. Accepting Warranty or Additional Liability. While redistributing\n the Work or Derivative Works thereof, You may choose to offer,\n and charge a fee for, acceptance of support, warranty, indemnity,\n or other liability obligations and/or rights consistent with this\n License. However, in accepting such obligations, You may act only\n on Your own behalf and on Your sole responsibility, not on behalf\n of any other Contributor, and only if You agree to indemnify,\n defend, and hold each Contributor harmless for any liability\n incurred by, or claims asserted against, such Contributor by reason\n of your accepting any such warranty or additional liability.\n \n END OF TERMS AND CONDITIONS\n APPENDIX: How to apply the Apache License to your work.\n \n To apply the Apache License to your work, attach the following\n boilerplate notice, with the fields enclosed by brackets \"[]\"\n replaced with your own identifying information. (Don't include\n the brackets!) The text should be enclosed in the appropriate\n comment syntax for the file format. We also recommend that a\n file or class name and description of purpose be included on the\n same \"printed page\" as the copyright notice for easier\n identification within third-party archives.\n \n Copyright [yyyy] [name of copyright owner]\n \n Licensed under the Apache License, Version 2.0 (the \"License\");\n you may not use this file except in compliance with the License.\n You may obtain a copy of the License at\n \n http://www.apache.org/licenses/LICENSE-2.0\n \n Unless required by applicable law or agreed to in writing, software\n distributed under the License is distributed on an \"AS IS\" BASIS,\n WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n See the License for the specific language governing permissions and\n limitations under the License.\n ", + "nvidia-riva-client:2.15.0": "MIT", + "numpy:1.26.0": "Copyright (c) 2005-2023, NumPy Developers. All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. * Neither the name of the NumPy Developers nor the names of any contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS \"AS IS\" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. ---- This binary distribution of NumPy also bundles the following software: Name: GCC runtime library Files: .dylibs/* Description: dynamically linked to files compiled with gcc Availability: https://gcc.gnu.org/viewcvs/gcc/ License: GPLv3 + runtime exception Copyright (C) 2002-2017 Free Software Foundation, Inc. Libgfortran is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3, or (at your option) any later version. Libgfortran is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. Under Section 7 of GPL version 3, you are granted additional permissions described in the GCC Runtime Library Exception, version 3.1, as published by the Free Software Foundation. You should have received a copy of the GNU General Public License and a copy of the GCC Runtime Library Exception along with this program; see the files COPYING3 and COPYING.RUNTIME respectively. If not, see . ---- Full text of license texts referred to above follows (that they are listed below does not necessarily imply the conditions apply to the present binary release): ---- GCC RUNTIME LIBRARY EXCEPTION Version 3.1, 31 March 2009 Copyright (C) 2009 Free Software Foundation, Inc. Everyone is permitted to copy and distribute verbatim copies of this license document, but changing it is not allowed. This GCC Runtime Library Exception (\"Exception\") is an additional permission under section 7 of the GNU General Public License, version 3 (\"GPLv3\"). It applies to a given file (the \"Runtime Library\") that bears a notice placed by the copyright holder of the file stating that the file is governed by GPLv3 along with this Exception. When you use GCC to compile a program, GCC may combine portions of certain GCC header files and runtime libraries with the compiled program. The purpose of this Exception is to allow compilation of non-GPL (including proprietary) programs to use, in this way, the header files and runtime libraries covered by this Exception. 0. Definitions. A file is an \"Independent Module\" if it either requires the Runtime Library for execution after a Compilation Process, or makes use of an interface provided by the Runtime Library, but is not otherwise based on the Runtime Library. \"GCC\" means a version of the GNU Compiler Collection, with or without modifications, governed by version 3 (or a specified later version) of the GNU General Public License (GPL) with the option of using any subsequent versions published by the FSF. \"GPL-compatible Software\" is software whose conditions of propagation, modification and use would permit combination with GCC in accord with the license of GCC. \"Target Code\" refers to output from any compiler for a real or virtual target processor architecture, in executable form or suitable for input to an assembler, loader, linker and/or execution phase. Notwithstanding that, Target Code does not include data in any format that is used as a compiler intermediate representation, or used for producing a compiler intermediate representation. The \"Compilation Process\" transforms code entirely represented in non-intermediate languages designed for human-written code, and/or in Java Virtual Machine byte code, into Target Code. Thus, for example, use of source code generators and preprocessors need not be considered part of the Compilation Process, since the Compilation Process can be understood as starting with the output of the generators or preprocessors. A Compilation Process is \"Eligible\" if it is done using GCC, alone or with other GPL-compatible software, or if it is done without using any work based on GCC. For example, using non-GPL-compatible Software to optimize any GCC intermediate representations would not qualify as an Eligible Compilation Process. 1. Grant of Additional Permission. You have permission to propagate a work of Target Code formed by combining the Runtime Library with Independent Modules, even if such propagation would otherwise violate the terms of GPLv3, provided that all Target Code was generated by Eligible Compilation Processes. You may then convey such a combination under terms of your choice, consistent with the licensing of the Independent Modules. 2. No Weakening of GCC Copyleft. The availability of this Exception does not imply any general presumption that third-party software is unaffected by the copyleft requirements of the license of GCC. ---- GNU GENERAL PUBLIC LICENSE Version 3, 29 June 2007 Copyright (C) 2007 Free Software Foundation, Inc. Everyone is permitted to copy and distribute verbatim copies of this license document, but changing it is not allowed. Preamble The GNU General Public License is a free, copyleft license for software and other kinds of works. The licenses for most software and other practical works are designed to take away your freedom to share and change the works. By contrast, the GNU General Public License is intended to guarantee your freedom to share and change all versions of a program--to make sure it remains free software for all its users. We, the Free Software Foundation, use the GNU General Public License for most of our software; it applies also to any other work released this way by its authors. You can apply it to your programs, too. When we speak of free software, we are referring to freedom, not price. Our General Public Licenses are designed to make sure that you have the freedom to distribute copies of free software (and charge for them if you wish), that you receive source code or can get it if you want it, that you can change the software or use pieces of it in new free programs, and that you know you can do these things. To protect your rights, we need to prevent others from denying you these rights or asking you to surrender the rights. Therefore, you have certain responsibilities if you distribute copies of the software, or if you modify it: responsibilities to respect the freedom of others. For example, if you distribute copies of such a program, whether gratis or for a fee, you must pass on to the recipients the same freedoms that you received. You must make sure that they, too, receive or can get the source code. And you must show them these terms so they know their rights. Developers that use the GNU GPL protect your rights with two steps: (1) assert copyright on the software, and (2) offer you this License giving you legal permission to copy, distribute and/or modify it. For the developers' and authors' protection, the GPL clearly explains that there is no warranty for this free software. For both users' and authors' sake, the GPL requires that modified versions be marked as changed, so that their problems will not be attributed erroneously to authors of previous versions. Some devices are designed to deny users access to install or run modified versions of the software inside them, although the manufacturer can do so. This is fundamentally incompatible with the aim of protecting users' freedom to change the software. The systematic pattern of such abuse occurs in the area of products for individuals to use, which is precisely where it is most unacceptable. Therefore, we have designed this version of the GPL to prohibit the practice for those products. If such problems arise substantially in other domains, we stand ready to extend this provision to those domains in future versions of the GPL, as needed to protect the freedom of users. Finally, every program is threatened constantly by software patents. States should not allow patents to restrict development and use of software on general-purpose computers, but in those that do, we wish to avoid the special danger that patents applied to a free program could make it effectively proprietary. To prevent this, the GPL assures that patents cannot be used to render the program non-free. The precise terms and conditions for copying, distribution and modification follow. TERMS AND CONDITIONS 0. Definitions. \"This License\" refers to version 3 of the GNU General Public License. \"Copyright\" also means copyright-like laws that apply to other kinds of works, such as semiconductor masks. \"The Program\" refers to any copyrightable work licensed under this License. Each licensee is addressed as \"you\". \"Licensees\" and \"recipients\" may be individuals or organizations. To \"modify\" a work means to copy from or adapt all or part of the work in a fashion requiring copyright permission, other than the making of an exact copy. The resulting work is called a \"modified version\" of the earlier work or a work \"based on\" the earlier work. A \"covered work\" means either the unmodified Program or a work based on the Program. To \"propagate\" a work means to do anything with it that, without permission, would make you directly or secondarily liable for infringement under applicable copyright law, except executing it on a computer or modifying a private copy. Propagation includes copying, distribution (with or without modification), making available to the public, and in some countries other activities as well. To \"convey\" a work means any kind of propagation that enables other parties to make or receive copies. Mere interaction with a user through a computer network, with no transfer of a copy, is not conveying. An interactive user interface displays \"Appropriate Legal Notices\" to the extent that it includes a convenient and prominently visible feature that (1) displays an appropriate copyright notice, and (2) tells the user that there is no warranty for the work (except to the extent that warranties are provided), that licensees may convey the work under this License, and how to view a copy of this License. If the interface presents a list of user commands or options, such as a menu, a prominent item in the list meets this criterion. 1. Source Code. The \"source code\" for a work means the preferred form of the work for making modifications to it. \"Object code\" means any non-source form of a work. A \"Standard Interface\" means an interface that either is an official standard defined by a recognized standards body, or, in the case of interfaces specified for a particular programming language, one that is widely used among developers working in that language. The \"System Libraries\" of an executable work include anything, other than the work as a whole, that (a) is included in the normal form of packaging a Major Component, but which is not part of that Major Component, and (b) serves only to enable use of the work with that Major Component, or to implement a Standard Interface for which an implementation is available to the public in source code form. A \"Major Component\", in this context, means a major essential component (kernel, window system, and so on) of the specific operating system (if any) on which the executable work runs, or a compiler used to produce the work, or an object code interpreter used to run it. The \"Corresponding Source\" for a work in object code form means all the source code needed to generate, install, and (for an executable work) run the object code and to modify the work, including scripts to control those activities. However, it does not include the work's System Libraries, or general-purpose tools or generally available free programs which are used unmodified in performing those activities but which are not part of the work. For example, Corresponding Source includes interface definition files associated with source files for the work, and the source code for shared libraries and dynamically linked subprograms that the work is specifically designed to require, such as by intimate data communication or control flow between those subprograms and other parts of the work. The Corresponding Source need not include anything that users can regenerate automatically from other parts of the Corresponding Source. The Corresponding Source for a work in source code form is that same work. 2. Basic Permissions. All rights granted under this License are granted for the term of copyright on the Program, and are irrevocable provided the stated conditions are met. This License explicitly affirms your unlimited permission to run the unmodified Program. The output from running a covered work is covered by this License only if the output, given its content, constitutes a covered work. This License acknowledges your rights of fair use or other equivalent, as provided by copyright law. You may make, run and propagate covered works that you do not convey, without conditions so long as your license otherwise remains in force. You may convey covered works to others for the sole purpose of having them make modifications exclusively for you, or provide you with facilities for running those works, provided that you comply with the terms of this License in conveying all material for which you do not control copyright. Those thus making or running the covered works for you must do so exclusively on your behalf, under your direction and control, on terms that prohibit them from making any copies of your copyrighted material outside their relationship with you. Conveying under any other circumstances is permitted solely under the conditions stated below. Sublicensing is not allowed; section 10 makes it unnecessary. 3. Protecting Users' Legal Rights From Anti-Circumvention Law. No covered work shall be deemed part of an effective technological measure under any applicable law fulfilling obligations under article 11 of the WIPO copyright treaty adopted on 20 December 1996, or similar laws prohibiting or restricting circumvention of such measures. When you convey a covered work, you waive any legal power to forbid circumvention of technological measures to the extent such circumvention is effected by exercising rights under this License with respect to the covered work, and you disclaim any intention to limit operation or modification of the work as a means of enforcing, against the work's users, your or third parties' legal rights to forbid circumvention of technological measures. 4. Conveying Verbatim Copies. You may convey verbatim copies of the Program's source code as you receive it, in any medium, provided that you conspicuously and appropriately publish on each copy an appropriate copyright notice; keep intact all notices stating that this License and any non-permissive terms added in accord with section 7 apply to the code; keep intact all notices of the absence of any warranty; and give all recipients a copy of this License along with the Program. You may charge any price or no price for each copy that you convey, and you may offer support or warranty protection for a fee. 5. Conveying Modified Source Versions. You may convey a work based on the Program, or the modifications to produce it from the Program, in the form of source code under the terms of section 4, provided that you also meet all of these conditions: a) The work must carry prominent notices stating that you modified it, and giving a relevant date. b) The work must carry prominent notices stating that it is released under this License and any conditions added under section 7. This requirement modifies the requirement in section 4 to \"keep intact all notices\". c) You must license the entire work, as a whole, under this License to anyone who comes into possession of a copy. This License will therefore apply, along with any applicable section 7 additional terms, to the whole of the work, and all its parts, regardless of how they are packaged. This License gives no permission to license the work in any other way, but it does not invalidate such permission if you have separately received it. d) If the work has interactive user interfaces, each must display Appropriate Legal Notices; however, if the Program has interactive interfaces that do not display Appropriate Legal Notices, your work need not make them do so. A compilation of a covered work with other separate and independent works, which are not by their nature extensions of the covered work, and which are not combined with it such as to form a larger program, in or on a volume of a storage or distribution medium, is called an \"aggregate\" if the compilation and its resulting copyright are not used to limit the access or legal rights of the compilation's users beyond what the individual works permit. Inclusion of a covered work in an aggregate does not cause this License to apply to the other parts of the aggregate. 6. Conveying Non-Source Forms. You may convey a covered work in object code form under the terms of sections 4 and 5, provided that you also convey the machine-readable Corresponding Source under the terms of this License, in one of these ways: a) Convey the object code in, or embodied in, a physical product (including a physical distribution medium), accompanied by the Corresponding Source fixed on a durable physical medium customarily used for software interchange. b) Convey the object code in, or embodied in, a physical product (including a physical distribution medium), accompanied by a written offer, valid for at least three years and valid for as long as you offer spare parts or customer support for that product model, to give anyone who possesses the object code either (1) a copy of the Corresponding Source for all the software in the product that is covered by this License, on a durable physical medium customarily used for software interchange, for a price no more than your reasonable cost of physically performing this conveying of source, or (2) access to copy the Corresponding Source from a network server at no charge. c) Convey individual copies of the object code with a copy of the written offer to provide the Corresponding Source. This alternative is allowed only occasionally and noncommercially, and only if you received the object code with such an offer, in accord with subsection 6b. d) Convey the object code by offering access from a designated place (gratis or for a charge), and offer equivalent access to the Corresponding Source in the same way through the same place at no further charge. You need not require recipients to copy the Corresponding Source along with the object code. If the place to copy the object code is a network server, the Corresponding Source may be on a different server (operated by you or a third party) that supports equivalent copying facilities, provided you maintain clear directions next to the object code saying where to find the Corresponding Source. Regardless of what server hosts the Corresponding Source, you remain obligated to ensure that it is available for as long as needed to satisfy these requirements. e) Convey the object code using peer-to-peer transmission, provided you inform other peers where the object code and Corresponding Source of the work are being offered to the general public at no charge under subsection 6d. A separable portion of the object code, whose source code is excluded from the Corresponding Source as a System Library, need not be included in conveying the object code work. A \"User Product\" is either (1) a \"consumer product\", which means any tangible personal property which is normally used for personal, family, or household purposes, or (2) anything designed or sold for incorporation into a dwelling. In determining whether a product is a consumer product, doubtful cases shall be resolved in favor of coverage. For a particular product received by a particular user, \"normally used\" refers to a typical or common use of that class of product, regardless of the status of the particular user or of the way in which the particular user actually uses, or expects or is expected to use, the product. A product is a consumer product regardless of whether the product has substantial commercial, industrial or non-consumer uses, unless such uses represent the only significant mode of use of the product. \"Installation Information\" for a User Product means any methods, procedures, authorization keys, or other information required to install and execute modified versions of a covered work in that User Product from a modified version of its Corresponding Source. The information must suffice to ensure that the continued functioning of the modified object code is in no case prevented or interfered with solely because modification has been made. If you convey an object code work under this section in, or with, or specifically for use in, a User Product, and the conveying occurs as part of a transaction in which the right of possession and use of the User Product is transferred to the recipient in perpetuity or for a fixed term (regardless of how the transaction is characterized), the Corresponding Source conveyed under this section must be accompanied by the Installation Information. But this requirement does not apply if neither you nor any third party retains the ability to install modified object code on the User Product (for example, the work has been installed in ROM). The requirement to provide Installation Information does not include a requirement to continue to provide support service, warranty, or updates for a work that has been modified or installed by the recipient, or for the User Product in which it has been modified or installed. Access to a network may be denied when the modification itself materially and adversely affects the operation of the network or violates the rules and protocols for communication across the network. Corresponding Source conveyed, and Installation Information provided, in accord with this section must be in a format that is publicly documented (and with an implementation available to the public in source code form), and must require no special password or key for unpacking, reading or copying. 7. Additional Terms. \"Additional permissions\" are terms that supplement the terms of this License by making exceptions from one or more of its conditions. Additional permissions that are applicable to the entire Program shall be treated as though they were included in this License, to the extent that they are valid under applicable law. If additional permissions apply only to part of the Program, that part may be used separately under those permissions, but the entire Program remains governed by this License without regard to the additional permissions. When you convey a copy of a covered work, you may at your option remove any additional permissions from that copy, or from any part of it. (Additional permissions may be written to require their own removal in certain cases when you modify the work.) You may place additional permissions on material, added by you to a covered work, for which you have or can give appropriate copyright permission. Notwithstanding any other provision of this License, for material you add to a covered work, you may (if authorized by the copyright holders of that material) supplement the terms of this License with terms: a) Disclaiming warranty or limiting liability differently from the terms of sections 15 and 16 of this License; or b) Requiring preservation of specified reasonable legal notices or author attributions in that material or in the Appropriate Legal Notices displayed by works containing it; or c) Prohibiting misrepresentation of the origin of that material, or requiring that modified versions of such material be marked in reasonable ways as different from the original version; or d) Limiting the use for publicity purposes of names of licensors or authors of the material; or e) Declining to grant rights under trademark law for use of some trade names, trademarks, or service marks; or f) Requiring indemnification of licensors and authors of that material by anyone who conveys the material (or modified versions of it) with contractual assumptions of liability to the recipient, for any liability that these contractual assumptions directly impose on those licensors and authors. All other non-permissive additional terms are considered \"further restrictions\" within the meaning of section 10. If the Program as you received it, or any part of it, contains a notice stating that it is governed by this License along with a term that is a further restriction, you may remove that term. If a license document contains a further restriction but permits relicensing or conveying under this License, you may add to a covered work material governed by the terms of that license document, provided that the further restriction does not survive such relicensing or conveying. If you add terms to a covered work in accord with this section, you must place, in the relevant source files, a statement of the additional terms that apply to those files, or a notice indicating where to find the applicable terms. Additional terms, permissive or non-permissive, may be stated in the form of a separately written license, or stated as exceptions; the above requirements apply either way. 8. Termination. You may not propagate or modify a covered work except as expressly provided under this License. Any attempt otherwise to propagate or modify it is void, and will automatically terminate your rights under this License (including any patent licenses granted under the third paragraph of section 11). However, if you cease all violation of this License, then your license from a particular copyright holder is reinstated (a) provisionally, unless and until the copyright holder explicitly and finally terminates your license, and (b) permanently, if the copyright holder fails to notify you of the violation by some reasonable means prior to 60 days after the cessation. Moreover, your license from a particular copyright holder is reinstated permanently if the copyright holder notifies you of the violation by some reasonable means, this is the first time you have received notice of violation of this License (for any work) from that copyright holder, and you cure the violation prior to 30 days after your receipt of the notice. Termination of your rights under this section does not terminate the licenses of parties who have received copies or rights from you under this License. If your rights have been terminated and not permanently reinstated, you do not qualify to receive new licenses for the same material under section 10. 9. Acceptance Not Required for Having Copies. You are not required to accept this License in order to receive or run a copy of the Program. Ancillary propagation of a covered work occurring solely as a consequence of using peer-to-peer transmission to receive a copy likewise does not require acceptance. However, nothing other than this License grants you permission to propagate or modify any covered work. These actions infringe copyright if you do not accept this License. Therefore, by modifying or propagating a covered work, you indicate your acceptance of this License to do so. 10. Automatic Licensing of Downstream Recipients. Each time you convey a covered work, the recipient automatically receives a license from the original licensors, to run, modify and propagate that work, subject to this License. You are not responsible for enforcing compliance by third parties with this License. An \"entity transaction\" is a transaction transferring control of an organization, or substantially all assets of one, or subdividing an organization, or merging organizations. If propagation of a covered work results from an entity transaction, each party to that transaction who receives a copy of the work also receives whatever licenses to the work the party's predecessor in interest had or could give under the previous paragraph, plus a right to possession of the Corresponding Source of the work from the predecessor in interest, if the predecessor has it or can get it with reasonable efforts. You may not impose any further restrictions on the exercise of the rights granted or affirmed under this License. For example, you may not impose a license fee, royalty, or other charge for exercise of rights granted under this License, and you may not initiate litigation (including a cross-claim or counterclaim in a lawsuit) alleging that any patent claim is infringed by making, using, selling, offering for sale, or importing the Program or any portion of it. 11. Patents. A \"contributor\" is a copyright holder who authorizes use under this License of the Program or a work on which the Program is based. The work thus licensed is called the contributor's \"contributor version\". A contributor's \"essential patent claims\" are all patent claims owned or controlled by the contributor, whether already acquired or hereafter acquired, that would be infringed by some manner, permitted by this License, of making, using, or selling its contributor version, but do not include claims that would be infringed only as a consequence of further modification of the contributor version. For purposes of this definition, \"control\" includes the right to grant patent sublicenses in a manner consistent with the requirements of this License. Each contributor grants you a non-exclusive, worldwide, royalty-free patent license under the contributor's essential patent claims, to make, use, sell, offer for sale, import and otherwise run, modify and propagate the contents of its contributor version. In the following three paragraphs, a \"patent license\" is any express agreement or commitment, however denominated, not to enforce a patent (such as an express permission to practice a patent or covenant not to sue for patent infringement). To \"grant\" such a patent license to a party means to make such an agreement or commitment not to enforce a patent against the party. If you convey a covered work, knowingly relying on a patent license, and the Corresponding Source of the work is not available for anyone to copy, free of charge and under the terms of this License, through a publicly available network server or other readily accessible means, then you must either (1) cause the Corresponding Source to be so available, or (2) arrange to deprive yourself of the benefit of the patent license for this particular work, or (3) arrange, in a manner consistent with the requirements of this License, to extend the patent license to downstream recipients. \"Knowingly relying\" means you have actual knowledge that, but for the patent license, your conveying the covered work in a country, or your recipient's use of the covered work in a country, would infringe one or more identifiable patents in that country that you have reason to believe are valid. If, pursuant to or in connection with a single transaction or arrangement, you convey, or propagate by procuring conveyance of, a covered work, and grant a patent license to some of the parties receiving the covered work authorizing them to use, propagate, modify or convey a specific copy of the covered work, then the patent license you grant is automatically extended to all recipients of the covered work and works based on it. A patent license is \"discriminatory\" if it does not include within the scope of its coverage, prohibits the exercise of, or is conditioned on the non-exercise of one or more of the rights that are specifically granted under this License. You may not convey a covered work if you are a party to an arrangement with a third party that is in the business of distributing software, under which you make payment to the third party based on the extent of your activity of conveying the work, and under which the third party grants, to any of the parties who would receive the covered work from you, a discriminatory patent license (a) in connection with copies of the covered work conveyed by you (or copies made from those copies), or (b) primarily for and in connection with specific products or compilations that contain the covered work, unless you entered into that arrangement, or that patent license was granted, prior to 28 March 2007. Nothing in this License shall be construed as excluding or limiting any implied license or other defenses to infringement that may otherwise be available to you under applicable patent law. 12. No Surrender of Others' Freedom. If conditions are imposed on you (whether by court order, agreement or otherwise) that contradict the conditions of this License, they do not excuse you from the conditions of this License. If you cannot convey a covered work so as to satisfy simultaneously your obligations under this License and any other pertinent obligations, then as a consequence you may not convey it at all. For example, if you agree to terms that obligate you to collect a royalty for further conveying from those to whom you convey the Program, the only way you could satisfy both those terms and this License would be to refrain entirely from conveying the Program. 13. Use with the GNU Affero General Public License. Notwithstanding any other provision of this License, you have permission to link or combine any covered work with a work licensed under version 3 of the GNU Affero General Public License into a single combined work, and to convey the resulting work. The terms of this License will continue to apply to the part which is the covered work, but the special requirements of the GNU Affero General Public License, section 13, concerning interaction through a network will apply to the combination as such. 14. Revised Versions of this License. The Free Software Foundation may publish revised and/or new versions of the GNU General Public License from time to time. Such new versions will be similar in spirit to the present version, but may differ in detail to address new problems or concerns. Each version is given a distinguishing version number. If the Program specifies that a certain numbered version of the GNU General Public License \"or any later version\" applies to it, you have the option of following the terms and conditions either of that numbered version or of any later version published by the Free Software Foundation. If the Program does not specify a version number of the GNU General Public License, you may choose any version ever published by the Free Software Foundation. If the Program specifies that a proxy can decide which future versions of the GNU General Public License can be used, that proxy's public statement of acceptance of a version permanently authorizes you to choose that version for the Program. Later license versions may give you additional or different permissions. However, no additional obligations are imposed on any author or copyright holder as a result of your choosing to follow a later version. 15. Disclaimer of Warranty. THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM \"AS IS\" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 16. Limitation of Liability. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. 17. Interpretation of Sections 15 and 16. If the disclaimer of warranty and limitation of liability provided above cannot be given local legal effect according to their terms, reviewing courts shall apply local law that most closely approximates an absolute waiver of all civil liability in connection with the Program, unless a warranty or assumption of liability accompanies a copy of the Program in return for a fee. END OF TERMS AND CONDITIONS How to Apply These Terms to Your New Programs If you develop a new program, and you want it to be of the greatest possible use to the public, the best way to achieve this is to make it free software which everyone can redistribute and change under these terms. To do so, attach the following notices to the program. It is safest to attach them to the start of each source file to most effectively state the exclusion of warranty; and each file should have at least the \"copyright\" line and a pointer to where the full notice is found. Copyright (C) This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program. If not, see . Also add information on how to contact you by electronic and paper mail. If the program does terminal interaction, make it output a short notice like this when it starts in an interactive mode: Copyright (C) This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. This is free software, and you are welcome to redistribute it under certain conditions; type `show c' for details. The hypothetical commands `show w' and `show c' should show the appropriate parts of the General Public License. Of course, your program's commands might be different; for a GUI interface, you would use an \"about box\". You should also get your employer (if you work as a programmer) or school, if any, to sign a \"copyright disclaimer\" for the program, if necessary. For more information on this, and how to apply and follow the GNU GPL, see . The GNU General Public License does not permit incorporating your program into proprietary programs. If your program is a subroutine library, you may consider it more useful to permit linking proprietary applications with the library. If this is what you want to do, use the GNU Lesser General Public License instead of this License. But first, please read .", + "vcrpy:8.1.1": "MIT", + "langchain-openai:1.1.14": "MIT", + "grpc-google-iam-v1:0.14.4": "Apache 2.0", + "tornado:6.5.5": "Apache-2.0" } \ No newline at end of file diff --git a/litellm-proxy-extras/litellm_proxy_extras/migrations/20260421135425_add_team_membership_total_spend/migration.sql b/litellm-proxy-extras/litellm_proxy_extras/migrations/20260421135425_add_team_membership_total_spend/migration.sql index 049bd513cd8..7d7c359bf2d 100644 --- a/litellm-proxy-extras/litellm_proxy_extras/migrations/20260421135425_add_team_membership_total_spend/migration.sql +++ b/litellm-proxy-extras/litellm_proxy_extras/migrations/20260421135425_add_team_membership_total_spend/migration.sql @@ -1,3 +1,3 @@ -- AlterTable -ALTER TABLE "LiteLLM_TeamMembership" ADD COLUMN "total_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0; +ALTER TABLE "LiteLLM_TeamMembership" ADD COLUMN IF NOT EXISTS "total_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0; diff --git a/litellm-proxy-extras/litellm_proxy_extras/migrations/20260513120000_add_delegate_auth_to_upstream_to_mcp_servers/migration.sql b/litellm-proxy-extras/litellm_proxy_extras/migrations/20260513120000_add_delegate_auth_to_upstream_to_mcp_servers/migration.sql new file mode 100644 index 00000000000..50a48743901 --- /dev/null +++ b/litellm-proxy-extras/litellm_proxy_extras/migrations/20260513120000_add_delegate_auth_to_upstream_to_mcp_servers/migration.sql @@ -0,0 +1,2 @@ +-- AlterTable +ALTER TABLE "LiteLLM_MCPServerTable" ADD COLUMN IF NOT EXISTS "delegate_auth_to_upstream" BOOLEAN NOT NULL DEFAULT false; diff --git a/litellm-proxy-extras/litellm_proxy_extras/migrations/20260514120000_add_blocked_to_proxy_model_table/migration.sql b/litellm-proxy-extras/litellm_proxy_extras/migrations/20260514120000_add_blocked_to_proxy_model_table/migration.sql new file mode 100644 index 00000000000..3253b63a884 --- /dev/null +++ b/litellm-proxy-extras/litellm_proxy_extras/migrations/20260514120000_add_blocked_to_proxy_model_table/migration.sql @@ -0,0 +1,4 @@ +-- AlterTable +-- Adds the admin-toggleable pause flag used by the router's blocked filter and the +-- credential lookup helpers; defaults to false so existing rows behave unchanged. +ALTER TABLE "LiteLLM_ProxyModelTable" ADD COLUMN IF NOT EXISTS "blocked" BOOLEAN NOT NULL DEFAULT false; diff --git a/litellm-proxy-extras/litellm_proxy_extras/schema.prisma b/litellm-proxy-extras/litellm_proxy_extras/schema.prisma index 84ce99557e3..78143fe0411 100644 --- a/litellm-proxy-extras/litellm_proxy_extras/schema.prisma +++ b/litellm-proxy-extras/litellm_proxy_extras/schema.prisma @@ -48,9 +48,10 @@ model LiteLLM_CredentialsTable { // Models on proxy model LiteLLM_ProxyModelTable { model_id String @id @default(uuid()) - model_name String + model_name String litellm_params Json - model_info Json? + model_info Json? + blocked Boolean @default(false) created_at DateTime @default(now()) @map("created_at") created_by String updated_at DateTime @default(now()) @updatedAt @map("updated_at") @@ -323,6 +324,7 @@ model LiteLLM_MCPServerTable { registration_url String? allow_all_keys Boolean @default(false) available_on_public_internet Boolean @default(true) + delegate_auth_to_upstream Boolean @default(false) is_byok Boolean @default(false) byok_description String[] @default([]) byok_api_key_help_url String? diff --git a/litellm-proxy-extras/pyproject.toml b/litellm-proxy-extras/pyproject.toml index cd569ca08ec..0654f17ec68 100644 --- a/litellm-proxy-extras/pyproject.toml +++ b/litellm-proxy-extras/pyproject.toml @@ -1,6 +1,6 @@ [project] name = "litellm-proxy-extras" -version = "0.4.71" +version = "0.4.73" description = "Additional files for the LiteLLM Proxy. Reduces the size of the main litellm package." readme = "README.md" requires-python = ">=3.9" @@ -26,7 +26,7 @@ required-version = ">=0.10.9" module-root = "" [tool.commitizen] -version = "0.4.71" +version = "0.4.73" version_files = [ "pyproject.toml:^version", "../pyproject.toml:litellm-proxy-extras==", diff --git a/litellm/__init__.py b/litellm/__init__.py index e1b367fb234..56d516536e8 100644 --- a/litellm/__init__.py +++ b/litellm/__init__.py @@ -225,6 +225,15 @@ use_chat_completions_url_for_anthropic_messages: bool = bool( route_all_chat_openai_to_responses: bool = ( os.getenv("LITELLM_ROUTE_ALL_CHAT_OPENAI_TO_RESPONSES", "false").lower() == "true" ) # When True, routes all OpenAI /chat/completions requests through the Responses API bridge +# When True, Gemini/Vertex Live setup is deferred until client `session.update`. +# Default False preserves historical behavior (auto-send setup on connect). +gemini_live_defer_setup: bool = ( + os.getenv("LITELLM_GEMINI_LIVE_DEFER_SETUP", "false").lower() == "true" +) +use_legacy_interactions_schema: bool = ( + os.getenv("LITELLM_USE_LEGACY_INTERACTIONS_SCHEMA", "false").lower() == "true" +) # When True, sends Api-Revision: 2026-05-07 to Google so responses use the legacy `outputs` +# schema instead of the new `steps` schema. Remove this flag after June 8, 2026. retry = True ### AUTH ### api_key: Optional[str] = None @@ -409,6 +418,12 @@ internal_user_budget_duration: Optional[str] = None tag_budget_config: Optional[Dict[str, "BudgetConfig"]] = None max_end_user_budget: Optional[float] = None max_end_user_budget_id: Optional[str] = None +# When True, end-user IDs extracted from requests are validated against +# LiteLLM_EndUserTable / LiteLLM_UserTable. Values that do not resolve to a +# known row are dropped before reaching spend logs. Defaults to False for +# backwards compatibility — arbitrary client-supplied identifiers still +# pass through unchanged. +validate_end_user_id_in_db: bool = False disable_end_user_cost_tracking: Optional[bool] = None disable_end_user_cost_tracking_prometheus_only: Optional[bool] = None enable_end_user_cost_tracking_prometheus_only: Optional[bool] = None @@ -416,6 +431,7 @@ custom_prometheus_metadata_labels: List[str] = [] custom_prometheus_tags: List[str] = [] prometheus_metrics_config: Optional[List] = None prometheus_emit_stream_label: bool = False +prometheus_user_budget_label_include_email_alias: bool = False prometheus_end_user_metrics_max_series_per_metric: Optional[int] = 10000 prometheus_end_user_metrics_ttl_seconds: Optional[float] = 3600.0 prometheus_end_user_metrics_cleanup_interval_seconds: Optional[float] = 60.0 @@ -631,6 +647,7 @@ minimax_models: Set = set() aws_polly_models: Set = set() gigachat_models: Set = set() llamagate_models: Set = set() +reducto_models: Set = set() bedrock_mantle_models: Set = set() @@ -898,6 +915,8 @@ def add_known_models(model_cost_map: Optional[Dict] = None): gigachat_models.add(key) elif value.get("litellm_provider") == "llamagate": llamagate_models.add(key) + elif value.get("litellm_provider") == "reducto": + reducto_models.add(key) elif value.get("litellm_provider") == "bedrock_mantle": bedrock_mantle_models.add(key) @@ -1009,6 +1028,7 @@ model_list = list( | ovhcloud_models | lemonade_models | docker_model_runner_models + | reducto_models | bedrock_mantle_models | set(clarifai_models) ) @@ -1115,6 +1135,7 @@ models_by_provider: dict = { "aws_polly": aws_polly_models, "gigachat": gigachat_models, "llamagate": llamagate_models, + "reducto": reducto_models, "bedrock_mantle": bedrock_mantle_models, } @@ -1287,6 +1308,18 @@ from .responses.main import * # Interactions API is available as litellm.interactions module # Usage: litellm.interactions.create(), litellm.interactions.get(), etc. from . import interactions +from .interactions.agents.main import ( + acreate as acreate_agent, + create as create_agent, + alist as alist_agents, + list as list_agents, + aget as aget_agent, + get as get_agent, + adelete as adelete_agent, + delete as delete_agent, + alist_versions as alist_agent_versions, + list_versions as list_agent_versions, +) from .skills.main import ( create_skill, acreate_skill, @@ -1849,6 +1882,9 @@ if TYPE_CHECKING: from .llms.azure.completion.transformation import ( AzureOpenAITextConfig as AzureOpenAITextConfig, ) + from .llms.azure.audio_transcription.transformation import ( + AzureSpeechAudioTranscriptionConfig as AzureSpeechAudioTranscriptionConfig, + ) from .llms.hosted_vllm.chat.transformation import ( HostedVLLMChatConfig as HostedVLLMChatConfig, ) @@ -1880,6 +1916,12 @@ if TYPE_CHECKING: from .llms.dashscope.chat.transformation import ( DashScopeChatConfig as DashScopeChatConfig, ) + from .llms.dashscope.embed.transformation import ( + DashScopeEmbeddingConfig as DashScopeEmbeddingConfig, + ) + from .llms.dashscope.rerank.transformation import ( + DashScopeRerankConfig as DashScopeRerankConfig, + ) from .llms.moonshot.chat.transformation import ( MoonshotChatConfig as MoonshotChatConfig, ) diff --git a/litellm/_lazy_imports_registry.py b/litellm/_lazy_imports_registry.py index 3531e8d96b9..17eb6609292 100644 --- a/litellm/_lazy_imports_registry.py +++ b/litellm/_lazy_imports_registry.py @@ -273,6 +273,7 @@ LLM_CONFIG_NAMES = ( "AzureOpenAIConfig", "AzureOpenAIGPT5Config", "AzureOpenAITextConfig", + "AzureSpeechAudioTranscriptionConfig", "HostedVLLMChatConfig", "HostedVLLMEmbeddingConfig", # Alias for backwards compatibility @@ -376,7 +377,6 @@ UTILS_MODULE_NAMES = ( "HTTPHandler", "get_num_retries_from_retry_policy", "reset_retry_policy", - "get_secret", "get_coroutine_checker", "get_litellm_logging_class", "get_set_callbacks", @@ -1055,6 +1055,10 @@ _LLM_CONFIGS_IMPORT_MAP = { ".llms.azure.completion.transformation", "AzureOpenAITextConfig", ), + "AzureSpeechAudioTranscriptionConfig": ( + ".llms.azure.audio_transcription.transformation", + "AzureSpeechAudioTranscriptionConfig", + ), "HostedVLLMChatConfig": ( ".llms.hosted_vllm.chat.transformation", "HostedVLLMChatConfig", @@ -1284,7 +1288,6 @@ _UTILS_MODULE_IMPORT_MAP = { "litellm.router_utils.get_retry_from_policy", "reset_retry_policy", ), - "get_secret": ("litellm.secret_managers.main", "get_secret"), "get_coroutine_checker": ( "litellm.litellm_core_utils.cached_imports", "get_coroutine_checker", diff --git a/litellm/_logging.py b/litellm/_logging.py index 5ddafd6c6af..6b99f50e014 100644 --- a/litellm/_logging.py +++ b/litellm/_logging.py @@ -404,6 +404,7 @@ def _turn_on_debug(): def _disable_debugging(): + """Disable the package, router, and proxy verbose loggers.""" verbose_logger.disabled = True verbose_router_logger.disabled = True verbose_proxy_logger.disabled = True diff --git a/litellm/_redis.py b/litellm/_redis.py index 1c11ea829ba..5ab551453bb 100644 --- a/litellm/_redis.py +++ b/litellm/_redis.py @@ -41,7 +41,7 @@ def _get_redis_kwargs(): "retry", } - include_args = [ + include_args = { "url", "redis_connect_func", "gcp_service_account", @@ -50,9 +50,9 @@ def _get_redis_kwargs(): "azure_client_id", "azure_tenant_id", "azure_client_secret", - ] + } - available_args = [x for x in arg_spec.args if x not in exclude_args] + include_args + available_args = {x for x in arg_spec.args if x not in exclude_args} | include_args return available_args @@ -84,23 +84,25 @@ def _get_redis_cluster_kwargs(client=None): # Only allow primitive arguments exclude_args = {"self", "connection_pool", "retry", "host", "port", "startup_nodes"} - available_args = [x for x in arg_spec.args if x not in exclude_args] - available_args.append("password") - available_args.append("username") - available_args.append("ssl") - available_args.append("ssl_cert_reqs") - available_args.append("ssl_check_hostname") - available_args.append("ssl_ca_certs") - available_args.append( - "redis_connect_func" - ) # Needed for sync clusters and IAM detection - available_args.append("gcp_service_account") - available_args.append("gcp_ssl_ca_certs") - available_args.append("azure_redis_ad_token") - available_args.append("azure_client_id") - available_args.append("azure_tenant_id") - available_args.append("azure_client_secret") - available_args.append("max_connections") + available_args = {x for x in arg_spec.args if x not in exclude_args} + available_args |= { + "password", + "username", + "ssl", + "ssl_cert_reqs", + "ssl_check_hostname", + "ssl_ca_certs", + "redis_connect_func", # Needed for sync clusters and IAM detection + "gcp_service_account", + "gcp_ssl_ca_certs", + "azure_redis_ad_token", + "azure_client_id", + "azure_tenant_id", + "azure_client_secret", + "max_connections", + "socket_timeout", + "socket_connect_timeout", + } return available_args @@ -479,10 +481,24 @@ def init_redis_cluster(redis_kwargs) -> redis.RedisCluster: return redis.RedisCluster(startup_nodes=new_startup_nodes, **cluster_kwargs) # type: ignore +def _get_redis_sentinel_connection_kwargs(redis_kwargs: dict) -> dict: + connection_kwargs = {} + args = _get_redis_kwargs() + for arg in redis_kwargs: + if arg in args: + connection_kwargs[arg] = redis_kwargs[arg] + + return connection_kwargs + + def _init_redis_sentinel(redis_kwargs) -> redis.Redis: sentinel_nodes = redis_kwargs.get("sentinel_nodes") sentinel_password = redis_kwargs.get("sentinel_password") service_name = redis_kwargs.get("service_name") + connection_kwargs = _get_redis_sentinel_connection_kwargs(redis_kwargs) + connection_kwargs.setdefault("socket_timeout", REDIS_SOCKET_TIMEOUT) + sentinel_kwargs = dict(connection_kwargs) + sentinel_kwargs["password"] = sentinel_password if not sentinel_nodes or not service_name: raise ValueError( @@ -494,19 +510,22 @@ def _init_redis_sentinel(redis_kwargs) -> redis.Redis: # Set up the Sentinel client sentinel = redis.Sentinel( sentinel_nodes, - socket_timeout=REDIS_SOCKET_TIMEOUT, - password=sentinel_password, + sentinel_kwargs=sentinel_kwargs, ) # Return the master instance for the given service - return sentinel.master_for(service_name) + return sentinel.master_for(service_name, **connection_kwargs) def _init_async_redis_sentinel(redis_kwargs) -> async_redis.Redis: sentinel_nodes = redis_kwargs.get("sentinel_nodes") sentinel_password = redis_kwargs.get("sentinel_password") service_name = redis_kwargs.get("service_name") + connection_kwargs = _get_redis_sentinel_connection_kwargs(redis_kwargs) + connection_kwargs.setdefault("socket_timeout", REDIS_SOCKET_TIMEOUT) + sentinel_kwargs = dict(connection_kwargs) + sentinel_kwargs["password"] = sentinel_password if not sentinel_nodes or not service_name: raise ValueError( @@ -518,13 +537,12 @@ def _init_async_redis_sentinel(redis_kwargs) -> async_redis.Redis: # Set up the Sentinel client sentinel = async_redis.Sentinel( sentinel_nodes, - socket_timeout=REDIS_SOCKET_TIMEOUT, - password=sentinel_password, + sentinel_kwargs=sentinel_kwargs, ) # Return the master instance for the given service - return sentinel.master_for(service_name) + return sentinel.master_for(service_name, **connection_kwargs) def get_redis_client(**env_overrides): diff --git a/litellm/_service_logger.py b/litellm/_service_logger.py index 1a3be203fec..5531c418799 100644 --- a/litellm/_service_logger.py +++ b/litellm/_service_logger.py @@ -24,6 +24,22 @@ else: UserAPIKeyAuth = Any +def _get_otel_v2_class() -> Optional[type]: + """Return the ``OpenTelemetryV2`` class, or ``None`` if the OTel SDK is absent. + + Imported lazily: ``litellm.integrations.otel.logger`` imports the OpenTelemetry + SDK at module scope, so importing it eagerly would break installs without the + SDK. The V2 logger only exists when ``LITELLM_OTEL_V2`` is enabled (which + requires the SDK), so a failed import simply means "no V2 logger in play". + """ + try: + from litellm.integrations.otel.logger import OpenTelemetryV2 + + return OpenTelemetryV2 + except Exception: + return None + + class ServiceLogging(CustomLogger): """ Separate class used for monitoring health of litellm-adjacent services (redis/postgres). @@ -38,6 +54,37 @@ class ServiceLogging(CustomLogger): if "prometheus_system" in litellm.service_callback: self.prometheusServicesLogger = PrometheusServicesLogger() + def _resolve_otel_service_logger(self, callback: Any) -> Optional[Any]: + """Resolve the OTel logger (legacy or V2) to emit a service span on. + + Returns the logger instance whose ``async_service_*_hook`` should fire for + this ``callback``, or ``None`` when ``callback`` is not an OTel callback. + + The V2 ``OpenTelemetryV2`` logger is a plain ``CustomLogger`` and is NOT a + subclass of the legacy ``OpenTelemetry``, so the legacy ``isinstance`` + check alone misses it — which is why redis/postgres service spans never + showed up under ``LITELLM_OTEL_V2``. Match both the legacy and V2 types, + whether the callback is the logger instance itself or the ``"otel"`` string + (which routes to the proxy's registered ``open_telemetry_logger``). + """ + otel_v2_cls = _get_otel_v2_class() + + def _is_otel_logger(obj: Any) -> bool: + if isinstance(obj, OpenTelemetry): + return True + return otel_v2_cls is not None and isinstance(obj, otel_v2_cls) + + if _is_otel_logger(callback): + return callback + if callback == "otel": + from litellm.proxy.proxy_server import open_telemetry_logger + + if open_telemetry_logger is not None and _is_otel_logger( + open_telemetry_logger + ): + return open_telemetry_logger + return None + def service_success_hook( self, service: ServiceTypes, @@ -129,6 +176,13 @@ class ServiceLogging(CustomLogger): event_metadata=event_metadata, ) + # OTel loggers already fired this event. ``service_callback`` can hold more + # than one reference that resolves to the *same* logger — the ``"otel"`` + # string AND the registered instance both map to ``open_telemetry_logger`` + # (the V2 logger self-registers its instance even when the string is + # present, unlike V1). Without this guard each such reference emits its own + # span, so a single DB call shows up as duplicate ``postgres ...`` spans. + emitted_otel_logger_ids: set = set() for callback in litellm.service_callback: if callback == "prometheus_system": await self.init_prometheus_services_logger_if_none() @@ -144,19 +198,18 @@ class ServiceLogging(CustomLogger): end_time=end_time, event_metadata=event_metadata, ) - elif callback == "otel" or isinstance(callback, OpenTelemetry): - _otel_logger_to_use: Optional[OpenTelemetry] = None - if isinstance(callback, OpenTelemetry): - _otel_logger_to_use = callback - else: - from litellm.proxy.proxy_server import open_telemetry_logger - - if open_telemetry_logger is not None and isinstance( - open_telemetry_logger, OpenTelemetry - ): - _otel_logger_to_use = open_telemetry_logger - - if _otel_logger_to_use is not None and parent_otel_span is not None: + else: + _otel_logger_to_use = self._resolve_otel_service_logger(callback) + # No ``parent_otel_span is not None`` gate: a background service + # call (no request on the stack) has no parent, and dropping it + # here is what hid those calls from traces entirely. The OTel + # logger decides what to do with a missing parent — legacy V1 + # no-ops, V2 emits a root span (and skips metrics-only pings). + if ( + _otel_logger_to_use is not None + and id(_otel_logger_to_use) not in emitted_otel_logger_ids + ): + emitted_otel_logger_ids.add(id(_otel_logger_to_use)) await _otel_logger_to_use.async_service_success_hook( payload=payload, parent_otel_span=parent_otel_span, @@ -238,6 +291,9 @@ class ServiceLogging(CustomLogger): event_metadata=event_metadata, ) + # Dedupe OTel loggers per event — see ``async_service_success_hook`` for why + # the same logger can be referenced twice in ``service_callback``. + emitted_otel_logger_ids: set = set() for callback in litellm.service_callback: if callback == "prometheus_system": await self.init_prometheus_services_logger_if_none() @@ -255,22 +311,19 @@ class ServiceLogging(CustomLogger): end_time=end_time, event_metadata=event_metadata, ) - elif callback == "otel" or isinstance(callback, OpenTelemetry): - _otel_logger_to_use: Optional[OpenTelemetry] = None - if isinstance(callback, OpenTelemetry): - _otel_logger_to_use = callback - else: - from litellm.proxy.proxy_server import open_telemetry_logger - - if open_telemetry_logger is not None and isinstance( - open_telemetry_logger, OpenTelemetry - ): - _otel_logger_to_use = open_telemetry_logger + else: + _otel_logger_to_use = self._resolve_otel_service_logger(callback) if not isinstance(error, str): error = str(error) - if _otel_logger_to_use is not None and parent_otel_span is not None: + # See the success hook: no parent gate, so background failures + # are traced too. V1 no-ops without a parent; V2 emits a root. + if ( + _otel_logger_to_use is not None + and id(_otel_logger_to_use) not in emitted_otel_logger_ids + ): + emitted_otel_logger_ids.add(id(_otel_logger_to_use)) await _otel_logger_to_use.async_service_failure_hook( payload=payload, error=error, diff --git a/litellm/_uuid.py b/litellm/_uuid.py index 52acf647dd8..2b7c3b82d35 100644 --- a/litellm/_uuid.py +++ b/litellm/_uuid.py @@ -6,7 +6,6 @@ Always uses fastuuid for performance. import fastuuid as _uuid # type: ignore - # Expose a module-like alias so callers can use: uuid.uuid4() uuid = _uuid diff --git a/litellm/a2a_protocol/litellm_completion_bridge/handler.py b/litellm/a2a_protocol/litellm_completion_bridge/handler.py index 4e66fe4ba67..67ffcf4f8f7 100644 --- a/litellm/a2a_protocol/litellm_completion_bridge/handler.py +++ b/litellm/a2a_protocol/litellm_completion_bridge/handler.py @@ -107,6 +107,14 @@ class A2ACompletionBridgeHandler: if k not in ("model", "custom_llm_provider") and k not in _AGENT_ONLY_PARAMS } completion_params.update(litellm_params_to_add) + # Apply forward metadata AFTER the litellm_params merge so the helper + # sees any agent-owner-configured ``extra_body.metadata`` and can keep + # those keys authoritative over the client-supplied A2A metadata. + A2ACompletionBridgeTransformation.apply_forward_metadata_to_completion_params( + completion_params=completion_params, + a2a_message=message, + params=params, + ) # Call litellm.acompletion response = await litellm.acompletion(**completion_params) @@ -214,6 +222,14 @@ class A2ACompletionBridgeHandler: if k not in ("model", "custom_llm_provider") and k not in _AGENT_ONLY_PARAMS } completion_params.update(litellm_params_to_add) + # Apply forward metadata AFTER the litellm_params merge so the helper + # sees any agent-owner-configured ``extra_body.metadata`` and can keep + # those keys authoritative over the client-supplied A2A metadata. + A2ACompletionBridgeTransformation.apply_forward_metadata_to_completion_params( + completion_params=completion_params, + a2a_message=message, + params=params, + ) # 1. Emit initial task event (kind: "task", status: "submitted") task_event = A2ACompletionBridgeTransformation.create_task_event(ctx) diff --git a/litellm/a2a_protocol/litellm_completion_bridge/transformation.py b/litellm/a2a_protocol/litellm_completion_bridge/transformation.py index 8a03569f689..06c0a8fc82f 100644 --- a/litellm/a2a_protocol/litellm_completion_bridge/transformation.py +++ b/litellm/a2a_protocol/litellm_completion_bridge/transformation.py @@ -45,10 +45,80 @@ class A2ACompletionBridgeTransformation: Static methods for transforming between A2A and OpenAI message formats. """ + @staticmethod + def _extract_text_from_a2a_parts(parts: List[Dict[str, Any]]) -> str: + """Extract text from A2A parts (with or without explicit ``kind``).""" + content_parts: List[str] = [] + for part in parts: + if not isinstance(part, dict): + continue + kind = part.get("kind") + text = part.get("text") + if text is None: + continue + if kind in (None, "", "text"): + content_parts.append(str(text)) + return "\n".join(content_parts) + + @staticmethod + def get_forward_metadata( + a2a_message: Dict[str, Any], + params: Optional[Dict[str, Any]] = None, + ) -> Optional[Dict[str, Any]]: + """ + Merge A2A metadata from MessageSendParams and the message for downstream providers. + + Forwarded once on the LangGraph run payload (``metadata``), not duplicated on + each input message — see ``apply_forward_metadata_to_completion_params``. + """ + merged: Dict[str, Any] = {} + if params and isinstance(params.get("metadata"), dict): + merged.update(params["metadata"]) + message_metadata = a2a_message.get("metadata") + if isinstance(message_metadata, dict): + merged.update(message_metadata) + return merged or None + + @staticmethod + def apply_forward_metadata_to_completion_params( + completion_params: Dict[str, Any], + a2a_message: Dict[str, Any], + params: Optional[Dict[str, Any]] = None, + ) -> None: + """ + Attach A2A metadata to completion kwargs for provider bridges (e.g. LangGraph). + + Uses ``extra_body`` so we do not collide with LiteLLM's spend-log ``metadata`` kwarg. + """ + forward_metadata = A2ACompletionBridgeTransformation.get_forward_metadata( + a2a_message=a2a_message, + params=params, + ) + if not forward_metadata: + return + + extra_body = completion_params.get("extra_body") + if not isinstance(extra_body, dict): + extra_body = {} + # Layer client-supplied A2A metadata under any agent-owner-configured + # ``extra_body.metadata`` so the configured keys remain authoritative + # and an A2A caller cannot overwrite server-set run metadata. + existing_metadata = extra_body.get("metadata") + existing_dict: Dict[str, Any] = ( + existing_metadata if isinstance(existing_metadata, dict) else {} + ) + merged_metadata: Dict[str, Any] = {**forward_metadata, **existing_dict} + extra_body = {**extra_body, "metadata": merged_metadata} + completion_params["extra_body"] = extra_body + + verbose_logger.debug( + f"A2A -> completion forward metadata keys={list(forward_metadata.keys())}" + ) + @staticmethod def a2a_message_to_openai_messages( a2a_message: Dict[str, Any], - ) -> List[Dict[str, str]]: + ) -> List[Dict[str, Any]]: """ Transform an A2A message to OpenAI message format. @@ -70,21 +140,20 @@ class A2ACompletionBridgeTransformation: elif role == "system": openai_role = "system" - # Extract text content from parts - content_parts = [] - for part in parts: - kind = part.get("kind", "") - if kind == "text": - text = part.get("text", "") - content_parts.append(text) + if not isinstance(parts, list): + parts = [] - content = "\n".join(content_parts) if content_parts else "" + content = A2ACompletionBridgeTransformation._extract_text_from_a2a_parts(parts) + + # Do not attach A2A message.metadata here — the completion bridge forwards it + # once at run level via extra_body.metadata (LangGraph POST /runs/wait shape). + openai_message: Dict[str, Any] = {"role": openai_role, "content": content} verbose_logger.debug( f"A2A -> OpenAI transform: role={role} -> {openai_role}, content_length={len(content)}" ) - return [{"role": openai_role, "content": content}] + return [openai_message] @staticmethod def openai_response_to_a2a_response( @@ -110,6 +179,7 @@ class A2ACompletionBridgeTransformation: # Build A2A message a2a_message = { + "kind": "message", "role": "agent", "parts": [{"kind": "text", "text": content}], "messageId": uuid4().hex, @@ -119,9 +189,7 @@ class A2ACompletionBridgeTransformation: a2a_response = { "jsonrpc": "2.0", "id": request_id, - "result": { - "message": a2a_message, - }, + "result": a2a_message, } verbose_logger.debug(f"OpenAI -> A2A transform: content_length={len(content)}") @@ -235,50 +303,3 @@ class A2ACompletionBridgeTransformation: "taskId": ctx.task_id, }, } - - @staticmethod - def openai_chunk_to_a2a_chunk( - chunk: Any, - request_id: Optional[str] = None, - is_final: bool = False, - ) -> Optional[Dict[str, Any]]: - """ - Transform a LiteLLM streaming chunk to A2A streaming format. - - NOTE: This method is deprecated for streaming. Use the event-based - methods (create_task_event, create_status_update_event, - create_artifact_update_event) instead for proper A2A streaming. - - Args: - chunk: LiteLLM ModelResponse chunk - request_id: Original A2A request ID - is_final: Whether this is the final chunk - - Returns: - A2A streaming chunk dict or None if no content - """ - # Extract delta content - content = "" - if chunk is not None and hasattr(chunk, "choices") and chunk.choices: - choice = chunk.choices[0] - if hasattr(choice, "delta") and choice.delta: - content = choice.delta.content or "" - - if not content and not is_final: - return None - - # Build A2A streaming chunk (legacy format) - a2a_chunk = { - "jsonrpc": "2.0", - "id": request_id, - "result": { - "message": { - "role": "agent", - "parts": [{"kind": "text", "text": content}], - "messageId": uuid4().hex, - }, - "final": is_final, - }, - } - - return a2a_chunk diff --git a/litellm/a2a_protocol/providers/litellm_completion/README.md b/litellm/a2a_protocol/providers/litellm_completion/README.md deleted file mode 100644 index a809e9bf55e..00000000000 --- a/litellm/a2a_protocol/providers/litellm_completion/README.md +++ /dev/null @@ -1,74 +0,0 @@ -# A2A to LiteLLM Completion Bridge - -Routes A2A protocol requests through `litellm.acompletion`, enabling any LiteLLM-supported provider to be invoked via A2A. - -## Flow - -``` -A2A Request → Transform → litellm.acompletion → Transform → A2A Response -``` - -## SDK Usage - -Use the existing `asend_message` and `asend_message_streaming` functions with `litellm_params`: - -```python -from litellm.a2a_protocol import asend_message, asend_message_streaming -from a2a.types import SendMessageRequest, SendStreamingMessageRequest, MessageSendParams -from uuid import uuid4 - -# Non-streaming -request = SendMessageRequest( - id=str(uuid4()), - params=MessageSendParams( - message={"role": "user", "parts": [{"kind": "text", "text": "Hello!"}], "messageId": uuid4().hex} - ) -) -response = await asend_message( - request=request, - api_base="http://localhost:2024", - litellm_params={"custom_llm_provider": "langgraph", "model": "agent"}, -) - -# Streaming -stream_request = SendStreamingMessageRequest( - id=str(uuid4()), - params=MessageSendParams( - message={"role": "user", "parts": [{"kind": "text", "text": "Hello!"}], "messageId": uuid4().hex} - ) -) -async for chunk in asend_message_streaming( - request=stream_request, - api_base="http://localhost:2024", - litellm_params={"custom_llm_provider": "langgraph", "model": "agent"}, -): - print(chunk) -``` - -## Proxy Usage - -Configure an agent with `custom_llm_provider` in `litellm_params`: - -```yaml -agents: - - agent_name: my-langgraph-agent - agent_card_params: - name: "LangGraph Agent" - url: "http://localhost:2024" # Used as api_base - litellm_params: - custom_llm_provider: langgraph - model: agent -``` - -When an A2A request hits `/a2a/{agent_id}/message/send`, the bridge: - -1. Detects `custom_llm_provider` in agent's `litellm_params` -2. Transforms A2A message → OpenAI messages -3. Calls `litellm.acompletion(model="langgraph/agent", api_base="http://localhost:2024")` -4. Transforms response → A2A format - -## Classes - -- `A2ACompletionBridgeTransformation` - Static methods for message format conversion -- `A2ACompletionBridgeHandler` - Static methods for handling requests (streaming/non-streaming) - diff --git a/litellm/a2a_protocol/providers/litellm_completion/__init__.py b/litellm/a2a_protocol/providers/litellm_completion/__init__.py deleted file mode 100644 index fc2fc17f54f..00000000000 --- a/litellm/a2a_protocol/providers/litellm_completion/__init__.py +++ /dev/null @@ -1,5 +0,0 @@ -""" -LiteLLM Completion bridge provider for A2A protocol. - -Routes A2A requests through litellm.acompletion based on custom_llm_provider. -""" diff --git a/litellm/a2a_protocol/providers/litellm_completion/handler.py b/litellm/a2a_protocol/providers/litellm_completion/handler.py deleted file mode 100644 index 730f8f6b36f..00000000000 --- a/litellm/a2a_protocol/providers/litellm_completion/handler.py +++ /dev/null @@ -1,301 +0,0 @@ -""" -Handler for A2A to LiteLLM completion bridge. - -Routes A2A requests through litellm.acompletion based on custom_llm_provider. - -A2A Streaming Events (in order): -1. Task event (kind: "task") - Initial task creation with status "submitted" -2. Status update (kind: "status-update") - Status change to "working" -3. Artifact update (kind: "artifact-update") - Content/artifact delivery -4. Status update (kind: "status-update") - Final status "completed" with final=true -""" - -from typing import Any, AsyncIterator, Dict, Optional - -import litellm -from litellm._logging import verbose_logger -from litellm.a2a_protocol.litellm_completion_bridge.pydantic_ai_transformation import ( - PydanticAITransformation, -) -from litellm.a2a_protocol.litellm_completion_bridge.transformation import ( - A2ACompletionBridgeTransformation, - A2AStreamingContext, -) - - -class A2ACompletionBridgeHandler: - """ - Static methods for handling A2A requests via LiteLLM completion. - """ - - @staticmethod - async def handle_non_streaming( - request_id: str, - params: Dict[str, Any], - litellm_params: Dict[str, Any], - api_base: Optional[str] = None, - ) -> Dict[str, Any]: - """ - Handle non-streaming A2A request via litellm.acompletion. - - Args: - request_id: A2A JSON-RPC request ID - params: A2A MessageSendParams containing the message - litellm_params: Agent's litellm_params (custom_llm_provider, model, etc.) - api_base: API base URL from agent_card_params - - Returns: - A2A SendMessageResponse dict - """ - # Check if this is a Pydantic AI agent request - custom_llm_provider = litellm_params.get("custom_llm_provider") - if custom_llm_provider == "pydantic_ai_agents": - if api_base is None: - raise ValueError("api_base is required for Pydantic AI agents") - - verbose_logger.info( - f"Pydantic AI: Routing to Pydantic AI agent at {api_base}" - ) - - # Send request directly to Pydantic AI agent - response_data = await PydanticAITransformation.send_non_streaming_request( - api_base=api_base, - request_id=request_id, - params=params, - ) - - return response_data - - # Extract message from params - message = params.get("message", {}) - - # Transform A2A message to OpenAI format - openai_messages = ( - A2ACompletionBridgeTransformation.a2a_message_to_openai_messages(message) - ) - - # Get completion params - custom_llm_provider = litellm_params.get("custom_llm_provider") - model = litellm_params.get("model", "agent") - - # Build full model string if provider specified - # Skip prepending if model already starts with the provider prefix - if custom_llm_provider and not model.startswith(f"{custom_llm_provider}/"): - full_model = f"{custom_llm_provider}/{model}" - else: - full_model = model - - verbose_logger.info( - f"A2A completion bridge: model={full_model}, api_base={api_base}" - ) - - # Build completion params dict - completion_params = { - "model": full_model, - "messages": openai_messages, - "api_base": api_base, - "stream": False, - } - # Add litellm_params (contains api_key, client_id, client_secret, tenant_id, etc.) - litellm_params_to_add = { - k: v - for k, v in litellm_params.items() - if k not in ("model", "custom_llm_provider") - } - completion_params.update(litellm_params_to_add) - - # Call litellm.acompletion - response = await litellm.acompletion(**completion_params) - - # Transform response to A2A format - a2a_response = ( - A2ACompletionBridgeTransformation.openai_response_to_a2a_response( - response=response, - request_id=request_id, - ) - ) - - verbose_logger.info(f"A2A completion bridge completed: request_id={request_id}") - - return a2a_response - - @staticmethod - async def handle_streaming( - request_id: str, - params: Dict[str, Any], - litellm_params: Dict[str, Any], - api_base: Optional[str] = None, - ) -> AsyncIterator[Dict[str, Any]]: - """ - Handle streaming A2A request via litellm.acompletion with stream=True. - - Emits proper A2A streaming events: - 1. Task event (kind: "task") - Initial task with status "submitted" - 2. Status update (kind: "status-update") - Status "working" - 3. Artifact update (kind: "artifact-update") - Content delivery - 4. Status update (kind: "status-update") - Final "completed" status - - Args: - request_id: A2A JSON-RPC request ID - params: A2A MessageSendParams containing the message - litellm_params: Agent's litellm_params (custom_llm_provider, model, etc.) - api_base: API base URL from agent_card_params - - Yields: - A2A streaming response events - """ - # Check if this is a Pydantic AI agent request - custom_llm_provider = litellm_params.get("custom_llm_provider") - if custom_llm_provider == "pydantic_ai_agents": - if api_base is None: - raise ValueError("api_base is required for Pydantic AI agents") - - verbose_logger.info( - f"Pydantic AI: Faking streaming for Pydantic AI agent at {api_base}" - ) - - # Get non-streaming response first - response_data = await PydanticAITransformation.send_non_streaming_request( - api_base=api_base, - request_id=request_id, - params=params, - ) - - # Convert to fake streaming - async for chunk in PydanticAITransformation.fake_streaming_from_response( - response_data=response_data, - request_id=request_id, - ): - yield chunk - - return - - # Extract message from params - message = params.get("message", {}) - - # Create streaming context - ctx = A2AStreamingContext( - request_id=request_id, - input_message=message, - ) - - # Transform A2A message to OpenAI format - openai_messages = ( - A2ACompletionBridgeTransformation.a2a_message_to_openai_messages(message) - ) - - # Get completion params - custom_llm_provider = litellm_params.get("custom_llm_provider") - model = litellm_params.get("model", "agent") - - # Build full model string if provider specified - # Skip prepending if model already starts with the provider prefix - if custom_llm_provider and not model.startswith(f"{custom_llm_provider}/"): - full_model = f"{custom_llm_provider}/{model}" - else: - full_model = model - - verbose_logger.info( - f"A2A completion bridge streaming: model={full_model}, api_base={api_base}" - ) - - # Build completion params dict - completion_params = { - "model": full_model, - "messages": openai_messages, - "api_base": api_base, - "stream": True, - } - # Add litellm_params (contains api_key, client_id, client_secret, tenant_id, etc.) - litellm_params_to_add = { - k: v - for k, v in litellm_params.items() - if k not in ("model", "custom_llm_provider") - } - completion_params.update(litellm_params_to_add) - - # 1. Emit initial task event (kind: "task", status: "submitted") - task_event = A2ACompletionBridgeTransformation.create_task_event(ctx) - yield task_event - - # 2. Emit status update (kind: "status-update", status: "working") - working_event = A2ACompletionBridgeTransformation.create_status_update_event( - ctx=ctx, - state="working", - final=False, - message_text="Processing request...", - ) - yield working_event - - # Call litellm.acompletion with streaming - response = await litellm.acompletion(**completion_params) - - # 3. Accumulate content and emit artifact update - accumulated_text = "" - chunk_count = 0 - async for chunk in response: # type: ignore[union-attr] - chunk_count += 1 - - # Extract delta content - content = "" - if chunk is not None and hasattr(chunk, "choices") and chunk.choices: - choice = chunk.choices[0] - if hasattr(choice, "delta") and choice.delta: - content = choice.delta.content or "" - - if content: - accumulated_text += content - - # Emit artifact update with accumulated content - if accumulated_text: - artifact_event = ( - A2ACompletionBridgeTransformation.create_artifact_update_event( - ctx=ctx, - text=accumulated_text, - ) - ) - yield artifact_event - - # 4. Emit final status update (kind: "status-update", status: "completed", final: true) - completed_event = A2ACompletionBridgeTransformation.create_status_update_event( - ctx=ctx, - state="completed", - final=True, - ) - yield completed_event - - verbose_logger.info( - f"A2A completion bridge streaming completed: request_id={request_id}, chunks={chunk_count}" - ) - - -# Convenience functions that delegate to the class methods -async def handle_a2a_completion( - request_id: str, - params: Dict[str, Any], - litellm_params: Dict[str, Any], - api_base: Optional[str] = None, -) -> Dict[str, Any]: - """Convenience function for non-streaming A2A completion.""" - return await A2ACompletionBridgeHandler.handle_non_streaming( - request_id=request_id, - params=params, - litellm_params=litellm_params, - api_base=api_base, - ) - - -async def handle_a2a_completion_streaming( - request_id: str, - params: Dict[str, Any], - litellm_params: Dict[str, Any], - api_base: Optional[str] = None, -) -> AsyncIterator[Dict[str, Any]]: - """Convenience function for streaming A2A completion.""" - async for chunk in A2ACompletionBridgeHandler.handle_streaming( - request_id=request_id, - params=params, - litellm_params=litellm_params, - api_base=api_base, - ): - yield chunk diff --git a/litellm/a2a_protocol/providers/litellm_completion/transformation.py b/litellm/a2a_protocol/providers/litellm_completion/transformation.py deleted file mode 100644 index 8a03569f689..00000000000 --- a/litellm/a2a_protocol/providers/litellm_completion/transformation.py +++ /dev/null @@ -1,284 +0,0 @@ -""" -Transformation utilities for A2A <-> OpenAI message format conversion. - -A2A Message Format: -{ - "role": "user", - "parts": [{"kind": "text", "text": "Hello!"}], - "messageId": "abc123" -} - -OpenAI Message Format: -{"role": "user", "content": "Hello!"} - -A2A Streaming Events: -- Task event (kind: "task") - Initial task creation with status "submitted" -- Status update (kind: "status-update") - Status changes (working, completed) -- Artifact update (kind: "artifact-update") - Content/artifact delivery -""" - -from datetime import datetime, timezone -from typing import Any, Dict, List, Optional -from uuid import uuid4 - -from litellm._logging import verbose_logger - - -class A2AStreamingContext: - """ - Context holder for A2A streaming state. - Tracks task_id, context_id, and message accumulation. - """ - - def __init__(self, request_id: str, input_message: Dict[str, Any]): - self.request_id = request_id - self.task_id = str(uuid4()) - self.context_id = str(uuid4()) - self.input_message = input_message - self.accumulated_text = "" - self.has_emitted_task = False - self.has_emitted_working = False - - -class A2ACompletionBridgeTransformation: - """ - Static methods for transforming between A2A and OpenAI message formats. - """ - - @staticmethod - def a2a_message_to_openai_messages( - a2a_message: Dict[str, Any], - ) -> List[Dict[str, str]]: - """ - Transform an A2A message to OpenAI message format. - - Args: - a2a_message: A2A message with role, parts, and messageId - - Returns: - List of OpenAI-format messages - """ - role = a2a_message.get("role", "user") - parts = a2a_message.get("parts", []) - - # Map A2A roles to OpenAI roles - openai_role = role - if role == "user": - openai_role = "user" - elif role == "assistant": - openai_role = "assistant" - elif role == "system": - openai_role = "system" - - # Extract text content from parts - content_parts = [] - for part in parts: - kind = part.get("kind", "") - if kind == "text": - text = part.get("text", "") - content_parts.append(text) - - content = "\n".join(content_parts) if content_parts else "" - - verbose_logger.debug( - f"A2A -> OpenAI transform: role={role} -> {openai_role}, content_length={len(content)}" - ) - - return [{"role": openai_role, "content": content}] - - @staticmethod - def openai_response_to_a2a_response( - response: Any, - request_id: Optional[str] = None, - ) -> Dict[str, Any]: - """ - Transform a LiteLLM ModelResponse to A2A SendMessageResponse format. - - Args: - response: LiteLLM ModelResponse object - request_id: Original A2A request ID - - Returns: - A2A SendMessageResponse dict - """ - # Extract content from response - content = "" - if hasattr(response, "choices") and response.choices: - choice = response.choices[0] - if hasattr(choice, "message") and choice.message: - content = choice.message.content or "" - - # Build A2A message - a2a_message = { - "role": "agent", - "parts": [{"kind": "text", "text": content}], - "messageId": uuid4().hex, - } - - # Build A2A response - a2a_response = { - "jsonrpc": "2.0", - "id": request_id, - "result": { - "message": a2a_message, - }, - } - - verbose_logger.debug(f"OpenAI -> A2A transform: content_length={len(content)}") - - return a2a_response - - @staticmethod - def _get_timestamp() -> str: - """Get current timestamp in ISO format with timezone.""" - return datetime.now(timezone.utc).isoformat() - - @staticmethod - def create_task_event( - ctx: A2AStreamingContext, - ) -> Dict[str, Any]: - """ - Create the initial task event with status 'submitted'. - - This is the first event emitted in an A2A streaming response. - """ - return { - "id": ctx.request_id, - "jsonrpc": "2.0", - "result": { - "contextId": ctx.context_id, - "history": [ - { - "contextId": ctx.context_id, - "kind": "message", - "messageId": ctx.input_message.get("messageId", uuid4().hex), - "parts": ctx.input_message.get("parts", []), - "role": ctx.input_message.get("role", "user"), - "taskId": ctx.task_id, - } - ], - "id": ctx.task_id, - "kind": "task", - "status": { - "state": "submitted", - }, - }, - } - - @staticmethod - def create_status_update_event( - ctx: A2AStreamingContext, - state: str, - final: bool = False, - message_text: Optional[str] = None, - ) -> Dict[str, Any]: - """ - Create a status update event. - - Args: - ctx: Streaming context - state: Status state ('working', 'completed') - final: Whether this is the final event - message_text: Optional message text for 'working' status - """ - status: Dict[str, Any] = { - "state": state, - "timestamp": A2ACompletionBridgeTransformation._get_timestamp(), - } - - # Add message for 'working' status - if state == "working" and message_text: - status["message"] = { - "contextId": ctx.context_id, - "kind": "message", - "messageId": str(uuid4()), - "parts": [{"kind": "text", "text": message_text}], - "role": "agent", - "taskId": ctx.task_id, - } - - return { - "id": ctx.request_id, - "jsonrpc": "2.0", - "result": { - "contextId": ctx.context_id, - "final": final, - "kind": "status-update", - "status": status, - "taskId": ctx.task_id, - }, - } - - @staticmethod - def create_artifact_update_event( - ctx: A2AStreamingContext, - text: str, - ) -> Dict[str, Any]: - """ - Create an artifact update event with content. - - Args: - ctx: Streaming context - text: The text content for the artifact - """ - return { - "id": ctx.request_id, - "jsonrpc": "2.0", - "result": { - "artifact": { - "artifactId": str(uuid4()), - "name": "response", - "parts": [{"kind": "text", "text": text}], - }, - "contextId": ctx.context_id, - "kind": "artifact-update", - "taskId": ctx.task_id, - }, - } - - @staticmethod - def openai_chunk_to_a2a_chunk( - chunk: Any, - request_id: Optional[str] = None, - is_final: bool = False, - ) -> Optional[Dict[str, Any]]: - """ - Transform a LiteLLM streaming chunk to A2A streaming format. - - NOTE: This method is deprecated for streaming. Use the event-based - methods (create_task_event, create_status_update_event, - create_artifact_update_event) instead for proper A2A streaming. - - Args: - chunk: LiteLLM ModelResponse chunk - request_id: Original A2A request ID - is_final: Whether this is the final chunk - - Returns: - A2A streaming chunk dict or None if no content - """ - # Extract delta content - content = "" - if chunk is not None and hasattr(chunk, "choices") and chunk.choices: - choice = chunk.choices[0] - if hasattr(choice, "delta") and choice.delta: - content = choice.delta.content or "" - - if not content and not is_final: - return None - - # Build A2A streaming chunk (legacy format) - a2a_chunk = { - "jsonrpc": "2.0", - "id": request_id, - "result": { - "message": { - "role": "agent", - "parts": [{"kind": "text", "text": content}], - "messageId": uuid4().hex, - }, - "final": is_final, - }, - } - - return a2a_chunk diff --git a/litellm/a2a_protocol/providers/pydantic_ai_agents/transformation.py b/litellm/a2a_protocol/providers/pydantic_ai_agents/transformation.py index e73b17ac3c0..bf68a01d98c 100644 --- a/litellm/a2a_protocol/providers/pydantic_ai_agents/transformation.py +++ b/litellm/a2a_protocol/providers/pydantic_ai_agents/transformation.py @@ -289,16 +289,16 @@ class PydanticAITransformation: Transform Pydantic AI task response to standard A2A non-streaming format. Pydantic AI returns a task with history/artifacts, but the standard A2A - non-streaming format expects: + non-streaming format expects ``result`` to be the Message directly + (``kind="message"``), per the A2A spec / ``SendMessageResponse``: { "jsonrpc": "2.0", "id": "...", "result": { - "message": { - "role": "agent", - "parts": [{"kind": "text", "text": "..."}], - "messageId": "..." - } + "kind": "message", + "role": "agent", + "parts": [{"kind": "text", "text": "..."}], + "messageId": "..." } } @@ -316,6 +316,7 @@ class PydanticAITransformation: # Build standard A2A message a2a_message = { + "kind": "message", "role": "agent", "parts": parts if parts else [{"kind": "text", "text": full_text}], "messageId": message_id, @@ -325,9 +326,7 @@ class PydanticAITransformation: return { "jsonrpc": "2.0", "id": request_id, - "result": { - "message": a2a_message, - }, + "result": a2a_message, } @staticmethod diff --git a/litellm/a2a_protocol/utils.py b/litellm/a2a_protocol/utils.py index 1cdbde97755..0dbd1eefc63 100644 --- a/litellm/a2a_protocol/utils.py +++ b/litellm/a2a_protocol/utils.py @@ -60,6 +60,12 @@ class A2ARequestUtils: if not isinstance(result, dict): return "" + # Direct message format (A2A spec): detect by explicit kind tag only. + # The "parts" heuristic is too broad and would match any future result + # type that happens to include a "parts" field. + if result.get("kind") == "message": + return A2ARequestUtils.extract_text_from_message(result) + message = result.get("message", {}) return A2ARequestUtils.extract_text_from_message(message) diff --git a/litellm/anthropic_interface/exceptions/exception_mapping_utils.py b/litellm/anthropic_interface/exceptions/exception_mapping_utils.py index 28020e763f4..4548185bbdc 100644 --- a/litellm/anthropic_interface/exceptions/exception_mapping_utils.py +++ b/litellm/anthropic_interface/exceptions/exception_mapping_utils.py @@ -9,7 +9,6 @@ from typing import Dict, Optional from .exceptions import AnthropicErrorResponse, AnthropicErrorType - # HTTP status code -> Anthropic error type # Source: https://docs.anthropic.com/en/api/errors ANTHROPIC_ERROR_TYPE_MAP: Dict[int, AnthropicErrorType] = { diff --git a/litellm/anthropic_interface/exceptions/exceptions.py b/litellm/anthropic_interface/exceptions/exceptions.py index 984390fa702..b289e493e6b 100644 --- a/litellm/anthropic_interface/exceptions/exceptions.py +++ b/litellm/anthropic_interface/exceptions/exceptions.py @@ -2,7 +2,6 @@ from typing_extensions import Literal, Required, TypedDict - # Known Anthropic error types # Source: https://docs.anthropic.com/en/api/errors AnthropicErrorType = Literal[ diff --git a/litellm/anthropic_interface/messages/__init__.py b/litellm/anthropic_interface/messages/__init__.py index 0996d62c866..f71279b226d 100644 --- a/litellm/anthropic_interface/messages/__init__.py +++ b/litellm/anthropic_interface/messages/__init__.py @@ -10,7 +10,7 @@ This is an __init__.py file to allow the following interface """ -from typing import Any, AsyncIterator, Coroutine, Dict, List, Optional, Union +from typing import Any, AsyncIterator, Coroutine, Dict, Iterator, List, Optional, Union from litellm.llms.anthropic.experimental_pass_through.messages.handler import ( anthropic_messages as _async_anthropic_messages, @@ -100,8 +100,11 @@ def create( **kwargs, ) -> Union[ AnthropicMessagesResponse, + Iterator[bytes], AsyncIterator[Any], - Coroutine[Any, Any, Union[AnthropicMessagesResponse, AsyncIterator[Any]]], + Coroutine[ + Any, Any, Union[AnthropicMessagesResponse, AsyncIterator[Any], Iterator[bytes]] + ], ]: """ Async wrapper for Anthropic's messages API diff --git a/litellm/batches/batch_utils.py b/litellm/batches/batch_utils.py index aaf083e75d6..74e753b09ea 100644 --- a/litellm/batches/batch_utils.py +++ b/litellm/batches/batch_utils.py @@ -113,8 +113,11 @@ def _batch_cost_calculator( """ Calculate the cost of a batch based on the output file id """ - # Handle Vertex AI with specialized method - if custom_llm_provider == "vertex_ai" and model_name: + if ( + custom_llm_provider == "vertex_ai" + and model_name + and getattr(litellm, "disable_vertex_batch_output_transformation", False) + ): batch_cost, _ = calculate_vertex_ai_batch_cost_and_usage( file_content_dictionary, model_name ) @@ -136,10 +139,13 @@ def calculate_vertex_ai_batch_cost_and_usage( model_name: Optional[str] = None, ) -> Tuple[float, Usage]: """ - Calculate both cost and usage from Vertex AI batch responses. + Calculate both cost and usage from raw Vertex AI batch responses. - Vertex AI batch output lines have format: - {"request": ..., "status": "", "response": {"candidates": [...], "usageMetadata": {...}}} + Used only when ``litellm.disable_vertex_batch_output_transformation = True``. + In that case the GCS predictions.jsonl is returned as-is, with each line in + the native Vertex format: + + {"request": ..., "response": {"candidates": [...], "usageMetadata": {...}}} usageMetadata contains promptTokenCount, candidatesTokenCount, totalTokenCount. """ @@ -362,8 +368,11 @@ def _get_batch_job_total_usage_from_file_content( """ Get the tokens of a batch job from the file content """ - # Handle Vertex AI with specialized method - if custom_llm_provider == "vertex_ai" and model_name: + if ( + custom_llm_provider == "vertex_ai" + and model_name + and getattr(litellm, "disable_vertex_batch_output_transformation", False) + ): _, batch_usage = calculate_vertex_ai_batch_cost_and_usage( file_content_dictionary, model_name ) diff --git a/litellm/budget_manager.py b/litellm/budget_manager.py index b25967579e0..bbebb6042cb 100644 --- a/litellm/budget_manager.py +++ b/litellm/budget_manager.py @@ -178,6 +178,18 @@ class BudgetManager: return list(self.user_dict.keys()) def reset_cost(self, user): + """ + Reset the tracked spend for a user back to zero. + + Clears both the aggregate ``current_cost`` and the per-model + ``model_cost`` breakdown stored for the given user. + + Args: + user: The user identifier whose cost should be reset. + + Returns: + dict: ``{"user": }`` reflecting the reset state. + """ self.user_dict[user]["current_cost"] = 0 self.user_dict[user]["model_cost"] = {} return {"user": self.user_dict[user]} diff --git a/litellm/caching/caching_handler.py b/litellm/caching/caching_handler.py index 3cf1d911d7f..3f4e54382c9 100644 --- a/litellm/caching/caching_handler.py +++ b/litellm/caching/caching_handler.py @@ -87,6 +87,16 @@ class CachingHandlerResponse(BaseModel): in_memory_cache_obj = InMemoryCache() +def _is_chat_completion_cached_dict(cached_result: dict) -> bool: + cached_id = cached_result.get("id") + if isinstance(cached_id, str) and cached_id.startswith("chatcmpl"): + return True + obj = cached_result.get("object") + if isinstance(obj, str): + return obj.startswith("chat.completion") + return "choices" in cached_result + + def _should_defer_streaming_cache_hit_callbacks(*, kwargs: Dict[str, Any]) -> bool: """ When stream=True, do not run success callbacks at cache-hit time. @@ -861,27 +871,47 @@ class LLMCachingHandler: elif (call_type == "aresponses" or call_type == "responses") and isinstance( cached_result, dict ): - from litellm.responses.streaming_iterator import ( - CachedResponsesAPIStreamingIterator, - ) - - response_obj = ResponsesAPIResponse(**cached_result) - if ( - hasattr(response_obj, "_hidden_params") - and response_obj._hidden_params is not None - and isinstance(response_obj._hidden_params, dict) - ): - response_obj._hidden_params["cache_hit"] = True - - if kwargs.get("stream", False) is True: - cached_result = CachedResponsesAPIStreamingIterator( - response=response_obj, - logging_obj=logging_obj, - request_data=kwargs, - call_type=call_type, - ) + use_chat_completion_cache = _is_chat_completion_cached_dict(cached_result) + if use_chat_completion_cache: + if kwargs.get("stream", False) is True: + bridge_call_type = ( + CallTypes.acompletion.value + if call_type == "aresponses" + else CallTypes.completion.value + ) + cached_result = self._convert_cached_stream_response( + cached_result=cached_result, + call_type=bridge_call_type, + logging_obj=logging_obj, + model=model, + ) + else: + cached_result = convert_to_model_response_object( + response_object=cached_result, + model_response_object=ModelResponse(), + ) else: - cached_result = response_obj + from litellm.responses.streaming_iterator import ( + CachedResponsesAPIStreamingIterator, + ) + + response_obj = ResponsesAPIResponse(**cached_result) + if ( + hasattr(response_obj, "_hidden_params") + and response_obj._hidden_params is not None + and isinstance(response_obj._hidden_params, dict) + ): + response_obj._hidden_params["cache_hit"] = True + + if kwargs.get("stream", False) is True: + cached_result = CachedResponsesAPIStreamingIterator( + response=response_obj, + logging_obj=logging_obj, + request_data=kwargs, + call_type=call_type, + ) + else: + cached_result = response_obj if ( hasattr(cached_result, "_hidden_params") diff --git a/litellm/completion_extras/litellm_responses_transformation/handler.py b/litellm/completion_extras/litellm_responses_transformation/handler.py index ce398ee8288..2de7bda6467 100644 --- a/litellm/completion_extras/litellm_responses_transformation/handler.py +++ b/litellm/completion_extras/litellm_responses_transformation/handler.py @@ -37,6 +37,15 @@ class ResponsesToCompletionBridgeHandler: stream = litellm_params.get("stream", False) return bool(stream) + @staticmethod + def _is_preformatted_cached_chat_stream(result: Any) -> bool: + from litellm.litellm_core_utils.streaming_handler import CustomStreamWrapper + + return ( + isinstance(result, CustomStreamWrapper) + and result.custom_llm_provider == "cached_response" + ) + @staticmethod def _coerce_response_object( response_obj: Any, @@ -177,6 +186,8 @@ class ResponsesToCompletionBridgeHandler: **request_data, ) + from litellm.types.utils import ModelResponse + stream = self._resolve_stream_flag(optional_params, litellm_params) if isinstance(result, ResponsesAPIResponse): return self.transformation_handler.transform_response( @@ -192,6 +203,8 @@ class ResponsesToCompletionBridgeHandler: api_key=kwargs.get("api_key"), json_mode=kwargs.get("json_mode"), ) + elif isinstance(result, ModelResponse): + return result elif not stream: responses_api_response = self._collect_response_from_stream(result) return self.transformation_handler.transform_response( @@ -208,6 +221,10 @@ class ResponsesToCompletionBridgeHandler: json_mode=kwargs.get("json_mode"), ) else: + if self._is_preformatted_cached_chat_stream(result): + return self._apply_post_stream_processing( + result, model, custom_llm_provider + ) completion_stream = self.transformation_handler.get_model_response_iterator( streaming_response=result, # type: ignore sync_stream=True, @@ -256,6 +273,8 @@ class ResponsesToCompletionBridgeHandler: aresponses=True, ) + from litellm.types.utils import ModelResponse + stream = self._resolve_stream_flag(optional_params, litellm_params) if isinstance(result, ResponsesAPIResponse): return self.transformation_handler.transform_response( @@ -271,6 +290,8 @@ class ResponsesToCompletionBridgeHandler: api_key=kwargs.get("api_key"), json_mode=kwargs.get("json_mode"), ) + elif isinstance(result, ModelResponse): + return result elif not stream: responses_api_response = await self._collect_response_from_stream_async( result @@ -289,6 +310,10 @@ class ResponsesToCompletionBridgeHandler: json_mode=kwargs.get("json_mode"), ) else: + if self._is_preformatted_cached_chat_stream(result): + return self._apply_post_stream_processing( + result, model, custom_llm_provider + ) completion_stream = self.transformation_handler.get_model_response_iterator( streaming_response=result, # type: ignore sync_stream=False, diff --git a/litellm/completion_extras/litellm_responses_transformation/transformation.py b/litellm/completion_extras/litellm_responses_transformation/transformation.py index 32423f23314..51abbbf729b 100644 --- a/litellm/completion_extras/litellm_responses_transformation/transformation.py +++ b/litellm/completion_extras/litellm_responses_transformation/transformation.py @@ -30,6 +30,11 @@ from litellm.llms.base_llm.base_model_iterator import BaseModelResponseIterator from litellm.llms.base_llm.bridges.completion_transformation import ( CompletionTransformationBridge, ) +from litellm.responses.sse_output_recovery import ( + parse_sse_json_chunk, + record_output_item_chunk, + record_output_text_chunk, +) from litellm.types.llms.openai import ( ChatCompletionAnnotation, ChatCompletionReasoningItem, @@ -97,7 +102,7 @@ def _build_reasoning_item( def _reasoning_item_to_response_input( - r_item: Union[ChatCompletionReasoningItem, Dict[str, Any]] + r_item: Union[ChatCompletionReasoningItem, Dict[str, Any]], ) -> Dict[str, Any]: """Convert a stored ChatCompletionReasoningItem back to a Responses API input item.""" r_input: Dict[str, Any] = { @@ -601,6 +606,79 @@ class LiteLLMResponsesTransformationHandler(CompletionTransformationBridge): return choices + @classmethod + def _extract_output_from_completed_event( + cls, parsed_chunk: Dict[str, Any] + ) -> Optional[List[Dict[str, Any]]]: + response_payload = parsed_chunk.get("response") + if not isinstance(response_payload, dict): + return None + response_output = response_payload.get("output") + if not isinstance(response_output, list) or len(response_output) == 0: + return None + return cast(List[Dict[str, Any]], response_output) + + @classmethod + def _recover_output_items_from_raw_sse( + cls, raw_sse: Optional[str] + ) -> List[Dict[str, Any]]: + if not raw_sse or not isinstance(raw_sse, str): + return [] + + recovered_output_items: Dict[int, Dict[str, Any]] = {} + recovered_text_only_items: Dict[int, Dict[str, Any]] = {} + + for chunk in raw_sse.splitlines(): + parsed_chunk = parse_sse_json_chunk(chunk) + if parsed_chunk is None: + continue + + event_type = parsed_chunk.get("type") + + if event_type == ResponsesAPIStreamEvents.RESPONSE_COMPLETED: + recovered_output = cls._extract_output_from_completed_event( + parsed_chunk + ) + if recovered_output is not None: + return recovered_output + continue + + if event_type == ResponsesAPIStreamEvents.OUTPUT_ITEM_DONE: + record_output_item_chunk( + parsed_chunk=parsed_chunk, + output_items=recovered_output_items, + ) + continue + + if event_type == ResponsesAPIStreamEvents.OUTPUT_TEXT_DONE: + record_output_text_chunk( + parsed_chunk=parsed_chunk, + output_items=recovered_output_items, + text_only_items=recovered_text_only_items, + ) + continue + + # Merge text-only items into the recovered output items. Real + # OUTPUT_ITEM_DONE events take precedence at any given output_index, + # but text-only items at indices without a matching OUTPUT_ITEM_DONE + # must still be preserved (e.g. multi-output responses where some + # indices only emitted OUTPUT_TEXT_DONE). + merged_items: Dict[int, Dict[str, Any]] = {**recovered_text_only_items} + merged_items.update(recovered_output_items) + + if merged_items: + return [item for _, item in sorted(merged_items.items())] + + return [] + + @classmethod + def _recover_output_items_from_logging( + cls, logging_obj: "LiteLLMLoggingObj" + ) -> List[Dict[str, Any]]: + model_call_details = getattr(logging_obj, "model_call_details", {}) or {} + original_response = model_call_details.get("original_response") + return cls._recover_output_items_from_raw_sse(original_response) + def transform_response( # noqa: PLR0915 self, model: str, @@ -625,9 +703,22 @@ class LiteLLMResponsesTransformationHandler(CompletionTransformationBridge): if raw_response.error is not None: raise ValueError(f"Error in response: {raw_response.error}") + output_items = raw_response.output + if len(output_items) == 0: + recovered_output_items = self._recover_output_items_from_logging( + logging_obj + ) + if recovered_output_items: + output_items = cast(Any, recovered_output_items) + raw_response.output = cast(Any, recovered_output_items) + verbose_logger.warning( + "Recovered empty Responses API output from raw SSE for model=%s", + model, + ) + # Convert response output to choices using the static helper choices = self._convert_response_output_to_choices( - output_items=raw_response.output, + output_items=output_items, handle_raw_dict_callback=self._handle_raw_dict_response_item, ) @@ -641,7 +732,7 @@ class LiteLLMResponsesTransformationHandler(CompletionTransformationBridge): ) else: raise ValueError( - f"Unknown items in responses API response: {raw_response.output}" + f"Unknown items in responses API response: {output_items}" ) setattr(model_response, "choices", choices) @@ -1141,6 +1232,14 @@ class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator): event_type = parsed_chunk.get("type") if isinstance(event_type, ResponsesAPIStreamEvents): event_type = event_type.value + + if parsed_chunk.get("object") == "chat.completion.chunk" or ( + event_type is None + and isinstance(parsed_chunk.get("choices"), list) + and parsed_chunk.get("choices") + ): + return ModelResponseStream(**parsed_chunk) + verbose_logger.debug(f"Chat provider: Processing event type: {event_type}") if event_type == "response.created": @@ -1229,7 +1328,7 @@ class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator): raise ValueError( f"Chat provider: Invalid function argument delta {parsed_chunk}" ) - elif event_type == "response.output_item.done": + elif event_type == ResponsesAPIStreamEvents.OUTPUT_ITEM_DONE: # New output item added output_item = parsed_chunk.get("item", {}) if output_item.get("type") == "function_call": diff --git a/litellm/compression/content_detection.py b/litellm/compression/content_detection.py index 0655a42daf5..975117eb608 100644 --- a/litellm/compression/content_detection.py +++ b/litellm/compression/content_detection.py @@ -5,7 +5,6 @@ Auto-detect content type per message: code, JSON, or text. import json import re - _CODE_KEYWORDS = re.compile( r"\b(?:def |function |class |import |from |require\(|#include|fn |func |const |let |var |public |private |static )\b" ) diff --git a/litellm/constants.py b/litellm/constants.py index 072c2c358f7..ae98b37d6e6 100644 --- a/litellm/constants.py +++ b/litellm/constants.py @@ -1147,6 +1147,7 @@ BEDROCK_CONVERSE_MODELS = [ "openai.gpt-oss-120b-1:0", "anthropic.claude-haiku-4-5-20251001-v1:0", "anthropic.claude-sonnet-4-5-20250929-v1:0", + "anthropic.claude-opus-4-8", "anthropic.claude-opus-4-7", "anthropic.claude-opus-4-6-v1:0", "anthropic.claude-opus-4-6-v1", @@ -1408,6 +1409,13 @@ LITELLM_INTERNAL_JOBS_SERVICE_ACCOUNT_NAME = "litellm_internal_jobs" # Prometheus metrics, audit trails, or any other downstream consumer. LITELLM_PROXY_MASTER_KEY_ALIAS = "litellm_proxy_master_key" +# Marker placed in ``model_call_details`` on a synthetic ``Logging`` object that +# records a proxy-gate error (auth/rate-limit rejection) for a request that never +# reached an upstream provider. Tracing callbacks key off it to avoid fabricating +# an LLM-call span for a call that did not happen. See +# ``ProxyLogging._handle_logging_proxy_only_error``. +LITELLM_LOGGING_NO_UPSTREAM_LLM_CALL = "litellm_no_upstream_llm_call" + # Key Rotation Constants LITELLM_KEY_ROTATION_ENABLED = os.getenv("LITELLM_KEY_ROTATION_ENABLED", "false") LITELLM_KEY_ROTATION_CHECK_INTERVAL_SECONDS = int( @@ -1443,6 +1451,12 @@ CLI_JWT_EXPIRATION_HOURS = int( or os.getenv("LITELLM_CLI_JWT_EXPIRATION_HOURS") or 24 ) +# Comma-separated allowlisted OIDC claim map for CLI SSO polling, e.g. +# "employment_type->acme_employment_type,org_info.department->department" +CLI_SSO_CLAIM_MAP = ( + os.getenv("CLI_SSO_CLAIM_MAP") or os.getenv("LITELLM_CLI_SSO_CLAIM_MAP") or "" +) +CLI_SSO_CLAIM_MAX_SCALAR_LENGTH = 1024 ########################### UI SESSION DURATION ########################### # Duration for UI login session (username/password, SSO, invitation links). Format: "30s", "30m", "24h", "7d" @@ -1569,6 +1583,15 @@ DEFAULT_MANAGEMENT_OBJECT_IN_MEMORY_CACHE_TTL = int( os.getenv("DEFAULT_MANAGEMENT_OBJECT_IN_MEMORY_CACHE_TTL", 60) ) DEFAULT_ACCESS_GROUP_CACHE_TTL = int(os.getenv("DEFAULT_ACCESS_GROUP_CACHE_TTL", 600)) +# Short TTL for negative MCP access-group existence lookups. Keeps unauthenticated +# callers from forcing a DB query per request for unknown names, while bounding +# staleness so a transient DB error (which surfaces as an empty list) cannot +# hide a real group for long. +DEFAULT_MCP_ACCESS_GROUP_NEGATIVE_CACHE_TTL = 10 +# Maximum number of comma-separated MCP server / access-group tokens accepted +# in a single ``/{name1,name2,...}/mcp`` URL. Bounds the per-request DB / cache +# fan-out an authenticated caller can trigger by stuffing the path with tokens. +DEFAULT_MCP_NAMESPACE_CSV_MAX_TOKENS = 16 # Sentry Scrubbing Configuration SENTRY_DENYLIST = [ diff --git a/litellm/cost_calculator.py b/litellm/cost_calculator.py index 9b4dd80265c..9a4b158b622 100644 --- a/litellm/cost_calculator.py +++ b/litellm/cost_calculator.py @@ -24,6 +24,7 @@ from litellm.litellm_core_utils.llm_cost_calc.usage_object_transformation import from litellm.litellm_core_utils.llm_cost_calc.utils import ( CostCalculatorUtils, _generic_cost_per_character, + _get_regional_uplift_multiplier, _get_service_tier_cost_key, _parse_prompt_tokens_details, calculate_cost_component, @@ -132,6 +133,8 @@ _VIDEO_CALL_TYPES = frozenset( { CallTypes.create_video.value, CallTypes.acreate_video.value, + CallTypes.video_edit.value, + CallTypes.avideo_edit.value, CallTypes.video_remix.value, CallTypes.avideo_remix.value, } @@ -173,17 +176,45 @@ def _cost_per_token_custom_pricing_helper( prompt_tokens: float = 0, completion_tokens: float = 0, response_time_ms: Optional[float] = 0.0, + cached_tokens: float = 0, + cache_creation_tokens: float = 0, ### CUSTOM PRICING ### custom_cost_per_token: Optional[CostPerToken] = None, custom_cost_per_second: Optional[float] = None, ) -> Optional[Tuple[float, float]]: - """Internal helper function for calculating cost, if custom pricing given""" + """Internal helper function for calculating cost, if custom pricing given. + + prompt_tokens is assumed to include both cached_tokens and cache_creation_tokens + (OpenAI-compatible convention). Anthropic-style usage where prompt_tokens excludes + cache tokens is handled at the caller (cost_per_token) before invoking this helper. + """ if custom_cost_per_token is None and custom_cost_per_second is None: return None if custom_cost_per_token is not None: - input_cost = custom_cost_per_token["input_cost_per_token"] * prompt_tokens - output_cost = custom_cost_per_token["output_cost_per_token"] * completion_tokens + input_cost_per_token = custom_cost_per_token["input_cost_per_token"] + output_cost_per_token = custom_cost_per_token["output_cost_per_token"] + + cache_read_input_token_cost = custom_cost_per_token.get( + "cache_read_input_token_cost", + input_cost_per_token, + ) + cache_creation_input_token_cost = custom_cost_per_token.get( + "cache_creation_input_token_cost", + input_cost_per_token, + ) + + regular_prompt_tokens = max( + prompt_tokens - cached_tokens - cache_creation_tokens, + 0, + ) + + input_cost = ( + regular_prompt_tokens * input_cost_per_token + + cached_tokens * cache_read_input_token_cost + + cache_creation_tokens * cache_creation_input_token_cost + ) + output_cost = completion_tokens * output_cost_per_token return input_cost, output_cost elif custom_cost_per_second is not None: output_cost = custom_cost_per_second * response_time_ms / 1000 # type: ignore @@ -284,6 +315,10 @@ def cost_per_token( # noqa: PLR0915 audio_transcription_file_duration: float = 0.0, # for audio transcription calls - the file time in seconds ### SERVICE TIER ### service_tier: Optional[str] = None, # for OpenAI service tier pricing + ### DATA RESIDENCY ### + data_residency: Optional[ + str + ] = None, # for OpenAI regional-processing uplift (e.g. "eu", "us") response: Optional[Any] = None, ### REQUEST MODEL ### request_model: Optional[str] = None, # original request model for router detection @@ -323,10 +358,56 @@ def cost_per_token( # noqa: PLR0915 ) ## CUSTOM PRICING ## + # Normalize cache token counts across providers: + # - OpenAI-compatible: usage.prompt_tokens_details.cached_tokens + # (prompt_tokens already INCLUDES cached_tokens) + # - Anthropic: usage.cache_read_input_tokens / cache_creation_input_tokens + # (prompt_tokens does NOT include these — adjust before calling helper) + _cache_read_tokens: float = 0 + _cache_creation_tokens: float = 0 + _is_anthropic_style = False + + if usage_object is not None: + _pt_details = getattr(usage_object, "prompt_tokens_details", None) + if _pt_details is not None: + _cache_read_tokens = float(getattr(_pt_details, "cached_tokens", 0) or 0) + # OpenAI-compatible providers report cache-write tokens under + # either `cache_write_tokens` (kimi-k2) or `cache_creation_tokens`. + # Mirror db_spend_update_writer to stay symmetric. + _cache_creation_tokens = float( + getattr(_pt_details, "cache_write_tokens", 0) + or getattr(_pt_details, "cache_creation_tokens", 0) + or 0 + ) + + _anthropic_read = getattr(usage_object, "cache_read_input_tokens", None) + _anthropic_create = getattr(usage_object, "cache_creation_input_tokens", None) + if _anthropic_read is not None or _anthropic_create is not None: + _is_anthropic_style = True + if _anthropic_read is not None: + _cache_read_tokens = float(_anthropic_read) + if _anthropic_create is not None: + _cache_creation_tokens = float(_anthropic_create) + + if not _cache_read_tokens and cache_read_input_tokens: + _cache_read_tokens = float(cache_read_input_tokens) + _is_anthropic_style = True + if not _cache_creation_tokens and cache_creation_input_tokens: + _cache_creation_tokens = float(cache_creation_input_tokens) + _is_anthropic_style = True + + # Anthropic reports prompt_tokens as input_tokens (excluding cache tokens). + # Adjust so the helper's "prompt_tokens includes cache tokens" invariant holds. + _normalized_prompt_tokens = float(prompt_tokens) + if _is_anthropic_style: + _normalized_prompt_tokens += _cache_read_tokens + _cache_creation_tokens + response_cost = _cost_per_token_custom_pricing_helper( - prompt_tokens=prompt_tokens, + prompt_tokens=_normalized_prompt_tokens, completion_tokens=completion_tokens, response_time_ms=response_time_ms, + cached_tokens=_cache_read_tokens, + cache_creation_tokens=_cache_creation_tokens, custom_cost_per_second=custom_cost_per_second, custom_cost_per_token=custom_cost_per_token, ) @@ -338,9 +419,36 @@ def cost_per_token( # noqa: PLR0915 prompt_tokens_cost_usd_dollar: float = 0 completion_tokens_cost_usd_dollar: float = 0 model_cost_ref = litellm.model_cost + # Only callers that explicitly pass `custom_llm_provider` get the + # dedup/prefix-join treatment. When provider is omitted, preserve legacy + # behavior: `model_with_provider` stays equal to the raw `model` string + # (provider is detected below for downstream use only). + caller_supplied_provider = custom_llm_provider is not None + + # `model` is normally a string, but callers that mock the transport can pass + # non-string objects. Only run the string-based dedup/prefix-join when it is + # actually a string — e.g. a MagicMock's `.startswith()` is always truthy and + # its slices return new mocks, which would spin the dedup loop forever. + model_is_str = isinstance(model, str) + + # Router/proxy deployments may repeat the provider segment (e.g. model_name + # "openai/openai/gpt-5.5"). Strip duplicated `{provider}/` chains before joining. + if caller_supplied_provider and model_is_str: + _dup_prefix = f"{custom_llm_provider}/" + while model.startswith(_dup_prefix): + _remainder = model[len(_dup_prefix) :] + if _remainder.startswith(_dup_prefix): + model = _remainder + else: + break + model_with_provider = model - if custom_llm_provider is not None: - model_with_provider = custom_llm_provider + "/" + model + if caller_supplied_provider: + _prov_prefix = f"{custom_llm_provider}/" + if model_is_str and model.startswith(_prov_prefix): + model_with_provider = model + else: + model_with_provider = f"{custom_llm_provider}/{model}" if region_name is not None: model_with_provider_and_region = ( f"{custom_llm_provider}/{region_name}/{model}" @@ -351,6 +459,9 @@ def cost_per_token( # noqa: PLR0915 model_with_provider = model_with_provider_and_region else: _, custom_llm_provider, _, _ = litellm.get_llm_provider(model=model) + + assert custom_llm_provider is not None # caller-supplied or get_llm_provider + model_without_prefix = model model_parts = model.split("/", 1) if len(model_parts) > 1: @@ -419,6 +530,7 @@ def cost_per_token( # noqa: PLR0915 usage=usage_block, custom_llm_provider=custom_llm_provider, service_tier=service_tier, + data_residency=data_residency, ) return prompt_cost, completion_cost @@ -447,7 +559,10 @@ def cost_per_token( # noqa: PLR0915 or call_type == CallTypes.retrieve_batch ): return batch_cost_calculator( - usage=usage_block, model=model, custom_llm_provider=custom_llm_provider + usage=usage_block, + model=model, + custom_llm_provider=custom_llm_provider, + data_residency=data_residency, ) elif call_type == "atranscription" or call_type == "transcription": if _transcription_usage_has_token_details(usage_block): @@ -455,6 +570,7 @@ def cost_per_token( # noqa: PLR0915 model=model_without_prefix, usage=usage_block, service_tier=service_tier, + data_residency=data_residency, ) return openai_cost_per_second( @@ -505,7 +621,10 @@ def cost_per_token( # noqa: PLR0915 ) elif custom_llm_provider == "openai": return openai_cost_per_token( - model=model, usage=usage_block, service_tier=service_tier + model=model, + usage=usage_block, + service_tier=service_tier, + data_residency=data_residency, ) elif custom_llm_provider == "databricks": return databricks_cost_per_token(model=model, usage=usage_block) @@ -557,6 +676,7 @@ def cost_per_token( # noqa: PLR0915 usage=usage_block, custom_llm_provider=custom_llm_provider, service_tier=service_tier, + data_residency=data_residency, ) if ( @@ -1043,6 +1163,10 @@ def completion_cost( # noqa: PLR0915 litellm_logging_obj: Optional[LitellmLoggingObject] = None, ### SERVICE TIER ### service_tier: Optional[str] = None, # for OpenAI service tier pricing + ### DATA RESIDENCY ### + data_residency: Optional[ + str + ] = None, # for OpenAI regional-processing uplift (e.g. "eu", "us") ) -> float: """ Calculate the cost of a given completion call fot GPT-3.5-turbo, llama2, any litellm supported llm. @@ -1442,6 +1566,7 @@ def completion_cost( # noqa: PLR0915 combined_usage_object=cost_per_token_usage_object, custom_llm_provider=custom_llm_provider, litellm_model_name=model, + data_residency=data_residency, ) elif call_type == _MCP_CALL_TYPE: from litellm.proxy._experimental.mcp_server.cost_calculator import ( @@ -1526,6 +1651,7 @@ def completion_cost( # noqa: PLR0915 audio_transcription_file_duration=audio_transcription_file_duration, rerank_billed_units=rerank_billed_units, service_tier=service_tier, + data_residency=data_residency, response=completion_response, request_model=request_model_for_cost, ) @@ -1737,6 +1863,10 @@ def response_cost_calculator( litellm_logging_obj: Optional[LitellmLoggingObject] = None, ### SERVICE TIER ### service_tier: Optional[str] = None, # for OpenAI service tier pricing + ### DATA RESIDENCY ### + data_residency: Optional[ + str + ] = None, # for OpenAI regional-processing uplift (e.g. "eu", "us") ) -> float: """ Returns @@ -1770,6 +1900,7 @@ def response_cost_calculator( router_model_id=router_model_id, litellm_logging_obj=litellm_logging_obj, service_tier=service_tier, + data_residency=data_residency, ) return response_cost except Exception as e: @@ -1805,10 +1936,6 @@ def ocr_cost( if response.usage_info is None: raise ValueError("OCR response usage_info is None") - pages_processed = response.usage_info.pages_processed - if pages_processed is None: - raise ValueError("OCR response pages_processed is None") - try: model_info: Optional[ModelInfo] = litellm.get_model_info( model=model, custom_llm_provider=custom_llm_provider @@ -1816,9 +1943,49 @@ def ocr_cost( except Exception: model_info = None - ocr_cost_per_page: float = 0.0 + credits = getattr(response.usage_info, "credits", None) + cost_per_credit = None if model_info is not None: - ocr_cost_per_page = model_info.get("ocr_cost_per_page") or 0.0 + cost_per_credit = model_info.get("ocr_cost_per_credit") + if credits is not None and cost_per_credit is not None: + return cost_per_credit * credits, 0.0 + + ocr_cost_per_page: Optional[float] = None + if model_info is not None: + ocr_cost_per_page = model_info.get("ocr_cost_per_page") + + pages_processed = response.usage_info.pages_processed + if pages_processed is None: + if cost_per_credit is not None or ocr_cost_per_page is None: + # Surface missing usage data instead of silently under-reporting + # cost. The previous behavior raised ValueError; we now return 0.0 + # for credit-priced or unpriced models, so log a warning to keep + # the regression visible to operators. + verbose_logger.warning( + "OCR cost: model=%s custom_llm_provider=%s response.usage_info." + "pages_processed is None and credits=%s; returning 0.0 cost.", + model, + custom_llm_provider, + credits, + ) + return 0.0, 0.0 + raise ValueError("OCR response pages_processed is None") + + if ocr_cost_per_page is None: + # No per-page pricing configured. Either the model is on credit-based + # pricing (and credits weren't returned, so the credit branch above did + # not match) or the model has no OCR pricing entry at all. Surface a + # warning so that missing pricing entries are visible rather than + # silently producing zero cost for billable usage. + verbose_logger.warning( + "OCR cost: model=%s custom_llm_provider=%s reported " + "pages_processed=%s but no ocr_cost_per_page is configured; " + "returning 0.0 cost.", + model, + custom_llm_provider, + pages_processed, + ) + return 0.0, 0.0 total_ocr_processing_cost: float = ocr_cost_per_page * pages_processed return total_ocr_processing_cost, 0.0 @@ -2092,6 +2259,7 @@ def batch_cost_calculator( model: str, custom_llm_provider: Optional[str] = None, model_info: Optional[ModelInfo] = None, + data_residency: Optional[str] = None, ) -> Tuple[float, float]: """ Calculate the cost of a batch job. @@ -2120,6 +2288,26 @@ def batch_cost_calculator( ) except Exception: model_info = None + elif not any( + model_info.get(k) is not None + for k in ( + "input_cost_per_token_batches", + "input_cost_per_token", + "output_cost_per_token_batches", + "output_cost_per_token", + ) + ): + # model_info was provided (e.g. deployment metadata with only id/db_model) + # but carries no pricing fields. Fall back to the global pricing table so + # that standard model pricing is used instead of silently returning $0. + try: + global_info = litellm.get_model_info( + model=model, custom_llm_provider=custom_llm_provider + ) + if global_info: + model_info = global_info + except Exception: + pass if not model_info: return 0.0, 0.0 @@ -2156,6 +2344,11 @@ def batch_cost_calculator( usage.completion_tokens * (output_cost_per_token) / 2 ) # batch cost is usually half of the regular token cost + uplift = _get_regional_uplift_multiplier(model_info, data_residency) + if uplift != 1.0: + total_prompt_cost *= uplift + total_completion_cost *= uplift + return total_prompt_cost, total_completion_cost @@ -2301,6 +2494,7 @@ def handle_realtime_stream_cost_calculation( combined_usage_object: Usage, custom_llm_provider: str, litellm_model_name: str, + data_residency: Optional[str] = None, ) -> float: """ Handles the cost calculation for realtime stream responses. @@ -2331,6 +2525,7 @@ def handle_realtime_stream_cost_calculation( model=model_name, usage=combined_usage_object, custom_llm_provider=custom_llm_provider, + data_residency=data_residency, ) except Exception: continue diff --git a/litellm/exceptions.py b/litellm/exceptions.py index 8b005291556..17f5b43c273 100644 --- a/litellm/exceptions.py +++ b/litellm/exceptions.py @@ -918,9 +918,11 @@ class GuardrailRaisedException(Exception): guardrail_name: Optional[str] = None, message: str = "", should_wrap_with_default_message: bool = True, + status_code: int = 400, ): default_message = f"Guardrail raised an exception, Guardrail: {guardrail_name}, Message: {message}" self.guardrail_name = guardrail_name + self.status_code = status_code self.message = default_message if should_wrap_with_default_message else message super().__init__(self.message) @@ -930,12 +932,14 @@ class BlockedPiiEntityError(Exception): self, entity_type: str, guardrail_name: Optional[str] = None, + status_code: int = 400, ): """ Raised when a blocked entity is detected by a guardrail. """ self.entity_type = entity_type self.guardrail_name = guardrail_name + self.status_code = status_code self.message = f"Blocked entity detected: {entity_type} by Guardrail: {guardrail_name}. This entity is not allowed to be used in this request." super().__init__(self.message) diff --git a/litellm/files/types.py b/litellm/files/types.py index 688bc86f0cf..ba42a39f666 100644 --- a/litellm/files/types.py +++ b/litellm/files/types.py @@ -1,6 +1,5 @@ from typing import AsyncIterator, Dict, Iterator, Literal, NamedTuple, Union - FileContentProvider = Literal[ "openai", "azure", "vertex_ai", "bedrock", "hosted_vllm", "anthropic", "manus" ] diff --git a/litellm/google_genai/adapters/__init__.py b/litellm/google_genai/adapters/__init__.py index bfa9e712678..6fbe7d95a55 100644 --- a/litellm/google_genai/adapters/__init__.py +++ b/litellm/google_genai/adapters/__init__.py @@ -1,10 +1,10 @@ """ Google GenAI Adapters for LiteLLM -This module provides adapters for transforming Google GenAI generate_content requests +This module provides adapters for transforming Google GenAI generate_content requests to/from LiteLLM completion format with full support for: - Text content transformation -- Tool calling (function declarations, function calls, function responses) +- Tool calling (function declarations, function calls, function responses) - Streaming (both regular and tool calling) - Mixed content (text + tool calls) """ diff --git a/litellm/integrations/SlackAlerting/batching_handler.py b/litellm/integrations/SlackAlerting/batching_handler.py index fdce2e04793..828f3eb4175 100644 --- a/litellm/integrations/SlackAlerting/batching_handler.py +++ b/litellm/integrations/SlackAlerting/batching_handler.py @@ -1,9 +1,9 @@ """ -Handles Batching + sending Httpx Post requests to slack +Handles Batching + sending Httpx Post requests to slack -Slack alerts are sent every 10s or when events are greater than X events +Slack alerts are sent every 10s or when events are greater than X events -see custom_batch_logger.py for more details / defaults +see custom_batch_logger.py for more details / defaults """ from typing import TYPE_CHECKING, Any diff --git a/litellm/integrations/SlackAlerting/utils.py b/litellm/integrations/SlackAlerting/utils.py index e695266c88b..e2580768178 100644 --- a/litellm/integrations/SlackAlerting/utils.py +++ b/litellm/integrations/SlackAlerting/utils.py @@ -18,7 +18,7 @@ else: def process_slack_alerting_variables( - alert_to_webhook_url: Optional[Dict[AlertType, Union[List[str], str]]] + alert_to_webhook_url: Optional[Dict[AlertType, Union[List[str], str]]], ) -> Optional[Dict[AlertType, Union[List[str], str]]]: """ process alert_to_webhook_url diff --git a/litellm/integrations/additional_logging_utils.py b/litellm/integrations/additional_logging_utils.py index 795afd81d41..59319140a18 100644 --- a/litellm/integrations/additional_logging_utils.py +++ b/litellm/integrations/additional_logging_utils.py @@ -1,5 +1,5 @@ """ -Base class for Additional Logging Utils for CustomLoggers +Base class for Additional Logging Utils for CustomLoggers - Health Check for the logging util - Get Request / Response Payload for the logging util diff --git a/litellm/integrations/arize/arize_phoenix.py b/litellm/integrations/arize/arize_phoenix.py index b8cd04836c3..d48dba8e7bb 100644 --- a/litellm/integrations/arize/arize_phoenix.py +++ b/litellm/integrations/arize/arize_phoenix.py @@ -1,5 +1,7 @@ import os -from typing import TYPE_CHECKING, Any, Optional, Union +import threading +from collections import OrderedDict +from typing import TYPE_CHECKING, Any, Optional, Tuple, Union from litellm._logging import verbose_logger from litellm.integrations.arize import _utils @@ -8,8 +10,10 @@ from litellm.types.integrations.arize_phoenix import ArizePhoenixConfig if TYPE_CHECKING: from opentelemetry.sdk.trace import TracerProvider + from opentelemetry.sdk.trace.export import SpanProcessor from opentelemetry.trace import Span as _Span from opentelemetry.trace import SpanKind + from opentelemetry.trace import Tracer from litellm.integrations.opentelemetry import OpenTelemetry as _OpenTelemetry from litellm.integrations.opentelemetry import ( @@ -21,20 +25,27 @@ if TYPE_CHECKING: OpenTelemetryConfig = _OpenTelemetryConfig Span = Union[_Span, Any] OpenTelemetry = _OpenTelemetry + LITELLM_TRACER_NAME: str else: Protocol = Any OpenTelemetryConfig = Any Span = Any + Tracer = Any TracerProvider = Any SpanKind = Any - # Import OpenTelemetry at runtime + SpanProcessor = Any try: - from litellm.integrations.opentelemetry import OpenTelemetry + from litellm.integrations.opentelemetry import ( + LITELLM_TRACER_NAME, + OpenTelemetry, + ) except ImportError: + LITELLM_TRACER_NAME = "litellm" OpenTelemetry = None # type: ignore ARIZE_HOSTED_PHOENIX_ENDPOINT = "https://otlp.arize.com/v1/traces" +_MAX_PROJECT_PROVIDERS = 64 class ArizePhoenixLogger(OpenTelemetry): # type: ignore @@ -48,37 +59,142 @@ class ArizePhoenixLogger(OpenTelemetry): # type: ignore def _init_tracing(self, tracer_provider): """ - Override to always create a *private* TracerProvider for Arize Phoenix. + Override to create per-project TracerProviders (LRU-cached) for Arize Phoenix. The base ``OpenTelemetry._init_tracing`` falls back to the global TracerProvider when one already exists. That causes whichever integration initialises second to silently reuse the first one's exporter, so spans only reach one destination. - - By creating our own provider we guarantee Arize Phoenix always gets - its own exporter pipeline, regardless of initialisation order. """ - from opentelemetry.sdk.trace import TracerProvider from opentelemetry.trace import SpanKind if tracer_provider is not None: - # Explicitly supplied (e.g. in tests) — honour it. - self.tracer = tracer_provider.get_tracer("litellm") + self._use_injected_tracer_provider = True + self._shared_span_processor = None + self.tracer = tracer_provider.get_tracer(LITELLM_TRACER_NAME) self.span_kind = SpanKind return - # Always create a dedicated provider — never touch the global one. - provider = TracerProvider(resource=self._get_litellm_resource(self.config)) - provider.add_span_processor(self._get_span_processor()) - self.tracer = provider.get_tracer("litellm") + self._use_injected_tracer_provider = False + self._project_providers: OrderedDict[str, TracerProvider] = OrderedDict() + self._project_providers_lock = threading.Lock() + self._shared_span_processor = self._get_span_processor() self.span_kind = SpanKind + + default_project = self._resolve_project_name({}) + self.tracer = self._get_tracer_for(default_project) verbose_logger.debug( - "ArizePhoenixLogger: Created dedicated TracerProvider " - "(endpoint=%s, exporter=%s)", + "ArizePhoenixLogger: Initialized per-project TracerProvider cache " + "(default_project=%s, endpoint=%s, exporter=%s)", + default_project, self.config.endpoint, self.config.exporter, ) + def flush_tracer_providers(self) -> None: + """ + Flush all cached per-project providers and the shared span processor. + + Call on graceful proxy shutdown. Do not call on LRU eviction — in-flight + spans may still reference evicted providers. + """ + if getattr(self, "_use_injected_tracer_provider", False): + return + + shared_processor = getattr(self, "_shared_span_processor", None) + if shared_processor is not None: + try: + shared_processor.force_flush() + except Exception as e: + verbose_logger.debug( + "ArizePhoenixLogger: shared span processor force_flush failed: %s", + e, + ) + + with getattr(self, "_project_providers_lock", threading.Lock()): + providers = list(getattr(self, "_project_providers", {}).values()) + + for provider in providers: + try: + provider.force_flush() + except Exception as e: + verbose_logger.debug( + "ArizePhoenixLogger: TracerProvider force_flush failed: %s", e + ) + + def _get_litellm_resource_for_project(self, project_name: str): + """ + Build an OTEL Resource with project routing attrs that win over env detector. + + Phoenix uses ``openinference.project.name``; Arize AX uses ``model_id`` and + ``service.name``. Project attrs are merged last so OTEL_RESOURCE_ATTRIBUTES + from init does not pin every provider to one project. + """ + from opentelemetry.sdk.resources import OTELResourceDetector, Resource + + project_attributes: dict[str, str] = { + "openinference.project.name": project_name, + "model_id": project_name, + "service.name": project_name, + } + deployment_environment = getattr(self.config, "deployment_environment", None) + if deployment_environment is not None: + project_attributes["deployment.environment"] = deployment_environment + + env_resource = OTELResourceDetector().detect() + project_resource = Resource.create(project_attributes) # type: ignore[arg-type] + return env_resource.merge(project_resource) + + def _build_tracer_provider_for_project(self, project_name: str) -> TracerProvider: + """Create a TracerProvider for *project_name* (caller holds no cache lock).""" + from opentelemetry.sdk.trace import TracerProvider + + provider = TracerProvider( + resource=self._get_litellm_resource_for_project(project_name) + ) + provider.add_span_processor(self._shared_span_processor) + return provider + + def _get_tracer_for(self, project_name: str) -> Tracer: + """Return a tracer for *project_name*, creating/caching a provider on miss.""" + if getattr(self, "_use_injected_tracer_provider", False): + return self.tracer + + with self._project_providers_lock: + if project_name in self._project_providers: + self._project_providers.move_to_end(project_name) + return self._project_providers[project_name].get_tracer( + LITELLM_TRACER_NAME + ) + + # OTELResourceDetector().detect() is synchronous; build outside the lock so + # concurrent requests for other projects are not blocked on cache misses. + new_provider = self._build_tracer_provider_for_project(project_name) + + with self._project_providers_lock: + if project_name in self._project_providers: + self._project_providers.move_to_end(project_name) + return self._project_providers[project_name].get_tracer( + LITELLM_TRACER_NAME + ) + + if len(self._project_providers) >= _MAX_PROJECT_PROVIDERS: + self._project_providers.popitem(last=False) + + self._project_providers[project_name] = new_provider + return new_provider.get_tracer(LITELLM_TRACER_NAME) + + def _resolve_tracer_for_kwargs(self, kwargs: dict) -> Tuple[str, Tracer]: + """Resolve project name once and return the matching tracer.""" + project_name = self._resolve_project_name(kwargs) + return project_name, self._get_tracer_for(project_name) + + def get_tracer_to_use_for_request(self, kwargs: dict) -> Tracer: + """Route guardrail/raw-request spans to the same per-project tracer as the request.""" + if getattr(self, "_use_injected_tracer_provider", False): + return self.tracer + return self._resolve_tracer_for_kwargs(kwargs)[1] + def _init_otel_logger_on_litellm_proxy(self): """ Override: Arize Phoenix should NOT overwrite the proxy's @@ -93,56 +209,109 @@ class ArizePhoenixLogger(OpenTelemetry): # type: ignore @staticmethod def set_arize_phoenix_attributes(span: Span, kwargs, response_obj): - from litellm.integrations.opentelemetry_utils.base_otel_llm_obs_attributes import ( - safe_set_attribute, - ) - _utils.set_attributes(span, kwargs, response_obj, ArizeOTELAttributes) - - # Dynamic project name: check metadata first, then fall back to env var config - dynamic_project_name = ArizePhoenixLogger._get_dynamic_project_name(kwargs) - if dynamic_project_name: - safe_set_attribute(span, "openinference.project.name", dynamic_project_name) - else: - # Fall back to static config from env var - config = ArizePhoenixLogger.get_arize_phoenix_config() - if config.project_name: - safe_set_attribute( - span, "openinference.project.name", config.project_name - ) - return @staticmethod - def _get_dynamic_project_name(kwargs) -> Optional[str]: - """ - Retrieve dynamic Phoenix project name from request metadata. + def _normalize_project_name(name: Optional[str]) -> Optional[str]: + if name is None: + return None + normalized = str(name).strip() + return normalized if normalized else None - Users can set `metadata.phoenix_project_name` in their request to route - traces to different Phoenix projects dynamically. - """ - standard_logging_payload = kwargs.get("standard_logging_object") - if isinstance(standard_logging_payload, dict): - metadata = standard_logging_payload.get("metadata") + @staticmethod + def _iter_metadata_dicts_from_kwargs(kwargs: dict): + """Yield request metadata dicts; standard_logging_object before litellm_params.""" + for key in ("standard_logging_object", "litellm_params"): + found_key = kwargs.get(key) + if not isinstance(found_key, dict): + continue + metadata = found_key.get("metadata") if isinstance(metadata, dict): - project_name = metadata.get("phoenix_project_name") - if project_name: - return str(project_name) + yield metadata - # Also check litellm_params.metadata for SDK usage + @staticmethod + def _is_proxy_request(kwargs: dict) -> bool: + """True when the call is routed through the LiteLLM proxy. + + Proxy mode is determined solely by the server-set ``proxy_server_request`` + field in ``litellm_params``. Checking request metadata for + ``user_api_key_auth_metadata`` is intentionally avoided: that field is + user-supplied and would let an authenticated caller fake proxy-mode + detection to route their telemetry into arbitrary Arize/Phoenix projects. + """ litellm_params = kwargs.get("litellm_params") - if isinstance(litellm_params, dict): - metadata = litellm_params.get("metadata") or {} - else: - metadata = {} - if isinstance(metadata, dict): - project_name = metadata.get("phoenix_project_name") - if project_name: - return str(project_name) + return isinstance(litellm_params, dict) and bool( + litellm_params.get("proxy_server_request") + ) + @staticmethod + def _project_from_metadata_dict( + metadata: dict, metadata_key: str, *, proxy_mode: bool + ) -> Optional[str]: + """ + Read a Phoenix project field from proxy/SDK metadata. + + On the proxy, only ``user_api_key_auth_metadata`` (team/key config) may + select the project. SDK callers may still set project fields directly on + ``metadata``. + """ + auth_metadata = metadata.get("user_api_key_auth_metadata") + if isinstance(auth_metadata, dict): + project = ArizePhoenixLogger._normalize_project_name( + auth_metadata.get(metadata_key) + ) + if project: + return project + + if not proxy_mode: + return ArizePhoenixLogger._normalize_project_name( + metadata.get(metadata_key) + ) return None - def _get_phoenix_context(self, kwargs): + @staticmethod + def _metadata_project_from_kwargs(kwargs: dict, metadata_key: str) -> Optional[str]: + proxy_mode = ArizePhoenixLogger._is_proxy_request(kwargs) + for metadata in ArizePhoenixLogger._iter_metadata_dicts_from_kwargs(kwargs): + project = ArizePhoenixLogger._project_from_metadata_dict( + metadata, metadata_key, proxy_mode=proxy_mode + ) + if project: + return project + return None + + @staticmethod + def _resolve_project_name(kwargs: dict) -> str: + """ + Resolve the target Phoenix/Arize project for this request. + + Proxy priority: ``user_api_key_auth_metadata.phoenix_project_name_override``, + ``user_api_key_auth_metadata.phoenix_project_name``, env, then ``default``. + SDK priority: request metadata fields, then env, then ``default``. + """ + override = ArizePhoenixLogger._metadata_project_from_kwargs( + kwargs, "phoenix_project_name_override" + ) + if override: + return override + + phoenix_name = ArizePhoenixLogger._metadata_project_from_kwargs( + kwargs, "phoenix_project_name" + ) + if phoenix_name: + return phoenix_name + + env_name = ArizePhoenixLogger._normalize_project_name( + os.environ.get("PHOENIX_PROJECT_NAME") + or os.environ.get("ARIZE_PROJECT_NAME") + ) + if env_name: + return env_name + + return "default" + + def _get_phoenix_context(self, kwargs, tracer: Optional[Tracer] = None): """ Build a trace context for Phoenix's dedicated TracerProvider. @@ -159,11 +328,13 @@ class ArizePhoenixLogger(OpenTelemetry): # type: ignore """ from opentelemetry import trace + if tracer is None: + tracer = self._resolve_tracer_for_kwargs(kwargs)[1] + litellm_params = kwargs.get("litellm_params", {}) or {} proxy_server_request = litellm_params.get("proxy_server_request", {}) or {} headers = proxy_server_request.get("headers", {}) or {} - # Propagate distributed trace context if the caller sent a traceparent traceparent_ctx = ( self.get_traceparent_from_header(headers=headers) if headers.get("traceparent") @@ -173,10 +344,8 @@ class ArizePhoenixLogger(OpenTelemetry): # type: ignore is_proxy_mode = bool(proxy_server_request) if is_proxy_mode: - # Create a parent span on Phoenix's own tracer so both parent - # and child are exported to Phoenix. start_time_val = kwargs.get("start_time", kwargs.get("api_call_start_time")) - parent_span = self.tracer.start_span( + parent_span = tracer.start_span( name="litellm_proxy_request", start_time=( self._to_ns(start_time_val) if start_time_val is not None else None @@ -187,100 +356,77 @@ class ArizePhoenixLogger(OpenTelemetry): # type: ignore ctx = trace.set_span_in_context(parent_span) return ctx, parent_span - # SDK mode — no parent span needed return traceparent_ctx, None def _handle_success(self, kwargs, response_obj, start_time, end_time): - """ - Override to always create spans on ArizePhoenixLogger's dedicated TracerProvider. - - The base class's ``_get_span_context`` would find the parent span created by - the ``otel`` callback on the *global* TracerProvider. That span is invisible - in Phoenix (different exporter pipeline), so we ignore it and build our own - hierarchy via ``_get_phoenix_context``. - """ - from opentelemetry.trace import Status, StatusCode - - verbose_logger.debug( - "ArizePhoenixLogger: Logging kwargs: %s, OTEL config settings=%s", - kwargs, - self.config, + self._handle_phoenix_trace( + kwargs, response_obj, start_time, end_time, success=True ) - ctx, parent_span = self._get_phoenix_context(kwargs) - - # Create litellm_request span (child of our parent when in proxy mode) - span = self.tracer.start_span( - name=self._get_span_name(kwargs), - start_time=self._to_ns(start_time), - context=ctx, - ) - span.set_status(Status(StatusCode.OK)) - self.set_attributes(span, kwargs, response_obj) - - # Raw-request sub-span (if enabled) — must be created before - # ending the parent span so the hierarchy is valid. - self._maybe_log_raw_request(kwargs, response_obj, start_time, end_time, span) - span.end(end_time=self._to_ns(end_time)) - - # Guardrail span - self._create_guardrail_span(kwargs=kwargs, context=ctx) - - # Annotate and close our proxy parent span - if parent_span is not None: - parent_span.set_status(Status(StatusCode.OK)) - self.set_attributes(parent_span, kwargs, response_obj) - parent_span.end(end_time=self._to_ns(end_time)) - - # Metrics & cost recording - self._record_metrics(kwargs, response_obj, start_time, end_time) - - # Semantic logs - if self.config.enable_events: - self._emit_semantic_logs(kwargs, response_obj, span) - def _handle_failure(self, kwargs, response_obj, start_time, end_time): - """ - Override to always create failure spans on ArizePhoenixLogger's dedicated - TracerProvider. Mirrors ``_handle_success`` but sets ERROR status. - """ + self._handle_phoenix_trace( + kwargs, response_obj, start_time, end_time, success=False + ) + + def _handle_phoenix_trace( + self, + kwargs, + response_obj, + start_time, + end_time, + *, + success: bool, + ): from opentelemetry.trace import Status, StatusCode verbose_logger.debug( - "ArizePhoenixLogger: Failure - Logging kwargs: %s, OTEL config settings=%s", + "ArizePhoenixLogger: %s - kwargs: %s, OTEL config settings=%s", + "success" if success else "failure", kwargs, self.config, ) - ctx, parent_span = self._get_phoenix_context(kwargs) + _project_name, tracer = self._resolve_tracer_for_kwargs(kwargs) + ctx, parent_span = self._get_phoenix_context(kwargs, tracer=tracer) - # Create litellm_request span (child of our parent when in proxy mode) - span = self.tracer.start_span( + status = Status(StatusCode.OK if success else StatusCode.ERROR) + + span = tracer.start_span( name=self._get_span_name(kwargs), start_time=self._to_ns(start_time), context=ctx, ) - span.set_status(Status(StatusCode.ERROR)) + span.set_status(status) self.set_attributes(span, kwargs, response_obj) - self._record_exception_on_span(span=span, kwargs=kwargs) + if not success: + self._record_exception_on_span(span=span, kwargs=kwargs) + + if success: + self._maybe_log_raw_request( + kwargs, response_obj, start_time, end_time, span + ) span.end(end_time=self._to_ns(end_time)) - # Guardrail span self._create_guardrail_span(kwargs=kwargs, context=ctx) - # Annotate and close our proxy parent span if parent_span is not None: - parent_span.set_status(Status(StatusCode.ERROR)) + parent_span.set_status(status) self.set_attributes(parent_span, kwargs, response_obj) - self._record_exception_on_span(span=parent_span, kwargs=kwargs) + if not success: + self._record_exception_on_span(span=parent_span, kwargs=kwargs) parent_span.end(end_time=self._to_ns(end_time)) + if success: + self._record_metrics(kwargs, response_obj, start_time, end_time) + + if self.config.enable_events: + self._emit_semantic_logs(kwargs, response_obj, span) + @staticmethod def get_arize_phoenix_config() -> ArizePhoenixConfig: """ Retrieves the Arize Phoenix configuration based on environment variables. Returns: - ArizePhoenixConfig: A Pydantic model containing Arize Phoenix configuration. """ api_key = os.environ.get("PHOENIX_API_KEY", None) @@ -295,18 +441,15 @@ class ArizePhoenixLogger(OpenTelemetry): # type: ignore protocol: Protocol = "otlp_http" if collector_endpoint: - # Parse the endpoint to determine protocol if collector_endpoint.startswith("grpc://") or ( ":4317" in collector_endpoint and "/v1/traces" not in collector_endpoint ): endpoint = collector_endpoint protocol = "otlp_grpc" else: - # Phoenix Cloud endpoints (app.phoenix.arize.com) include the space in the URL if "app.phoenix.arize.com" in collector_endpoint: endpoint = collector_endpoint protocol = "otlp_http" - # For other HTTP endpoints, ensure they have the correct path elif "/v1/traces" not in collector_endpoint: if collector_endpoint.endswith("/v1"): endpoint = collector_endpoint + "/traces" @@ -318,7 +461,6 @@ class ArizePhoenixLogger(OpenTelemetry): # type: ignore endpoint = collector_endpoint protocol = "otlp_http" else: - # If no endpoint specified, self hosted phoenix endpoint = "http://localhost:6006/v1/traces" protocol = "otlp_http" verbose_logger.debug( @@ -329,12 +471,11 @@ class ArizePhoenixLogger(OpenTelemetry): # type: ignore if api_key is not None: otlp_auth_headers = f"Authorization=Bearer {api_key}" elif "app.phoenix.arize.com" in endpoint: - # Phoenix Cloud requires an API key raise ValueError( "PHOENIX_API_KEY must be set when using Phoenix Cloud (app.phoenix.arize.com)." ) - project_name = os.environ.get("PHOENIX_PROJECT_NAME", "default") + project_name = os.environ.get("PHOENIX_PROJECT_NAME") or "default" return ArizePhoenixConfig( otlp_auth_headers=otlp_auth_headers, @@ -343,8 +484,6 @@ class ArizePhoenixLogger(OpenTelemetry): # type: ignore project_name=project_name, ) - ## cannot suppress additional proxy server spans, removed previous methods. - async def async_health_check(self): config = self.get_arize_phoenix_config() diff --git a/litellm/integrations/custom_batch_logger.py b/litellm/integrations/custom_batch_logger.py index f9d4496c21f..8f4844501c3 100644 --- a/litellm/integrations/custom_batch_logger.py +++ b/litellm/integrations/custom_batch_logger.py @@ -1,5 +1,5 @@ """ -Custom Logger that handles batching logic +Custom Logger that handles batching logic Use this if you want your logs to be stored in memory and flushed periodically. """ @@ -14,22 +14,38 @@ from litellm.integrations.custom_logger import CustomLogger class CustomBatchLogger(CustomLogger): + preserve_events_added_during_flush = False + + # Default cap on the in-memory log queue. Prevents unbounded memory growth + # if ``async_send_batch`` consistently fails (e.g. the destination is + # unreachable) and events are preserved across flush attempts. Subclasses + # may override by passing ``max_queue_size`` or by setting the attribute + # directly (see ``RubrikLogger`` for an example). + DEFAULT_MAX_QUEUE_SIZE = 50_000 + def __init__( self, flush_lock: Optional[asyncio.Lock] = None, batch_size: Optional[int] = None, flush_interval: Optional[int] = None, + max_queue_size: Optional[int] = None, **kwargs, ) -> None: """ Args: flush_lock (Optional[asyncio.Lock], optional): Lock to use when flushing the queue. Defaults to None. Only used for custom loggers that do batching + max_queue_size (Optional[int], optional): Maximum number of events to retain in ``log_queue``. When the limit is exceeded (e.g. because the send destination is unreachable and events are preserved for retry), the oldest events are dropped. Defaults to ``DEFAULT_MAX_QUEUE_SIZE``. """ self.log_queue: List = [] self.flush_interval = flush_interval or litellm.DEFAULT_FLUSH_INTERVAL_SECONDS self.batch_size: int = batch_size or litellm.DEFAULT_BATCH_SIZE self.last_flush_time = time.time() self.flush_lock = flush_lock + self.max_queue_size: int = ( + max_queue_size + if max_queue_size is not None + else self.DEFAULT_MAX_QUEUE_SIZE + ) super().__init__(**kwargs) @@ -47,11 +63,40 @@ class CustomBatchLogger(CustomLogger): async with self.flush_lock: if self.log_queue: + log_queue_length = len(self.log_queue) verbose_logger.debug( "CustomLogger: Flushing batch of %s events", len(self.log_queue) ) - await self.async_send_batch() - self.log_queue.clear() + try: + await self.async_send_batch() + except Exception: + # If the underlying batch send raised, do NOT drop the + # in-flight events. They will be retried on the next flush. + # Most existing async_send_batch implementations swallow + # their own errors, so this only affects loggers that opt + # in to surfacing failures (e.g. Rubrik). + verbose_logger.exception( + "CustomLogger: async_send_batch raised; preserving " + "%s events in queue for retry", + log_queue_length, + ) + # Guard against unbounded queue growth if the destination + # is persistently unreachable. Drop the oldest events + # beyond ``max_queue_size``. + overflow = len(self.log_queue) - self.max_queue_size + if overflow > 0: + del self.log_queue[:overflow] + verbose_logger.warning( + "CustomLogger: log queue exceeded max_queue_size=%s; " + "dropped %s oldest events.", + self.max_queue_size, + overflow, + ) + return + if self.preserve_events_added_during_flush: + del self.log_queue[:log_queue_length] + else: + self.log_queue.clear() self.last_flush_time = time.time() async def async_send_batch(self, *args, **kwargs): diff --git a/litellm/integrations/custom_guardrail.py b/litellm/integrations/custom_guardrail.py index a03aef481e7..82a35f2eedd 100644 --- a/litellm/integrations/custom_guardrail.py +++ b/litellm/integrations/custom_guardrail.py @@ -43,7 +43,11 @@ if TYPE_CHECKING: dc = DualCache() -from litellm.exceptions import ModifyResponseException as ModifyResponseException +from litellm.exceptions import ( + BlockedPiiEntityError, + GuardrailRaisedException, + ModifyResponseException, +) class CustomGuardrail(CustomLogger): @@ -737,12 +741,15 @@ class CustomGuardrail(CustomLogger): (this was logged previously as an API failure - guardrail_failed_to_respond). Guardrails signal intentional blocks by raising: + - GuardrailRaisedException (generic guardrail API, tool permission) + - BlockedPiiEntityError (Presidio PII detection) - HTTPException with status 400 (content policy violation) - ModifyResponseException (passthrough mode violation) """ - if isinstance(e, ModifyResponseException): return True + if isinstance(e, (GuardrailRaisedException, BlockedPiiEntityError)): + return True if ( HTTPException is not None and isinstance(e, HTTPException) @@ -888,6 +895,15 @@ def log_guardrail_information(func): - pre_call - during_call - post_call + + Some guardrails (e.g. ``block_code_execution``) call + ``add_standard_logging_guardrail_information_to_request_data`` directly + from inside the wrapped function so they can record a richer payload + (structured detections, tracing detail) than this decorator's + "allow"/"mask"/raw-response default. To avoid double-recording in that + case (which would emit two spans, two Datadog records, two spend-log + entries, etc.), snapshot the entry count before invocation: if the + wrapped function already appended its own entry, skip the auto-record. """ import functools import inspect @@ -907,6 +923,16 @@ def log_guardrail_information(func): return GuardrailEventHooks.post_call return None + def _count_recorded_guardrail_entries(request_data: dict) -> int: + total = 0 + for container_key in ("metadata", "litellm_metadata"): + container = request_data.get(container_key) + if isinstance(container, dict): + entries = container.get("standard_logging_guardrail_information") + if isinstance(entries, list): + total += len(entries) + return total + @functools.wraps(func) async def async_wrapper(*args, **kwargs): start_time = datetime.now() # Move start_time inside the wrapper @@ -919,8 +945,11 @@ def log_guardrail_information(func): if func.__name__ == "apply_guardrail" and "inputs" in kwargs: original_inputs = kwargs.get("inputs") + entries_before = _count_recorded_guardrail_entries(request_data) try: response = await func(*args, **kwargs) + if _count_recorded_guardrail_entries(request_data) > entries_before: + return response return self._process_response( response=response, request_data=request_data, @@ -931,6 +960,8 @@ def log_guardrail_information(func): original_inputs=original_inputs, ) except Exception as e: + if _count_recorded_guardrail_entries(request_data) > entries_before: + raise return self._process_error( e=e, request_data=request_data, @@ -952,8 +983,11 @@ def log_guardrail_information(func): if func.__name__ == "apply_guardrail" and "inputs" in kwargs: original_inputs = kwargs.get("inputs") + entries_before = _count_recorded_guardrail_entries(request_data) try: response = func(*args, **kwargs) + if _count_recorded_guardrail_entries(request_data) > entries_before: + return response return self._process_response( response=response, request_data=request_data, @@ -962,6 +996,8 @@ def log_guardrail_information(func): original_inputs=original_inputs, ) except Exception as e: + if _count_recorded_guardrail_entries(request_data) > entries_before: + raise return self._process_error( e=e, request_data=request_data, diff --git a/litellm/integrations/custom_logger.py b/litellm/integrations/custom_logger.py index 300c311f36d..481cf7fce8e 100644 --- a/litellm/integrations/custom_logger.py +++ b/litellm/integrations/custom_logger.py @@ -697,6 +697,27 @@ class CustomLogger: # https://docs.litellm.ai/docs/observability/custom_callbac """ return AgenticLoopPlan(run_agentic_loop=False) + async def async_post_agentic_loop_response_hook( + self, + response: Any, + plan: AgenticLoopPlan, + kwargs: Dict, + ) -> Any: + """ + Post-process the response returned by the agentic-loop follow-up call. + + Called after BaseLLMHTTPHandler executes ``AgenticLoopPlan.request_patch`` + and receives the final response from the provider. Lets callbacks shape + what the client sees without bypassing the loop's safety / observability + machinery (depth tracking, fingerprinting, etc.). + + Use ``plan.metadata`` to carry whatever the build step decided to expose + for post-processing (e.g. native tool_result blocks to inject). + + Default returns ``response`` unchanged. + """ + return response + async def async_should_run_chat_completion_agentic_loop( self, response: Any, diff --git a/litellm/integrations/datadog/datadog_cost_management.py b/litellm/integrations/datadog/datadog_cost_management.py index a961d4f9244..0f954eb1ce0 100644 --- a/litellm/integrations/datadog/datadog_cost_management.py +++ b/litellm/integrations/datadog/datadog_cost_management.py @@ -2,10 +2,17 @@ import asyncio import os import time from datetime import datetime -from typing import Dict, List, Optional, Tuple +from typing import Any, Dict, List, Optional, Tuple, cast from litellm._logging import verbose_logger from litellm.integrations.custom_batch_logger import CustomBatchLogger +from litellm.integrations.datadog.datadog_handler import ( + get_datadog_env, + get_datadog_hostname, + get_datadog_pod_name, + get_datadog_service, +) +from litellm.litellm_core_utils.safe_json_dumps import safe_dumps from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, httpxSpecialProvider, @@ -15,9 +22,30 @@ from litellm.types.integrations.datadog_cost_management import ( ) from litellm.types.utils import StandardLoggingPayload +# Reserved tag keys whose values come from trusted sources (infra env, LiteLLM +# core payload fields, or proxy-controlled auth metadata). User-supplied +# request_tags / metadata cannot overwrite these, even when the key is +# allowlisted via cost_tag_keys, because that would let an authenticated caller +# spoof cost attribution (e.g. request_tags=["team:victim-team"]). +_RESERVED_TAG_KEYS: frozenset = frozenset( + { + "env", + "service", + "host", + "pod_name", + "provider", + "model", + "model_id", + "team", + "user", + "model_group", + } +) + class DatadogCostManagementLogger(CustomBatchLogger): - def __init__(self, **kwargs): + def __init__(self, cost_tag_keys: Optional[List[str]] = None, **kwargs): + self.cost_tag_keys: List[str] = list(cost_tag_keys) if cost_tag_keys else [] self.dd_api_key = os.getenv("DD_API_KEY") self.dd_app_key = os.getenv("DD_APP_KEY") self.dd_site = os.getenv("DD_SITE", "datadoghq.com") @@ -68,20 +96,21 @@ class DatadogCostManagementLogger(CustomBatchLogger): if not self.log_queue: return + batch_to_send = self.log_queue[:] + self.log_queue = [] + try: - # Aggregate costs from the batch - aggregated_entries = self._aggregate_costs(self.log_queue) - + aggregated_entries = self._aggregate_costs(batch_to_send) if not aggregated_entries: + verbose_logger.debug( + "Datadog Cost Management: batch produced no aggregable entries; " + "dropping %d log(s) from queue.", + len(batch_to_send), + ) return - - # Send to Datadog await self._upload_to_datadog(aggregated_entries) - - # Clear queue only on success (or if we decide to drop on failure) - # CustomBatchLogger clears queue in flush_queue, so we just process here - except Exception as e: + self.log_queue = batch_to_send + self.log_queue verbose_logger.exception( f"Datadog Cost Management: Error in async_send_batch: {str(e)}" ) @@ -151,45 +180,81 @@ class DatadogCostManagementLogger(CustomBatchLogger): return list(aggregator.values()) def _extract_tags(self, log: StandardLoggingPayload) -> Dict[str, str]: - from litellm.integrations.datadog.datadog_handler import ( - get_datadog_env, - get_datadog_hostname, - get_datadog_pod_name, - get_datadog_service, - ) - - tags = { + tags: Dict[str, str] = { "env": get_datadog_env(), "service": get_datadog_service(), "host": get_datadog_hostname(), "pod_name": get_datadog_pod_name(), } - # Add metadata as tags - metadata = log.get("metadata", {}) - if metadata: - # Add user info - # Add user info - if metadata.get("user_api_key_alias"): - tags["user"] = str(metadata["user_api_key_alias"]) + # Always-on canonical FOCUS dimensions from top-level payload fields. + # Non-sensitive and required for Datadog Custom Costs per-model attribution. + self._add_tag(tags, "provider", log.get("custom_llm_provider")) + self._add_tag(tags, "model", log.get("model")) + self._add_tag(tags, "model_id", log.get("model_id")) - # Add Team Tag - team_tag = ( - metadata.get("user_api_key_team_alias") - or metadata.get("team_alias") # type: ignore - or metadata.get("user_api_key_team_id") - or metadata.get("team_id") # type: ignore - ) + # cast because StandardLoggingMetadata is a TypedDict; we iterate it + # as a generic mapping below. + metadata: Dict[str, Any] = cast(Dict[str, Any], log.get("metadata") or {}) - if team_tag: - tags["team"] = str(team_tag) - # model_group is not in StandardLoggingMetadata TypedDict, so we need to access it via dict.get() - model_group = metadata.get("model_group") # type: ignore[misc] - if model_group: - tags["model_group"] = str(model_group) + # Backwards-compat: team/user/model_group preserved regardless of allowlist. + if metadata.get("user_api_key_alias"): + tags["user"] = str(metadata["user_api_key_alias"]) + team_tag = ( + metadata.get("user_api_key_team_alias") + or metadata.get("team_alias") + or metadata.get("user_api_key_team_id") + or metadata.get("team_id") + ) + if team_tag: + tags["team"] = str(team_tag) + if metadata.get("model_group"): + tags["model_group"] = str(metadata["model_group"]) + + # Allowlist-gated: request_tags (split on `:`) and arbitrary metadata.*. + # Reserved keys are hard-blocked here regardless of allowlist membership — + # see _RESERVED_TAG_KEYS for the rationale. + if self.cost_tag_keys: + allow = set(self.cost_tag_keys) + for rt in log.get("request_tags") or []: + if not isinstance(rt, str) or ":" not in rt: + continue + k, _, v = rt.partition(":") + if k in allow and v: + self._set_custom_tag(tags, k, v) + for k, v in metadata.items(): + if k in allow and v is not None and not isinstance(v, (dict, list)): + self._set_custom_tag(tags, k, str(v)) + for nested_key in ("spend_logs_metadata", "requester_metadata"): + nested = metadata.get(nested_key) + if isinstance(nested, dict): + for k, v in nested.items(): + if ( + k in allow + and v is not None + and not isinstance(v, (dict, list)) + ): + self._set_custom_tag(tags, k, str(v)) return tags + @staticmethod + def _set_custom_tag(tags: Dict[str, str], key: str, value: str) -> None: + if key in _RESERVED_TAG_KEYS: + verbose_logger.debug( + "Datadog Cost Management: dropping user-supplied tag %r=%r — " + "key is reserved for trusted cost attribution.", + key, + value, + ) + return + tags[key] = value + + @staticmethod + def _add_tag(tags: Dict[str, str], key: str, value: Any) -> None: + if value: + tags[key] = str(value) + async def _upload_to_datadog(self, payload: List[Dict]): if not self.dd_api_key or not self.dd_app_key: return @@ -201,8 +266,6 @@ class DatadogCostManagementLogger(CustomBatchLogger): } # The API endpoint expects a list of objects directly in the body (file content behavior) - from litellm.litellm_core_utils.safe_json_dumps import safe_dumps - data_json = safe_dumps(payload) response = await self.async_client.put( diff --git a/litellm/integrations/datadog/datadog_metrics.py b/litellm/integrations/datadog/datadog_metrics.py index fcf40701e28..d7847027d7e 100644 --- a/litellm/integrations/datadog/datadog_metrics.py +++ b/litellm/integrations/datadog/datadog_metrics.py @@ -144,7 +144,26 @@ class DatadogMetricsLogger(CustomBatchLogger): } self.log_queue.append(series_llm_latency) - # 3. Request Count / Status Code + # 3. LiteLLM Overhead Latency Metric (total - llm_api time) + hidden_params = log.get("hidden_params", {}) or {} + litellm_overhead_time_ms = hidden_params.get("litellm_overhead_time_ms") + if litellm_overhead_time_ms is not None: + overhead_tags = self._extract_tags(log) # no status_code on latency metric + series_overhead: DatadogMetricSeries = { + "metric": "litellm.overhead.latency", + "type": 3, # gauge + "points": [ + { + "timestamp": timestamp, + "value": litellm_overhead_time_ms + / 1000, # convert ms → seconds + } + ], + "tags": overhead_tags, + } + self.log_queue.append(series_overhead) + + # 4. Request Count / Status Code series_count: DatadogMetricSeries = { "metric": "litellm.llm_api.request_count", "type": 1, # count diff --git a/litellm/integrations/focus/transformer.py b/litellm/integrations/focus/transformer.py index b7d28e3dbb9..6f4433b4a05 100644 --- a/litellm/integrations/focus/transformer.py +++ b/litellm/integrations/focus/transformer.py @@ -9,7 +9,6 @@ import polars as pl from .schema import FOCUS_NORMALIZED_SCHEMA - _TAG_KEYS = ( "team_id", "team_alias", diff --git a/litellm/integrations/galileo.py b/litellm/integrations/galileo.py index e99d5f23a4c..a598124f612 100644 --- a/litellm/integrations/galileo.py +++ b/litellm/integrations/galileo.py @@ -1,18 +1,29 @@ +import json import os -from typing import Any, Dict, List, Optional +import re +from typing import Any, Dict, List, Optional, Tuple, cast from pydantic import BaseModel, Field import litellm from litellm._logging import verbose_logger from litellm.integrations.custom_logger import CustomLogger +from litellm.litellm_core_utils.prompt_templates.common_utils import ( + convert_content_list_to_str, + get_content_from_model_response, +) from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, httpxSpecialProvider, ) +from litellm.types.llms.openai import AllMessageValues + +GALILEO_CLOUD_API_BASE_URL = "https://api.galileo.ai" +# Cap the in-memory buffer so persistent flush failures (e.g. Galileo +# unavailable, invalid credentials) cannot leak memory unboundedly. +GALILEO_MAX_IN_MEMORY_RECORDS = 1000 -# from here: https://docs.rungalileo.io/galileo/gen-ai-studio-products/galileo-observe/how-to/logging-data-via-restful-apis#structuring-your-records class LLMResponse(BaseModel): latency_ms: int status_code: int @@ -37,65 +48,190 @@ class GalileoObserve(CustomLogger): def __init__(self) -> None: self.in_memory_records: List[dict] = [] self.batch_size = 1 - self.base_url = os.getenv("GALILEO_BASE_URL", None) - self.project_id = os.getenv("GALILEO_PROJECT_ID", None) + self.api_key = os.getenv("GALILEO_API_KEY") + self.project_id = os.getenv("GALILEO_PROJECT_ID") + self.log_stream_id = os.getenv("GALILEO_LOG_STREAM_ID") + self.username = os.getenv("GALILEO_USERNAME") + self.password = os.getenv("GALILEO_PASSWORD") + self.base_url = self._normalize_base_url(os.getenv("GALILEO_BASE_URL")) + if self.api_key and not self.base_url: + self.base_url = GALILEO_CLOUD_API_BASE_URL + self.use_v2_api = bool(self.api_key) self.headers: Optional[Dict[str, str]] = None self.async_httpx_handler = get_async_httpx_client( llm_provider=httpxSpecialProvider.LoggingCallback ) - pass - def set_galileo_headers(self): - # following https://docs.rungalileo.io/galileo/gen-ai-studio-products/galileo-observe/how-to/logging-data-via-restful-apis#logging-your-records + @staticmethod + def _normalize_base_url(base_url: Optional[str]) -> Optional[str]: + if base_url: + return base_url.rstrip("/") + return None - headers = { - "accept": "application/json", - "Content-Type": "application/x-www-form-urlencoded", - } - galileo_login_response = litellm.module_level_client.post( + def _is_configured(self) -> bool: + if not self.project_id or not self.base_url: + return False + if self.use_v2_api: + return bool(self.api_key) + return bool(self.username and self.password) + + async def async_set_galileo_headers(self) -> None: + galileo_login_response = await self.async_httpx_handler.post( url=f"{self.base_url}/login", - headers=headers, + headers={ + "accept": "application/json", + "Content-Type": "application/x-www-form-urlencoded", + }, data={ - "username": os.getenv("GALILEO_USERNAME"), - "password": os.getenv("GALILEO_PASSWORD"), + "username": self.username, + "password": self.password, }, ) - + galileo_login_response.raise_for_status() access_token = galileo_login_response.json()["access_token"] - self.headers = { "accept": "application/json", "Content-Type": "application/json", "Authorization": f"Bearer {access_token}", } - def get_output_str_from_response(self, response_obj, kwargs): - output = None - if response_obj is not None and ( - kwargs.get("call_type", None) == "embedding" - or isinstance(response_obj, litellm.EmbeddingResponse) - ): - output = None - elif response_obj is not None and isinstance( - response_obj, litellm.ModelResponse - ): - output = response_obj["choices"][0]["message"].json() - elif response_obj is not None and isinstance( - response_obj, litellm.TextCompletionResponse - ): - output = response_obj.choices[0].text - elif response_obj is not None and isinstance( - response_obj, litellm.ImageResponse - ): - output = response_obj["data"] + async def _ensure_headers(self) -> bool: + if self.headers is not None: + return True - return output + if self.use_v2_api: + if not self.api_key: + return False + self.headers = { + "accept": "application/json", + "Content-Type": "application/json", + "Galileo-API-Key": self.api_key, + } + return True + + if not (self.username and self.password and self.base_url): + return False + + try: + await self.async_set_galileo_headers() + return True + except Exception as e: + verbose_logger.debug("Galileo Logger: failed to authenticate: %s", e) + return False + + @staticmethod + def _galileo_input_messages( + messages: Optional[List[Any]], input_text: str + ) -> List[Dict[str, str]]: + if not messages: + return [{"role": "user", "content": input_text}] + + galileo_messages: List[Dict[str, str]] = [] + for message in messages: + if not isinstance(message, dict): + continue + role = message.get("role") + if not role: + continue + galileo_messages.append( + { + "role": str(role), + "content": convert_content_list_to_str( + message=cast(AllMessageValues, message) + ), + } + ) + + if galileo_messages: + return galileo_messages + return [{"role": "user", "content": input_text}] + + @staticmethod + def _record_to_v2_span(record: Dict[str, Any]) -> Dict[str, Any]: + created_at = record.get("created_at", "") + if created_at and not re.search(r"(Z|[+-]\d{2}:?\d{2})$", created_at): + created_at = f"{created_at}Z" + + span: Dict[str, Any] = { + "type": "llm", + "name": record.get("node_type", "litellm"), + "created_at": created_at, + "input": GalileoObserve._galileo_input_messages( + record.get("messages"), record.get("input_text", "") + ), + "output": { + "role": "assistant", + "content": record.get("output_text", ""), + }, + "status_code": record.get("status_code", 200), + "model": record.get("model"), + "metrics": { + "duration_ns": int(record.get("latency_ms", 0)) * 1_000_000, + "num_input_tokens": record.get("num_input_tokens"), + "num_output_tokens": record.get("num_output_tokens"), + }, + } + if record.get("tags"): + span["tags"] = record["tags"] + return span + + def _get_ingest_request(self) -> Optional[Tuple[str, Dict[str, Any]]]: + if not self.base_url or not self.project_id: + return None + + # Snapshot the records to be sent into a new list so concurrent appends + # during the network round-trip (across the await points in + # flush_in_memory_records) aren't silently dropped when we later clear + # the in-memory buffer. + records = list(self.in_memory_records) + + if self.use_v2_api: + payload: Dict[str, Any] = { + "spans": [self._record_to_v2_span(record) for record in records], + "reliable": False, + } + if self.log_stream_id: + payload["log_stream_id"] = self.log_stream_id + return ( + f"{self.base_url}/v2/projects/{self.project_id}/spans", + payload, + ) + + return ( + f"{self.base_url}/projects/{self.project_id}/observe/ingest", + {"records": records}, + ) + + def get_output_str_from_response( + self, response_obj: Any, kwargs: Dict[str, Any] + ) -> Optional[str]: + if response_obj is None: + return None + if kwargs.get("call_type", None) == "embedding" or isinstance( + response_obj, litellm.EmbeddingResponse + ): + return None + if isinstance(response_obj, litellm.TextCompletionResponse): + return response_obj.choices[0].text + if isinstance(response_obj, litellm.ImageResponse): + return json.dumps(response_obj["data"], default=str) + if isinstance(response_obj, (litellm.ModelResponse, dict)): + return get_content_from_model_response(response_obj) + return None async def async_log_success_event( self, kwargs: Any, response_obj: Any, start_time: Any, end_time: Any ): verbose_logger.debug("On Async Success") + if not self._is_configured(): + verbose_logger.debug( + "Galileo Logger: skipping flush — set GALILEO_PROJECT_ID and " + "either GALILEO_API_KEY (hosted) or GALILEO_USERNAME/GALILEO_PASSWORD " + "(enterprise Observe)." + ) + return + _latency_ms = int((end_time - start_time).total_seconds() * 1000) _call_type = kwargs.get("call_type", "litellm") input_text = litellm.utils.get_formatted_prompt( @@ -125,26 +261,69 @@ class GalileoObserve(CustomLogger): ), # timestamp str constructed in "%Y-%m-%dT%H:%M:%S" format ) - # dump to dict request_dict = request_record.model_dump() + messages = kwargs.get("messages") + if messages: + request_dict["messages"] = messages self.in_memory_records.append(request_dict) + # Bound the buffer so persistent flush failures cannot grow it + # without limit. Drop the oldest records once we exceed the cap. + if len(self.in_memory_records) > GALILEO_MAX_IN_MEMORY_RECORDS: + dropped = len(self.in_memory_records) - GALILEO_MAX_IN_MEMORY_RECORDS + self.in_memory_records = self.in_memory_records[ + -GALILEO_MAX_IN_MEMORY_RECORDS: + ] + verbose_logger.warning( + "Galileo Logger: in-memory buffer exceeded %s records; " + "dropped %s oldest record(s). Check Galileo connectivity/credentials.", + GALILEO_MAX_IN_MEMORY_RECORDS, + dropped, + ) + if len(self.in_memory_records) >= self.batch_size: await self.flush_in_memory_records() async def flush_in_memory_records(self): - verbose_logger.debug("flushing in memory records") - response = await self.async_httpx_handler.post( - url=f"{self.base_url}/projects/{self.project_id}/observe/ingest", - headers=self.headers, - json={"records": self.in_memory_records}, - ) + if not self.in_memory_records: + return - if response.status_code == 200: + # Capture the number of records that will be sent BEFORE any await so + # that concurrent appends made by other asyncio tasks during the + # network round-trip aren't silently dropped on the success-clear. + records_in_payload = len(self.in_memory_records) + + ingest_request = self._get_ingest_request() + if ingest_request is None: verbose_logger.debug( - "Galileo Logger:successfully flushed in memory records" + "Galileo Logger: missing GALILEO_BASE_URL or GALILEO_PROJECT_ID" ) - self.in_memory_records = [] + return + + if not await self._ensure_headers(): + verbose_logger.debug("Galileo Logger: could not set request headers") + return + + url, payload = ingest_request + verbose_logger.debug("flushing in memory records to %s", url) + + try: + response = await self.async_httpx_handler.post( + url=url, + headers=self.headers, + json=payload, + ) + except Exception as e: + verbose_logger.debug( + "Galileo Logger: failed to flush in memory records: %s", e + ) + return + + if response.is_success: + verbose_logger.debug( + "Galileo Logger: successfully flushed in memory records" + ) + del self.in_memory_records[:records_in_payload] else: verbose_logger.debug("Galileo Logger: failed to flush in memory records") verbose_logger.debug( @@ -152,6 +331,13 @@ class GalileoObserve(CustomLogger): response.text, response.status_code, ) + # Legacy enterprise auth caches a bearer token obtained from + # /login. If the request was rejected for auth reasons, drop the + # cached headers so the next flush re-authenticates instead of + # silently failing forever on a stale token. The v2 API key path + # uses a long-lived static key, so leave its headers in place. + if not self.use_v2_api and response.status_code in (401, 403): + self.headers = None async def async_log_failure_event(self, kwargs, response_obj, start_time, end_time): verbose_logger.debug("On Async Failure") diff --git a/litellm/integrations/opentelemetry.py b/litellm/integrations/opentelemetry.py index abce3c3e1c7..814da344f03 100644 --- a/litellm/integrations/opentelemetry.py +++ b/litellm/integrations/opentelemetry.py @@ -1,7 +1,7 @@ import os -from dataclasses import dataclass +from dataclasses import dataclass, field from datetime import datetime -from typing import TYPE_CHECKING, Any, Dict, List, Optional, Union, cast +from typing import TYPE_CHECKING, Any, Dict, List, Optional, Set, Union, cast import litellm from litellm._logging import verbose_logger @@ -10,6 +10,12 @@ from litellm.integrations._types.open_inference import ( SpanAttributes, ) from litellm.integrations.custom_logger import CustomLogger +from litellm.integrations.opentelemetry_utils.gen_ai_semconv import ( + OTEL_SEMCONV_STABILITY_OPT_IN_ENV, + OTELGenAISemconvMixin, + OTELSemconvCategory, + parse_semconv_opt_in, +) from litellm.litellm_core_utils.safe_json_dumps import safe_dumps from litellm.secret_managers.main import get_secret_bool, str_to_bool from litellm.types.services import ServiceLoggerPayload @@ -53,6 +59,14 @@ LITELLM_TRACER_NAME = os.getenv("OTEL_TRACER_NAME", "litellm") LITELLM_METER_NAME = os.getenv("LITELLM_METER_NAME", "litellm") LITELLM_LOGGER_NAME = os.getenv("LITELLM_LOGGER_NAME", "litellm") LITELLM_PROXY_REQUEST_SPAN_NAME = "Received Proxy Server Request" +# OTel-standard names. status is also kept under error.code for back compat. +HTTP_RESPONSE_STATUS_CODE_ATTRIBUTE = "http.response.status_code" +HTTP_ROUTE_ATTRIBUTE = "http.route" +URL_PATH_ATTRIBUTE = "url.path" +PREPROCESSING_DURATION_MS_ATTRIBUTE = "litellm.preprocessing.duration_ms" +TEAM_METADATA_ATTRIBUTE = "litellm.team.metadata" +MODEL_GROUP_ATTRIBUTE = "litellm.model_group" +PROVIDER_MODEL_ATTRIBUTE = "litellm.provider.model" # Remove the hardcoded LITELLM_RESOURCE dictionary - we'll create it properly later RAW_REQUEST_SPAN_NAME = "raw_gen_ai_request" LITELLM_REQUEST_SPAN_NAME = "litellm_request" @@ -85,6 +99,7 @@ class OpenTelemetryConfig: # Programmatic override for OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT. # One of NO_CONTENT, SPAN_ONLY, EVENT_ONLY, SPAN_AND_EVENT (or "true" as legacy alias). capture_message_content: Optional[str] = None + semconv_stability_opt_in: Set[OTELSemconvCategory] = field(default_factory=set) def __post_init__(self) -> None: # If endpoint is specified but exporter is still the default "console", @@ -110,6 +125,11 @@ class OpenTelemetryConfig: self.ignore_context_propagation = str_to_bool( os.getenv("OTEL_IGNORE_CONTEXT_PROPAGATION") ) + # Resolve the env opt-in once here so self.semconv_stability_opt_in is the + # single source of truth: the union of programmatic and env categories. + self.semconv_stability_opt_in |= parse_semconv_opt_in( + os.getenv(OTEL_SEMCONV_STABILITY_OPT_IN_ENV) + ) @classmethod def from_env(cls): @@ -157,7 +177,7 @@ class OpenTelemetryConfig: ) -class OpenTelemetry(CustomLogger): +class OpenTelemetry(OTELGenAISemconvMixin, CustomLogger): def __init__( self, config: Optional[OpenTelemetryConfig] = None, @@ -237,7 +257,14 @@ class OpenTelemetry(CustomLogger): not isinstance(cb, OpenTelemetry) for cb in litellm.service_callback ): litellm.service_callback.append(self) - setattr(proxy_server, "open_telemetry_logger", self) + # avoid proxy logger ownership being overwritten by later + # handlers. Multiple integrations (default OTEL, Langfuse OTEL, + # Arize OTEL, etc.) may initialize in sequence; without this guard, + # the last one silently replaces the first and breaks expected + # routing for proxy_server.open_telemetry_logger consumers. + # Behavior: first-registered wins. + if getattr(proxy_server, "open_telemetry_logger", None) is None: + setattr(proxy_server, "open_telemetry_logger", self) def _get_or_create_provider( self, @@ -648,6 +675,48 @@ class OpenTelemetry(CustomLogger): parent_otel_span = user_api_key_dict.parent_otel_span if parent_otel_span is not None: parent_otel_span.set_status(Status(StatusCode.ERROR)) + + # Stamp team attributes onto the SERVER (root) span too, so the + # trace root is team-filterable on the failure path like the + # child exception span below. + self._set_team_attributes_on_span( + span=parent_otel_span, + team_id=user_api_key_dict.team_id, + team_alias=user_api_key_dict.team_alias, + ) + + # Stamp structured error attrs on the SERVER span itself; the + # failure path otherwise only sets its status (_handle_failure + # records on the litellm_request child span). Inline import: + # litellm_logging <-> integrations is circular. + from litellm.litellm_core_utils.litellm_logging import ( + StandardLoggingPayloadSetup, + ) + + error_information = StandardLoggingPayloadSetup.get_error_information( + original_exception=original_exception, + traceback_str=traceback_str, + ) + self._record_exception_on_span( + span=parent_otel_span, + kwargs={ + "exception": original_exception, + "standard_logging_object": {"error_information": error_information}, + }, + ) + + # _record_exception_on_span only stamps when error_code is set; + # bare TypeError etc. has none, and the span is about to be ended. + error_code = ( + error_information.get("error_code") if error_information else None + ) + if not error_code: + self.set_response_status_code_attribute(parent_otel_span, 500) + + # Pre-request latency (request_data carries the propagated + # metadata on the failure path; omitted if it failed before handoff). + self.set_preprocessing_duration_attribute(parent_otel_span, request_data) + _span_name = "Failed Proxy Server Request" # Exception Logging Child Span @@ -660,12 +729,65 @@ class OpenTelemetry(CustomLogger): key="exception", value=str(original_exception), ) + self._set_team_attributes_on_span( + span=exception_logging_span, + team_id=user_api_key_dict.team_id, + team_alias=user_api_key_dict.team_alias, + ) exception_logging_span.set_status(Status(StatusCode.ERROR)) exception_logging_span.end(end_time=self._to_ns(datetime.now())) + # Emit guardrail spans for any guardrail invocations that + # ran during this request. _handle_failure typically does this, + # but for pre-call guardrail blocks the standard_logging_object + # may not carry guardrail_information by the time _handle_failure + # fires (the data lives only in request_data["metadata"]). Pull + # directly from request_data so the span is recorded either way; + # _emit_once dedupes if _handle_failure already emitted it. + self._emit_guardrail_spans_from_request_data( + request_data=request_data, + parent_span=parent_otel_span, + ) + # End Parent OTEL Sspan parent_otel_span.end(end_time=self._to_ns(datetime.now())) + def _emit_guardrail_spans_from_request_data( + self, + request_data: dict, + parent_span: Optional[Any], + ) -> None: + """Emit ``guardrail`` spans from ``request_data["metadata"] + ["standard_logging_guardrail_information"]``. + + Routed through ``_create_guardrail_span`` so the dedupe state in + ``_otel_internal`` is honoured — if ``_handle_failure`` already + emitted these spans for the same kwargs, this is a no-op. + """ + from opentelemetry import trace as _trace + + metadata = (request_data or {}).get("metadata") or {} + guardrail_information = metadata.get("standard_logging_guardrail_information") + if not guardrail_information: + return + + # _create_guardrail_span reads guardrail_information from + # kwargs["standard_logging_object"] and shares its dedupe state via + # kwargs["litellm_params"]["metadata"]["_otel_internal"]. Pass the + # SAME metadata dict the proxy populated so _handle_failure and + # this hook see the same dedupe markers. + kwargs: Dict[str, Any] = { + "litellm_params": {"metadata": metadata}, + "standard_logging_object": { + "guardrail_information": guardrail_information, + "metadata": metadata, + }, + } + context = ( + _trace.set_span_in_context(parent_span) if parent_span is not None else None + ) + self._create_guardrail_span(kwargs=kwargs, context=context) + async def async_post_call_success_hook( self, data: dict, @@ -684,6 +806,9 @@ class OpenTelemetry(CustomLogger): ctx, _ = self._get_span_context(kwargs, default_span=parent_span) + # Pre-request latency on the SERVER span (success path). + self.set_preprocessing_duration_attribute(parent_span, kwargs) + # 3. Guardrail span self._create_guardrail_span(kwargs=kwargs, context=ctx) @@ -794,12 +919,108 @@ class OpenTelemetry(CustomLogger): # End of Team/Key Based Logging Control Flow ######################################################### + def _emit_once(self, kwargs: dict, *scope: object) -> bool: + """Return True the first time this handler is asked to emit a span + for the given (handler, scope) on this kwargs; False on repeats. + + Used to suppress duplicate span emission for two distinct patterns: + + 1. **Handler-level dual-fire**: streaming code paths trigger both + the sync and async callback for one request, so ``_handle_success`` + / ``_handle_failure`` would otherwise produce two + ``litellm_request`` spans. Scope: ``("success",)`` / ``("failure",)``. + 2. **Payload-driven multi-entrypoint emission**: a span loop that + reads entries from ``standard_logging_payload`` (currently only + guardrails) is invoked from multiple lifecycle points + (post-call hooks, success callback, failure callback). The list + can be re-read with mutated entries between calls, so dedupe + must be at entry granularity. Scope: the entry's stable identity. + + ``scope`` parts can be any hashable identity. The marker is stored + in ``kwargs["litellm_params"]["metadata"]["_otel_internal"]`` so it + is request-local (kwargs is shared across the sync/async callbacks + and lifecycle hooks for one request). + """ + litellm_params = kwargs.get("litellm_params") + if not isinstance(litellm_params, dict): + litellm_params = {} + kwargs["litellm_params"] = litellm_params + + _metadata = litellm_params.get("metadata") + if not isinstance(_metadata, dict): + _metadata = {} + litellm_params["metadata"] = _metadata + + _otel_internal = _metadata.get("_otel_internal") + if not isinstance(_otel_internal, dict): + _otel_internal = {} + _metadata["_otel_internal"] = _otel_internal + + spans_logged = _otel_internal.get("spans_logged") + if not isinstance(spans_logged, dict): + spans_logged = {} + _otel_internal["spans_logged"] = spans_logged + + dedupe_key = (self.__class__.__name__, id(self), *scope) + if spans_logged.get(dedupe_key) is True: + return False + + spans_logged[dedupe_key] = True + return True + + def _end_proxy_span_from_kwargs(self, kwargs: dict, end_time) -> None: + """Close the proxy-level parent span if it is still recording. + + This helper retrieves the proxy span directly from kwargs metadata + and closes it after all child spans have been recorded. + + Only called from the success path. The failure path deliberately + leaves the proxy span open so ``async_post_call_failure_hook`` can + append the ``"Failed Proxy Server Request"`` child span before + closing it. + + Only spans named ``LITELLM_PROXY_REQUEST_SPAN_NAME`` are closed — + externally provided spans must not be closed by LiteLLM. + """ + litellm_params = kwargs.get("litellm_params", {}) or {} + _metadata = litellm_params.get("metadata", {}) or {} + proxy_span = _metadata.get("litellm_parent_otel_span", None) + if ( + proxy_span is not None + and getattr(proxy_span, "name", None) == LITELLM_PROXY_REQUEST_SPAN_NAME + and hasattr(proxy_span, "is_recording") + and proxy_span.is_recording() + ): + self._close_proxy_span_ok(proxy_span, end_time) + + def _close_proxy_span_ok(self, span: Span, end_time) -> None: + """Stamp http.response.status_code=200 + status=OK, then end the span.""" + from opentelemetry.trace import Status, StatusCode + + self.set_response_status_code_attribute(span, 200) + span.set_status(Status(StatusCode.OK)) + span.end(end_time=self._to_ns(end_time)) + def _handle_success(self, kwargs, response_obj, start_time, end_time): + """Create the litellm_request span then close the proxy span.""" verbose_logger.debug( "OpenTelemetry Logger: Logging kwargs: %s, OTEL config settings=%s", kwargs, self.config, ) + + # sync + async success handlers can both fire for one + # request (notably in streaming code paths). Guard against duplicate + # span writes — but still close the proxy span on the skip path so + # the trace doesn't leak an open root span. + if not self._emit_once(kwargs, "success"): + verbose_logger.debug( + "OpenTelemetry: skipping duplicate success span for handler=%s", + self.__class__.__name__, + ) + self._end_proxy_span_from_kwargs(kwargs, end_time) + return + ctx, parent_span = self._get_span_context(kwargs) if self.config.ignore_context_propagation: @@ -859,13 +1080,24 @@ class OpenTelemetry(CustomLogger): # 6. Do NOT end parent span - it should be managed by its creator # External spans (from Langfuse, user code, HTTP headers, global context) must not be closed by LiteLLM - # However, proxy-created spans should be closed here + # However, proxy-created spans should be closed here. if ( parent_span is not None and hasattr(parent_span, "name") and parent_span.name == LITELLM_PROXY_REQUEST_SPAN_NAME + and hasattr(parent_span, "is_recording") + and parent_span.is_recording() ): - parent_span.end(end_time=self._to_ns(end_time)) + self._close_proxy_span_ok(parent_span, end_time) + + # Stamp team attributes onto the SERVER (root) span before it is + # closed, so the trace root carries them like every child span. + self._set_team_attributes_on_proxy_span_from_kwargs(kwargs) + + # close the proxy span explicitly from kwargs metadata + # after all child spans (litellm_request, guardrail, raw_request) + # have been fully recorded and exported. + self._end_proxy_span_from_kwargs(kwargs, end_time) def _start_primary_span( self, @@ -879,13 +1111,14 @@ class OpenTelemetry(CustomLogger): otel_tracer: Tracer = self.get_tracer_to_use_for_request(kwargs) - # Always create a new span - # The parent relationship is preserved through the context parameter - span = otel_tracer.start_span( - name=self._get_span_name(kwargs), - start_time=self._to_ns(start_time), - context=context, - ) + span_kwargs: Dict[str, Any] = { + "name": self._get_span_name(kwargs), + "start_time": self._to_ns(start_time), + "context": context, + } + if self._gen_ai_semconv_latest_experimental: + span_kwargs["kind"] = self.span_kind.CLIENT + span = otel_tracer.start_span(**span_kwargs) span.set_status(Status(StatusCode.OK)) self.set_attributes(span, kwargs, response_obj) @@ -898,6 +1131,10 @@ class OpenTelemetry(CustomLogger): from opentelemetry import trace from opentelemetry.trace import Status, StatusCode + # raw_gen_ai_request is non-standard in semconv mode. + if self._gen_ai_semconv_latest_experimental: + return + if not self._capture_in_span(): return @@ -915,15 +1152,143 @@ class OpenTelemetry(CustomLogger): ) raw_span.set_status(Status(StatusCode.OK)) self.set_raw_request_attributes(raw_span, kwargs, response_obj) + self._set_team_attributes_from_kwargs(raw_span, kwargs) raw_span.end(end_time=self._to_ns(end_time)) + def _set_team_attributes_on_span( + self, + span: Span, + team_id: Optional[str], + team_alias: Optional[str], + ) -> None: + """Stamp team_id / team_alias onto a span so every child span of a + litellm_request trace carries them, not just the root span. + + Empty strings are treated as absent: a request made with the master + key or a team-less virtual key carries ``user_api_key_team_id=""`` + in ``standard_logging_object.metadata``; propagating that to every + span only adds noise that makes traces look mis-instrumented. + """ + if team_id: + self.safe_set_attribute( + span=span, + key="metadata.user_api_key_team_id", + value=team_id, + ) + if team_alias: + self.safe_set_attribute( + span=span, + key="metadata.user_api_key_team_alias", + value=team_alias, + ) + + def _set_team_attributes_from_kwargs(self, span: Span, kwargs: dict) -> None: + """Pull team_id / team_alias from the standard logging metadata in kwargs and stamp them onto span.""" + std_log = kwargs.get("standard_logging_object") + md: dict = {} + if isinstance(std_log, dict): + md = std_log.get("metadata") or {} + elif std_log is not None: + md = getattr(std_log, "metadata", None) or {} + self._set_team_attributes_on_span( + span=span, + team_id=md.get("user_api_key_team_id"), + team_alias=md.get("user_api_key_team_alias"), + ) + + def _set_team_attributes_on_proxy_span_from_kwargs(self, kwargs: dict) -> None: + """Stamp team attributes onto the proxy SERVER (root) span so the + trace root is filterable by team, not just its children. The root + span is created in auth before the team is resolved and is + otherwise only closed (never re-attributed) on the success path. + + Guarded to the LiteLLM-created proxy span (by name + recording) so + externally provided parent spans are never mutated. + """ + litellm_params = kwargs.get("litellm_params") or {} + metadata = litellm_params.get("metadata") or {} + proxy_span = metadata.get("litellm_parent_otel_span") + if ( + proxy_span is not None + and getattr(proxy_span, "name", None) == LITELLM_PROXY_REQUEST_SPAN_NAME + and hasattr(proxy_span, "is_recording") + and proxy_span.is_recording() + ): + self._set_team_attributes_from_kwargs(proxy_span, kwargs) + + def _set_inference_identity_attributes( + self, + span: Span, + standard_logging_payload: StandardLoggingPayload, + litellm_params: dict, + ) -> None: + """Stamp request-identity attributes onto an inference span so every + LLM-call span is filterable by the route it came in on, the team's + metadata, and both the user-facing (model_group alias) and the + dispatched (provider) model names. Empty/absent values are skipped. + """ + metadata = standard_logging_payload.get("metadata") or {} + + http_route = metadata.get("user_api_key_request_route") + if http_route: + self.safe_set_attribute( + span=span, key=HTTP_ROUTE_ATTRIBUTE, value=http_route + ) + + # ``user_api_key_team_metadata`` is dropped from the standard logging + # payload metadata, so read it from the raw request metadata in kwargs. + # ``metadata`` and ``litellm_metadata`` are alternate names for the same + # full metadata dict (the name varies by endpoint), so first-truthy wins. + raw_metadata = ( + litellm_params.get("metadata") + or litellm_params.get("litellm_metadata") + or {} + ) + team_metadata = self._team_metadata_json( + raw_metadata.get("user_api_key_team_metadata") + ) + if team_metadata: + self.safe_set_attribute( + span=span, key=TEAM_METADATA_ATTRIBUTE, value=team_metadata + ) + + model_group = standard_logging_payload.get("model_group") + if model_group: + self.safe_set_attribute( + span=span, key=MODEL_GROUP_ATTRIBUTE, value=model_group + ) + + hidden_params = standard_logging_payload.get("hidden_params") or {} + provider_model = hidden_params.get( + "litellm_model_name" + ) or standard_logging_payload.get("model") + if provider_model: + self.safe_set_attribute( + span=span, key=PROVIDER_MODEL_ATTRIBUTE, value=provider_model + ) + + @staticmethod + def _team_metadata_json(value: Any) -> Optional[str]: + """JSON-serialize a team's metadata dict for a single span attribute. + + Returns ``None`` for a missing, non-dict, or empty mapping so the + empty case is dropped rather than stamping a useless ``"{}"``. + """ + if not isinstance(value, dict) or not value: + return None + return safe_dumps(value) + def _record_metrics(self, kwargs, response_obj, start_time, end_time): duration_s = (end_time - start_time).total_seconds() params = kwargs.get("litellm_params") or {} provider = params.get("custom_llm_provider", "Unknown") common_attrs = { - "gen_ai.operation.name": "chat", + "gen_ai.operation.name": ( + self._gen_ai_operation_name(kwargs) + if self._gen_ai_semconv_latest_experimental + else "chat" + ), "gen_ai.system": provider, "gen_ai.request.model": kwargs.get("model"), "gen_ai.framework": "litellm", @@ -948,8 +1313,13 @@ class OpenTelemetry(CustomLogger): "mcp_tool_call_metadata", "vector_store_request_metadata", ]: - if md.get(key) is not None: - common_attrs[f"metadata.{key}"] = str(md[key]) + value = md.get(key) + if value is None: + continue + if isinstance(value, (dict, list)): + common_attrs[f"metadata.{key}"] = safe_dumps(value) + else: + common_attrs[f"metadata.{key}"] = str(value) # get hidden params hidden_params = getattr(std_log, "hidden_params", None) or (std_log or {}).get( @@ -1146,6 +1516,24 @@ class OpenTelemetry(CustomLogger): response_duration_seconds, attributes=common_attrs ) + @staticmethod + def _otel_log_types(): + """Resolve ``(LogRecord, SeverityNumber)`` across OTEL SDK versions. + + ``LogRecord`` moved out of ``opentelemetry.sdk._logs`` in OTEL >= 1.39.0 + (open-telemetry/opentelemetry-python#4676). Imports stay function-local + because the SDK is an optional dependency. + """ + from opentelemetry._logs import SeverityNumber + + try: + from opentelemetry.sdk._logs import LogRecord # OTEL < 1.39.0 + except ImportError: + from opentelemetry.sdk._logs._internal import ( # OTEL >= 1.39.0 + LogRecord, + ) + return LogRecord, SeverityNumber + def _emit_semantic_logs(self, kwargs, response_obj, span: Span): if not self.config.enable_events: return @@ -1159,16 +1547,7 @@ class OpenTelemetry(CustomLogger): # See: https://github.com/open-telemetry/opentelemetry-python/pull/4676 # TODO: Refactor to use the proper OTEL Logs API instead of directly creating SDK LogRecords - from opentelemetry._logs import SeverityNumber - - try: - from opentelemetry.sdk._logs import ( # type: ignore[attr-defined] # OTEL < 1.39.0 - LogRecord as SdkLogRecord, - ) - except ImportError: - from opentelemetry.sdk._logs._internal import ( - LogRecord as SdkLogRecord, # type: ignore[attr-defined] # OTEL >= 1.39.0 - ) + SdkLogRecord, SeverityNumber = self._otel_log_types() # Resolve through the handler's own LoggerProvider (which may be a # private one when skip_set_global=True) rather than the module-level @@ -1180,6 +1559,16 @@ class OpenTelemetry(CustomLogger): "custom_llm_provider", "Unknown" ) + if self._gen_ai_semconv_latest_experimental: + self._emit_inference_details_event( + kwargs=kwargs, + response_obj=response_obj, + provider=provider, + otel_logger=otel_logger, + parent_ctx=parent_ctx, + ) + return + # per-message events for msg in kwargs.get("messages", []): role = msg.get("role", "user") @@ -1296,6 +1685,21 @@ class OpenTelemetry(CustomLogger): for guardrail_information in guardrail_information_list: start_time_float = guardrail_information.get("start_time") end_time_float = guardrail_information.get("end_time") + + # ``_create_guardrail_span`` is called from three lifecycle + # points (``async_post_call_success_hook``, ``_handle_success``, + # ``_handle_failure``) and re-reads the (mutating) entry list + # each time. Dedupe at entry granularity so a single real + # guardrail invocation produces exactly one span per handler. + if not self._emit_once( + kwargs, + "guardrail", + guardrail_information.get("guardrail_name"), + start_time_float, + guardrail_information.get("guardrail_mode"), + ): + continue + start_time_datetime = datetime.now() if start_time_float is not None: start_time_datetime = datetime.fromtimestamp(start_time_float) @@ -1333,12 +1737,45 @@ class OpenTelemetry(CustomLogger): "masked_entity_count", safe_dumps(masked_entity_count) ) + guardrail_response = guardrail_information.get("guardrail_response") + if guardrail_response is not None: + guardrail_span.set_attribute( + "guardrail_response", safe_dumps(guardrail_response) + ) + + # Surface guardrail_status (success / guardrail_intervened / + # guardrail_failed_to_respond / not_run) as a top-level span + # attribute so trace backends can filter on it without parsing + # guardrail_response. self.safe_set_attribute( span=guardrail_span, - key="guardrail_response", - value=guardrail_information.get("guardrail_response"), + key="guardrail_status", + value=guardrail_information.get("guardrail_status"), ) + # Provider's raw top-level action (e.g. Bedrock's + # ``GUARDRAIL_INTERVENED`` / ``NONE``). Populated by the provider + # hook onto StandardLoggingGuardrailInformation so this integration + # stays provider-agnostic — we only read a normalised string. + guardrail_action = guardrail_information.get("guardrail_action") + if guardrail_action: + guardrail_span.set_attribute("guardrail_action", guardrail_action) + + # The provider hook (e.g. Bedrock) extracts violation_categories + # from the raw response BEFORE redaction and stamps them onto + # StandardLoggingGuardrailInformation. Surfacing them here as a + # queryable attribute lets dashboards group by violation category + # without parsing the redacted guardrail_response blob. + violation_categories = guardrail_information.get("violation_categories") + if violation_categories: + # OTel sequence attributes must be homogeneous primitives; + # serialise to JSON once so set_attribute never coerces. + guardrail_span.set_attribute( + "guardrail_violation_categories", safe_dumps(violation_categories) + ) + + self._set_team_attributes_from_kwargs(guardrail_span, kwargs) + guardrail_span.end(end_time=self._to_ns(end_time_datetime)) def _handle_failure(self, kwargs, response_obj, start_time, end_time): @@ -1349,6 +1786,21 @@ class OpenTelemetry(CustomLogger): kwargs, self.config, ) + + # sync + async failure handlers can both fire for one + # request (notably in streaming code paths), producing two + # semantically identical ERROR spans. Unlike the success path, the + # proxy span is intentionally left open here so that + # ``async_post_call_failure_hook`` can append the + # "Failed Proxy Server Request" child span before closing it — + # there is no proxy-span side-effect to preserve on the skip path. + if not self._emit_once(kwargs, "failure"): + verbose_logger.debug( + "OpenTelemetry: skipping duplicate failure span for handler=%s", + self.__class__.__name__, + ) + return + _parent_context, parent_otel_span = self._get_span_context(kwargs) if self.config.ignore_context_propagation: @@ -1366,11 +1818,14 @@ class OpenTelemetry(CustomLogger): if should_create_primary_span: # Span 1: Request sent to litellm SDK otel_tracer: Tracer = self.get_tracer_to_use_for_request(kwargs) - span = otel_tracer.start_span( - name=self._get_span_name(kwargs), - start_time=self._to_ns(start_time), - context=_parent_context, - ) + span_kwargs: Dict[str, Any] = { + "name": self._get_span_name(kwargs), + "start_time": self._to_ns(start_time), + "context": _parent_context, + } + if self._gen_ai_semconv_latest_experimental: + span_kwargs["kind"] = self.span_kind.CLIENT + span = otel_tracer.start_span(**span_kwargs) span.set_status(Status(StatusCode.ERROR)) self.set_attributes(span, kwargs, response_obj) @@ -1454,6 +1909,19 @@ class OpenTelemetry(CustomLogger): value=error_information["error_code"], ) + # Also expose under the OTel-standard name as an int + # (error_code is a str, may be non-numeric). + _error_code_val = error_information["error_code"] + if _error_code_val is not None: + try: + self.safe_set_attribute( + span=span, + key=HTTP_RESPONSE_STATUS_CODE_ATTRIBUTE, + value=int(_error_code_val), + ) + except (ValueError, TypeError): + pass + if error_information.get("error_class"): self.safe_set_attribute( span=span, @@ -1620,6 +2088,12 @@ class OpenTelemetry(CustomLogger): key="hidden_params", value=safe_dumps(hidden_params), ) + + self._set_inference_identity_attributes( + span=span, + standard_logging_payload=standard_logging_payload, + litellm_params=litellm_params, + ) # Cost breakdown tracking cost_breakdown: Optional[CostBreakdown] = standard_logging_payload.get( "cost_breakdown" @@ -1652,11 +2126,21 @@ class OpenTelemetry(CustomLogger): ) # The Generative AI Provider: Azure, OpenAI, etc. - self.safe_set_attribute( - span=span, - key=SpanAttributes.LLM_SYSTEM.value, - value=litellm_params.get("custom_llm_provider", "Unknown"), - ) + provider_name = litellm_params.get("custom_llm_provider", "Unknown") + # Latest-experimental semconv replaced gen_ai.system with + # gen_ai.provider.name; emit only the conformant key in that mode. + if self._gen_ai_semconv_latest_experimental: + self.safe_set_attribute( + span=span, + key="gen_ai.provider.name", + value=provider_name, + ) + else: + self.safe_set_attribute( + span=span, + key=SpanAttributes.LLM_SYSTEM.value, + value=provider_name, + ) # The maximum number of tokens the LLM generates for a request. if optional_params.get("max_tokens"): @@ -1682,11 +2166,17 @@ class OpenTelemetry(CustomLogger): value=optional_params.get("top_p"), ) - self.safe_set_attribute( - span=span, - key=SpanAttributes.LLM_IS_STREAMING.value, - value=str(optional_params.get("stream", False)), - ) + if self._gen_ai_semconv_latest_experimental: + # Semconv emits gen_ai.request.stream (only when streaming) via + # _set_semconv_request_attributes; skip the legacy llm.is_streaming. + self._set_semconv_request_attributes(span, optional_params) + self._set_semconv_cache_token_attributes(span, standard_logging_payload) + else: + self.safe_set_attribute( + span=span, + key=SpanAttributes.LLM_IS_STREAMING.value, + value=str(optional_params.get("stream", False)), + ) if optional_params.get("user"): self.safe_set_attribute( @@ -1807,14 +2297,18 @@ class OpenTelemetry(CustomLogger): value=safe_dumps(transformed_system_instructions), ) - self.safe_set_attribute( - span=span, - key=SpanAttributes.GEN_AI_OPERATION_NAME.value, - value=( + if self._gen_ai_semconv_latest_experimental: + operation_name = self._gen_ai_operation_name(kwargs) + else: + operation_name = ( "chat" if standard_logging_payload.get("call_type") == "completion" else standard_logging_payload.get("call_type") or "chat" - ), + ) + self.safe_set_attribute( + span=span, + key=SpanAttributes.GEN_AI_OPERATION_NAME.value, + value=operation_name, ) if standard_logging_payload.get("request_id"): @@ -2151,6 +2645,10 @@ class OpenTelemetry(CustomLogger): if generation_name: return generation_name + if self._gen_ai_semconv_latest_experimental: + model = kwargs.get("model") or "unknown" + return f"{self._gen_ai_operation_name(kwargs)} {model}" + return LITELLM_REQUEST_SPAN_NAME def get_traceparent_from_header(self, headers): @@ -2188,7 +2686,7 @@ class OpenTelemetry(CustomLogger): verbose_logger.debug( "OpenTelemetry: Using explicit parent span from metadata" ) - return trace.set_span_in_context(parent_otel_span), parent_otel_span + return trace.set_span_in_context(parent_otel_span), None # Priority 2: HTTP traceparent header if traceparent is not None: @@ -2627,6 +3125,11 @@ class OpenTelemetry(CustomLogger): management_endpoint_span.set_status(Status(StatusCode.OK)) management_endpoint_span.end(end_time=_end_time_ns) + # The management wrapper has no other hook that closes the SERVER span. + self.set_response_status_code_attribute(parent_otel_span, 200) + parent_otel_span.set_status(Status(StatusCode.OK)) + parent_otel_span.end(end_time=_end_time_ns) + async def async_management_endpoint_failure_hook( self, logging_payload: ManagementEndpointLoggingPayload, @@ -2677,6 +3180,24 @@ class OpenTelemetry(CustomLogger): management_endpoint_span.set_status(Status(StatusCode.ERROR)) management_endpoint_span.end(end_time=_end_time_ns) + # The management wrapper has no other hook that closes the SERVER span. + from litellm.litellm_core_utils.litellm_logging import ( + StandardLoggingPayloadSetup, + ) + + error_information = StandardLoggingPayloadSetup.get_error_information( + original_exception=_exception, + ) + parent_otel_span.set_status(Status(StatusCode.ERROR)) + self._record_exception_on_span( + span=parent_otel_span, + kwargs={ + "exception": _exception, + "standard_logging_object": {"error_information": error_information}, + }, + ) + parent_otel_span.end(end_time=_end_time_ns) + def create_litellm_proxy_request_started_span( self, start_time: datetime, @@ -2692,3 +3213,86 @@ class OpenTelemetry(CustomLogger): context=self.get_traceparent_from_header(headers=headers), kind=self.span_kind.SERVER, ) + + def set_proxy_request_route_attributes( + self, + span: Optional[Span], + *, + url_path: Optional[str] = None, + http_route: Optional[str] = None, + ) -> None: + """ + Set OTel-standard ``http.route`` / ``url.path`` on the proxy SERVER + span. Called from the auth path, the only point where both the + SERVER span and the request are in hand. No-op if span/value missing. + """ + if span is None: + return + if url_path: + self.safe_set_attribute(span=span, key=URL_PATH_ATTRIBUTE, value=url_path) + if http_route: + self.safe_set_attribute( + span=span, key=HTTP_ROUTE_ATTRIBUTE, value=http_route + ) + + def set_response_status_code_attribute( + self, span: Optional[Span], status_code: Optional[int] + ) -> None: + """ + Set OTel-standard ``http.response.status_code`` (int) on the proxy + SERVER span. The failure path sets this from the error code in + ``_record_exception_on_span``; this is the success-path counterpart + so the attribute is present on every SERVER span regardless of + outcome (required by the HTTP semconv, and needed for error-ratio / + status-breakdown dashboards). No-op if span/value missing. + """ + if span is None or status_code is None: + return + self.safe_set_attribute( + span=span, + key=HTTP_RESPONSE_STATUS_CODE_ATTRIBUTE, + value=int(status_code), + ) + + def set_preprocessing_duration_attribute( + self, span: Optional[Span], container: Any + ) -> None: + """ + Set ``litellm.preprocessing.duration_ms`` (proxy-receive -> first + provider handoff) on the proxy SERVER span. ``litellm_received_at`` + rides request metadata; ``first_api_call_start_time`` is the + set-once first-handoff instant (retries/backoff excluded). Works + uniformly for the success (model_call_details) and failure + (request_data) containers. No-op if span/either anchor is missing. + """ + if span is None or not isinstance(container, dict): + return + received_at = None + # first_api_call_start_time is top-level (never in user metadata). + first_handoff = container.get("first_api_call_start_time") + _lp = container.get("litellm_params") + for _md in ( + (_lp or {}).get("metadata") if isinstance(_lp, dict) else None, + container.get("metadata"), + container.get("litellm_metadata"), + ): + if isinstance(_md, dict): + received_at = received_at or _md.get("litellm_received_at") + if received_at is None or first_handoff is None: + return + try: + start_ts = self._to_timestamp(received_at) + end_ts = self._to_timestamp(first_handoff) + except Exception: + return + if start_ts is None or end_ts is None: + return + duration_ms = (end_ts - start_ts) * 1000.0 + # Clock skew → omit rather than emit a negative latency. + if duration_ms < 0: + return + self.safe_set_attribute( + span=span, + key=PREPROCESSING_DURATION_MS_ATTRIBUTE, + value=duration_ms, + ) diff --git a/litellm/integrations/opentelemetry_utils/gen_ai_semconv.py b/litellm/integrations/opentelemetry_utils/gen_ai_semconv.py new file mode 100644 index 00000000000..e45fe149e13 --- /dev/null +++ b/litellm/integrations/opentelemetry_utils/gen_ai_semconv.py @@ -0,0 +1,271 @@ +"""OTEL GenAI ``gen_ai_latest_experimental`` semantic conventions. + +Setting ``OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental`` switches the +emitted traces to the experimental OTEL GenAI conventions +(https://opentelemetry.io/docs/specs/semconv/gen-ai/). Concretely, versus the +default LiteLLM output: + +Request span: + +- name is ``{operation} {model}`` (e.g. ``chat gpt-4``) instead of + ``litellm_request``; span kind is ``CLIENT``. +- ``gen_ai.operation.name`` is the actual operation (``chat`` / + ``text_completion`` / ``embeddings``) instead of always ``chat``. +- the provider is reported as ``gen_ai.provider.name``; the superseded + ``gen_ai.system`` and the legacy ``llm.is_streaming`` are dropped. +- adds ``gen_ai.request.{frequency_penalty,presence_penalty,top_k,seed}``, + ``gen_ai.request.stop_sequences`` (a string array), + ``gen_ai.request.stream`` (only when streaming), + ``gen_ai.request.choice.count`` (only when n > 1), and + ``gen_ai.usage.cache_{creation,read}.input_tokens``. +- the non-standard ``raw_gen_ai_request`` child span is no longer created. + +Events: + +- the per-message ``gen_ai.content.prompt`` / per-choice + ``gen_ai.content.completion`` log events are replaced by a single + ``gen_ai.client.inference.operation.details`` log event carrying + ``gen_ai.input.messages`` / ``gen_ai.output.messages`` (message content + included only when content capture is enabled). +""" + +from datetime import datetime +from enum import Enum +from typing import TYPE_CHECKING, Any, Dict, List, Optional, Set, Tuple, Union + +from litellm.litellm_core_utils.safe_json_dumps import safe_dumps + +if TYPE_CHECKING: + from opentelemetry.trace import Span as _Span + + from litellm.integrations.opentelemetry import OpenTelemetryConfig + + Span = Union[_Span, Any] +else: + Span = Any + + +# OTEL_SEMCONV_STABILITY_OPT_IN is a comma-separated list of category-specific +# opt-in values. See https://opentelemetry.io/docs/specs/semconv/gen-ai/ +OTEL_SEMCONV_STABILITY_OPT_IN_ENV = "OTEL_SEMCONV_STABILITY_OPT_IN" + + +class OTELSemconvCategory(Enum): + GEN_AI_LATEST_EXPERIMENTAL = "gen_ai_latest_experimental" + + +# Reverse lookup: opt-in token string -> OTELSemconvCategory. +_SEMCONV_CATEGORY_BY_VALUE = { + category.value: category for category in OTELSemconvCategory +} + + +# LiteLLM optional_params key -> OTEL gen_ai semconv span attribute. +_SEMCONV_REQUEST_ATTRIBUTES = { + "frequency_penalty": "gen_ai.request.frequency_penalty", + "presence_penalty": "gen_ai.request.presence_penalty", + "top_k": "gen_ai.request.top_k", + "seed": "gen_ai.request.seed", +} + +# usage_object key -> OTEL gen_ai semconv cache-token span attribute. +_SEMCONV_CACHE_TOKEN_ATTRIBUTES = { + "cache_creation_input_tokens": "gen_ai.usage.cache_creation.input_tokens", + "cache_read_input_tokens": "gen_ai.usage.cache_read.input_tokens", +} + +# Name of the consolidated GenAI inference event (replaces the legacy +# per-message gen_ai.content.prompt / per-choice gen_ai.content.completion). +_INFERENCE_DETAILS_EVENT_NAME = "gen_ai.client.inference.operation.details" + + +def parse_semconv_opt_in(raw: Optional[str]) -> Set[OTELSemconvCategory]: + """Parse the comma-separated OTEL_SEMCONV_STABILITY_OPT_IN value into the + set of recognized categories. Unknown tokens are ignored per the spec.""" + if not raw: + return set() + return { + _SEMCONV_CATEGORY_BY_VALUE[token] + for token in (part.strip() for part in raw.split(",")) + if token in _SEMCONV_CATEGORY_BY_VALUE + } + + +class OTELGenAISemconvMixin: + """OTEL GenAI ``gen_ai_latest_experimental`` semantic-convention behavior. + + Mixed into ``OpenTelemetry`` (its only host). Every member is internal to + the OTEL integration; the leading underscore marks "subsystem-internal", + not "class-private" (the host lives in a sibling module). + + Members the host calls (the mixin -> host contract): + + - ``_gen_ai_semconv_latest_experimental`` -- opt-in gate; guards every + semconv code path in ``opentelemetry.py``. + - ``_gen_ai_operation_name`` -- LiteLLM ``call_type`` -> spec + ``gen_ai.operation.name``. + - ``_set_semconv_request_attributes`` / + ``_set_semconv_cache_token_attributes`` -- add the ``gen_ai.request.*`` + / ``gen_ai.usage.cache_*`` span attributes. + - ``_emit_inference_details_event`` -- emit the consolidated event. + + Helpers the host must provide (declared under ``TYPE_CHECKING`` below): + ``config``, ``safe_set_attribute``, ``_capture_in_event``, + ``_transform_messages_to_otel_semantic_conventions``, + ``_transform_choices_to_otel_semantic_conventions``, ``_to_ns``, + ``_otel_log_types``. + """ + + if TYPE_CHECKING: + config: "OpenTelemetryConfig" + + def safe_set_attribute(self, span: Span, key: str, value: Any) -> None: ... + + def _capture_in_event(self) -> bool: ... + + def _transform_messages_to_otel_semantic_conventions( + self, messages: Union[List[dict], str] + ) -> List[dict]: ... + + def _transform_choices_to_otel_semantic_conventions( + self, choices: List[dict] + ) -> List[dict]: ... + + def _to_ns(self, dt: datetime) -> int: ... + + def _otel_log_types(self) -> Tuple[Any, Any]: ... + + @property + def _gen_ai_semconv_latest_experimental(self) -> bool: + """Whether the ``gen_ai_latest_experimental`` opt-in is active. + + Every semconv behavior is gated on this; ``False`` => legacy output. + """ + return ( + OTELSemconvCategory.GEN_AI_LATEST_EXPERIMENTAL + in self.config.semconv_stability_opt_in + ) + + @staticmethod + def _gen_ai_operation_name(kwargs: dict) -> str: + """Map a LiteLLM ``call_type`` to spec ``gen_ai.operation.name``. + + Substring match (e.g. ``aembedding`` -> ``embeddings``); defaults to + ``chat``. + """ + call_type = kwargs.get("call_type", "") or "" + match call_type: + case s if "embedding" in s: + return "embeddings" + case s if "text_completion" in s: + return "text_completion" + case _: + return "chat" + + def _set_semconv_request_attributes( + self, span: Span, optional_params: dict + ) -> None: + """Add ``gen_ai.request.*`` span attributes from ``optional_params``. + + Covers the sampling params plus the conditionally-required + ``stop_sequences`` / ``stream`` / ``choice.count`` per the spec. + """ + for source_key, semconv_key in _SEMCONV_REQUEST_ATTRIBUTES.items(): + value = optional_params.get(source_key) + if value is not None: + self.safe_set_attribute(span=span, key=semconv_key, value=value) + + stop = optional_params.get("stop") + if stop is not None: + # Spec types this as string[]. safe_set_attribute coerces to a + # primitive, so set the array directly via the span API. + stop_list = stop if isinstance(stop, list) else [stop] + span.set_attribute( + "gen_ai.request.stop_sequences", [str(s) for s in stop_list] + ) + + # Conditionally required: set only when the request is streaming. + if optional_params.get("stream"): + self.safe_set_attribute(span=span, key="gen_ai.request.stream", value=True) + + # Conditionally required per spec ("if available and != 1"). Valid n is + # an int >= 1, so n > 1 is equivalent for conformant input while + # suppressing nonsensical values (0, negative, non-int). + n = optional_params.get("n") + if isinstance(n, int) and n > 1: + self.safe_set_attribute( + span=span, key="gen_ai.request.choice.count", value=n + ) + + def _set_semconv_cache_token_attributes( + self, span: Span, standard_logging_payload + ) -> None: + """Add ``gen_ai.usage.cache_*.input_tokens`` from the usage object. + + No-op when the payload or the usage values are missing/zero. + """ + if not standard_logging_payload: + return + usage = (standard_logging_payload.get("metadata") or {}).get( + "usage_object" + ) or {} + for source_key, semconv_key in _SEMCONV_CACHE_TOKEN_ATTRIBUTES.items(): + value = usage.get(source_key) + if value: + self.safe_set_attribute(span=span, key=semconv_key, value=value) + + def _build_inference_details_attrs( + self, kwargs: dict, response_obj: dict, provider: str + ) -> Dict[str, Any]: + """Build the attribute payload for the inference-details event. + + Always includes provider/operation; input/output messages are added + only when content capture is enabled and non-empty. Mixin-internal. + """ + attrs: Dict[str, Any] = { + "event_name": _INFERENCE_DETAILS_EVENT_NAME, + "gen_ai.provider.name": provider, + "gen_ai.operation.name": self._gen_ai_operation_name(kwargs), + } + if not self._capture_in_event(): + return attrs + + input_messages = self._transform_messages_to_otel_semantic_conventions( + kwargs.get("messages") or [] + ) + output_messages = self._transform_choices_to_otel_semantic_conventions( + response_obj.get("choices", []) + ) + if input_messages: + attrs["gen_ai.input.messages"] = safe_dumps(input_messages) + if output_messages: + attrs["gen_ai.output.messages"] = safe_dumps(output_messages) + return attrs + + def _emit_inference_details_event( + self, + kwargs: dict, + response_obj: dict, + provider: str, + otel_logger, + parent_ctx, + ) -> None: + """Emit the consolidated ``gen_ai.client.inference.operation.details`` + log event, correlated to the request span via ``parent_ctx``. + + Replaces the legacy per-message / per-choice content events. + """ + LogRecord, SeverityNumber = self._otel_log_types() + log_record = LogRecord( + timestamp=self._to_ns(datetime.now()), + trace_id=parent_ctx.trace_id, + span_id=parent_ctx.span_id, + trace_flags=parent_ctx.trace_flags, + severity_number=SeverityNumber.INFO, + severity_text="INFO", + body=None, + attributes=self._build_inference_details_attrs( + kwargs, response_obj, provider + ), + ) + otel_logger.emit(log_record) diff --git a/litellm/integrations/opik/utils.py b/litellm/integrations/opik/utils.py index b0ab5991c91..43577505c11 100644 --- a/litellm/integrations/opik/utils.py +++ b/litellm/integrations/opik/utils.py @@ -105,7 +105,7 @@ def _remove_nulls(x: Dict[str, Any]) -> Dict[str, Any]: def get_traces_and_spans_from_payload( - payload: List[Dict[str, Any]] + payload: List[Dict[str, Any]], ) -> Tuple[List[Dict[str, Any]], List[Dict[str, Any]]]: """ Separate traces and spans from payload. diff --git a/litellm/integrations/otel/README.md b/litellm/integrations/otel/README.md new file mode 100644 index 00000000000..99b3ecea162 --- /dev/null +++ b/litellm/integrations/otel/README.md @@ -0,0 +1,258 @@ +# OpenTelemetry instrumentation + +This package produces OpenTelemetry traces for LiteLLM. It is enabled by the +`LITELLM_OTEL_V2` environment variable (`is_otel_v2_enabled()` in +[`config.py`](./model/config.py)); when unset, nothing in this package runs. + +## What gets traced + +A traced proxy request produces one trace with two kinds of spans: + +``` +SERVER span "POST /v1/chat/completions" ← FastAPI instrumentation +├── INTERNAL span "auth /v1/chat/completions" ← auth phase ┐ +│ ├── CLIENT span "postgres get_key_object" ← datastore call │ +│ └── CLIENT span "postgres get_team_membership" │ +├── INTERNAL span "execute_guardrail …" ← guardrail │ this package +├── CLIENT span "chat gpt-4o" ← LLM call │ +└── CLIENT span "batch_write_to_db …" ← spend write ┘ +``` + +The gen-ai spans are siblings under the server span. In particular the guardrail +span is a sibling of the LLM call, not a child of it: pre/during/post-call +guardrail hooks are part of the request lifecycle (a pre-call guardrail runs +before the LLM call even starts), so they belong directly under the server span, +alongside the LLM call. + +Request-level spans (LLM call, guardrail) parent to the server span via an +**explicit anchor** — `context.set_request_root_span` captures the server span +once at request entry, and `resolve_request_span_context` reads it — rather than +to whatever span is momentarily active. Ambient-only parenting was wrong at two +boundaries: inside the live `auth` phase span the active span is `auth` (so the +span would nest under auth), and a pass-through request closes its span from a +detached `asyncio.create_task` where the server span is no longer active (so the +span orphaned into its own trace). The anchor — a contextvar inherited by those +child tasks — gives a stable parent in both cases. DB/service spans keep ambient +parenting so an auth DB lookup still nests under `auth`. + +**Which service calls become spans (`spans.span_role_for_service`).** LiteLLM's +service-logging layer instruments many internal functions, but only some are +traceable units of work: + +- **`DB_CALL` (CLIENT)** — outbound datastore calls (redis, postgres, + `batch_write_to_db`), carrying `db.system.name` / `db.operation.name` semconv. +- **`SERVICE` (INTERNAL)** — genuine internal work worth a span (background + budget/reset jobs, pod-lock manager). +- **metrics-only (no span)** — `self` (the `track_llm_api_timing` wrapper, which + duplicates the LLM-call span), `router` (duplicates the request), and + `proxy_pre_call` (a guardrail's real span is `execute_guardrail …`). These + still feed Prometheus/Datadog through their own hooks; they just never enter + the trace. `auth` is also excluded here because it gets a **live phase span** + instead (see below). + +Spans are named `"{service} {call_type}"` (e.g. `"redis set"`) so repeated calls +to one service stay distinguishable. Like every other span they parent to the +**ambient** context, falling back to the threaded `litellm_parent_otel_span` only +when ambient has no live span; a background job with neither starts its own root +trace. Caller-supplied `event_metadata` is **sanitized** before it reaches a span +(primitives only, no live objects, no secrets/headers, bounded) — see +`payloads.sanitize_event_metadata`. + +**Live phase spans.** `auth` is wrapped in a real, active span +(`logger.phase_span`) for the duration of authentication, so the DB lookups it +triggers nest **under** it instead of flattening onto the server span. Identity +Baggage (team/key/user) is seeded once the key resolves, so every post-auth span +inherits it; auth-internal DB lookups that run before the key is known stay +unlabeled, which is correct. + +**Status.** On success a span's status is left `UNSET` (the semconv default, +matching the FastAPI server span); only a genuine error sets `ERROR`. + +- **Server spans** (one per HTTP route) are created by the + `opentelemetry-instrumentation-fastapi` package. It stamps `http.*` attributes + and extracts inbound `traceparent` headers. This package does **not** create + or modify server spans — request routes never touch spans. +- **Gen-AI spans** (LLM calls, guardrails, internal service calls) are created + by this package from LiteLLM's logging callbacks. Request-level spans parent to + the server span via the captured anchor; DB/service spans parent to the active + span (ambient) so they nest under the request phase that triggered them. + +Both kinds share a single `TracerProvider`, so they belong to the same trace +and export through the same configured exporters. FastAPI middleware can only be +added before the app starts serving, so the app is instrumented at +import time **without** a provider — it binds to the OTel global +`ProxyTracerProvider`. Once config (and the callbacks) is loaded, the proxy +publishes the chosen logger's `TracerProvider` as the global via +`trace.set_tracer_provider(...)`, and the server spans delegate to it. When a +preset callback (`arize`, `langfuse_otel`, …) is configured, its provider +becomes the global, so server spans export to that backend too. + +## How a request flows + +1. **App creation** (`proxy_server` import): when the gate is on, + `mount.instrument_fastapi_app(app)` calls `FastAPIInstrumentor.instrument_app` + with no provider (the middleware stack is frozen once the app serves, so this + can't wait for startup). It binds to the OTel global `ProxyTracerProvider`. Noisy + non-LLM routes are excluded by default (`mount._DEFAULT_EXCLUDED_ROUTES`): health + checks (`/health*`), the Prometheus scrape (`/metrics`), and static UI/docs assets + (`/litellm-asset-prefix`, `/_next`, `/ui`, `/swagger`, `/docs`, `/redoc`, + `/openapi.json`, favicons, `/.well-known`) — so load-balancer polling, metric + scrapes, and asset fetches don't flood traces. Entries are substring-matched, so + `/metrics` also drops the `/model/metrics` admin-analytics spans. Set + `OTEL_PYTHON_FASTAPI_EXCLUDED_URLS` to override the whole set (e.g. `""` to trace + everything, or your own comma-separated path list). +2. **Startup** (`proxy_server.proxy_startup_event`): after the config (and + callbacks) is loaded, the already-registered preset `OpenTelemetryV2` logger + is reused — or a generic one reading `OTEL_*` envs is built when no preset is + configured — and its `TracerProvider` is published as the OTel global with + `trace.set_tracer_provider(...)`. The proxy tracer then delegates to it, so + server spans and gen-ai spans share one provider and the same trace. +3. **Request**: the FastAPI instrumentation starts the server span and makes it + the active context for the request task. The proxy's first call into the V2 + logger (`create_litellm_proxy_request_started_span`, at the auth boundary) + **captures it as the request anchor** (`set_request_root_span`), so every later + request-level span has a stable explicit parent regardless of what is active + when it emits. +4. **LLM call span (born at the boundary)**: `OpenTelemetryV2.log_pre_api_call` + runs synchronously in the request task, just before the upstream call, and + **opens** the LLM-call span there, parented to the anchored server span + (`resolve_request_span_context`). The open span is held in a bounded cache keyed + by `litellm_call_id` (a primitive the callback kwargs carry at both `pre_call` + and close), so no live `Span` ever travels through a `litellm_params` metadata + dict. For the boundary hook to fire at all, the logger is registered into + `litellm.input_callback` — the list `Logging.pre_call` iterates. The async + success/failure callback later + **closes** it: it builds an `LLMCallSpanData` from the typed + `standard_logging_object` (token usage and cost are computed only by then), + stamps the attributes, sets status, and ends the span. The sync callback is a + no-op (closing is async-only). When `pre_call` runs off the request task — a + sync-only provider driven through a thread pool, where contextvars (and so the + anchor) don't follow — no parent is visible there, so creation is **deferred** + to the async callback, whose worker context was copied from the request task at + enqueue and so still carries the anchor. **Pass-through** endpoints call + `logging_obj.pre_call` in the request task too, then close from a detached + `asyncio.create_task`; the anchor (not the by-then-inactive server span) keeps + their LLM-call span in the request's trace. `pre_call` is litellm's generic + "log the attempt" hook, so it also fires for synthetic proxy-gate error logs + (auth/rate-limit rejections); those carry `LITELLM_LOGGING_NO_UPSTREAM_LLM_CALL` + and are skipped, so a request rejected before reaching a provider never produces + a phantom CLIENT span. +5. **Guardrails / services**: the post-call and service hooks emit guardrail and + service spans the same way — typed data → engine → span. Service spans + (Redis/Postgres) are dispatched by `litellm/_service_logger.py`, which + recognizes the V2 `OpenTelemetryV2` logger (a plain `CustomLogger`, not a + subclass of the legacy `OpenTelemetry`). It hands every service call to the + logger — including calls with no parent span — and the V2 adapter decides the + role (`DB_CALL` vs `SERVICE`), the parent (ambient → threaded → root), and + whether the call is a traceable operation or a metrics-only ping. Guardrail + span data is built from the typed, provider-agnostic + `StandardLoggingGuardrailInformation` — no single provider's field shape is + assumed. +6. **Export**: each span ends and is handed to the provider's span processors, + which export to the configured backends (OTLP, console, in-memory, …). + +## Components + +### Sources of truth (`model/`, no OpenTelemetry import) + +These define the shape of a span without depending on the OTel SDK, so they can +be imported anywhere. They live in [`model/`](./model) and form a closed set — +nothing here imports outside it: + +- [`semconv.py`](./model/semconv.py) — attribute-key constants (`gen_ai.*`, `http.*`, + `litellm.*`), the GenAI operation/provider enums, and the functions that map + LiteLLM provider/call-type strings onto convention values. +- [`spans.py`](./model/spans.py) — the span registry: every span role, its OTel span + kind, its place in the hierarchy, and its name builder. +- [`payloads.py`](./model/payloads.py) — frozen dataclasses (`LLMCallSpanData`, + `GuardrailSpanData`, `ServiceSpanData`, …) built from heterogeneous logging + payloads via `from_*` classmethods. +- [`config.py`](./model/config.py) — `OpenTelemetryV2Config`, a pydantic-settings + model that reads `OTEL_*` / `LITELLM_OTEL_*` env vars, plus the feature gate. + `capture_span_content` gates whether prompt/response bodies may be written as + span attributes; it defaults **off** (`no_content`). The Baggage allowlists are + configurable, not hard-coded: set `LITELLM_OTEL_BAGGAGE_PROMOTED_KEYS` / + `LITELLM_OTEL_BAGGAGE_METADATA_KEYS` (comma-separated) as env vars, or + `baggage_promoted_keys` / `baggage_metadata_keys` (YAML lists) under + `callback_settings.otel` in `config.yaml` — the latter reach the config through + the logger's constructor kwargs. +- [`baggage.py`](./model/baggage.py) — the single definition of which request-identity + values are promoted into Baggage (so child spans inherit them) and under which + attribute keys. +- [`utils.py`](./model/utils.py) — value coercion, JSON serialization, and + extractor-table application, shared across the package. + +### Engine + +- [`emitter.py`](./emitter.py) — `SpanEmitter.emit(role, data)`: dedupe → start + the span → run the mapper chain to stamp attributes → set status → end. It + owns no attribute keys. The dedupe set (which coalesces the sync+async firing + of one request) is a bounded LRU so it can't grow without limit. +- [`mappers/`](./mappers) — each mapper turns typed span data into a flat + `{attribute key: value}` dict. They compose: listing several mapper names in + the config layers multiple attribute vocabularies onto the same span. + - `genai` — the canonical OpenTelemetry GenAI vocabulary, always present. + - `legacy` — an additional vocabulary using the older semconv-ai / Traceloop + attribute key names, for backends that read those. + - `openinference`, `langfuse`, `weave`, `langtrace` — vendor vocabularies. + - `resolve_mappers(names)` turns config names into mapper instances. + +### Plumbing (`plumbing/`) + +The OTel-SDK wiring. Everything here imports only `model/` and each other; it +lives in [`plumbing/`](./plumbing): + +- [`providers.py`](./plumbing/providers.py) — builds the `TracerProvider`, its exporters + (from `ExporterSpec`s), and the span processor that copies allowlisted Baggage + entries onto every span. `register_exporter_factory(kind, factory)` lets a + preset contribute a custom exporter `kind` (e.g. one that fetches an auth + token lazily) without coupling this module to any vendor. +- [`context.py`](./plumbing/context.py) — trace-context and Baggage read/write helpers. +- [`routing.py`](./plumbing/routing.py) — `TenantTracerCache`: when a request carries + team/key-scoped vendor credentials, route its spans through a credential-keyed + `TracerProvider` so one logger serves many tenants. The cache is a bounded LRU + that flushes + shuts down evicted providers, since the key derives from + request-supplied credentials and must not grow (or leak threads) without limit. +- [`metrics.py`](./plumbing/metrics.py) — GenAI client metric instruments. + +### Adapter + +- [`logger.py`](./logger.py) — `OpenTelemetryV2`, a `CustomLogger` that + translates LiteLLM's logging callbacks into typed span data and hands them to + the engine. The LLM-call span is opened at the `log_pre_api_call` boundary + (parented to the live server span via ambient context) and closed at the async + success/failure callback; the open span is held in a bounded cache keyed by + `litellm_call_id`, never threaded through a metadata dict. The logger registers + itself into `litellm.input_callback` so `Logging.pre_call` fires the boundary + hook. +- [`mount.py`](./mount.py) — `instrument_fastapi_app(app)`, the single call site + that attaches `opentelemetry-instrumentation-fastapi` for SERVER spans. It owns + the health-check exclusion default (`OTEL_PYTHON_FASTAPI_EXCLUDED_URLS`) and the + passthrough span-naming hook (`PASSTHROUGH_PREFIXES`) so `proxy_server` carries + no OTel detail. A safe no-op when the gate is off or the instrumentation package + is absent; must be called at app-creation time (the middleware stack freezes + once the app serves). + +### Presets + +- [`presets/`](./presets) — each preset reads one integration's env vars and + returns an `OpenTelemetryV2Config` (exporter destination + mapper vocabularies + + resource attributes). `PRESET_BY_CALLBACK` maps a callback name (`"arize"`, + `"langfuse_otel"`, …) to its preset. Integrations that support team/key-scoped + credentials also provide a per-request OTLP header builder + (`DYNAMIC_HEADERS_BY_CALLBACK`). Presets do **no** network I/O at build time: + AgentOps, for example, mints its JWT lazily inside a custom exporter on the + first export (in the `BatchSpanProcessor` worker thread), never on the event + loop. + +## Extending + +- **A new attribute vocabulary for a backend**: add a mapper in `mappers/` + (a class with a `map(data) -> AttributeMap` method, typically built from + `key -> extractor` tables) and register it in `mappers/__init__._MAPPER_BY_NAME`. +- **A new integration**: add a preset in `presets/` that returns an + `OpenTelemetryV2Config`, and register it in `presets/__init__.PRESET_BY_CALLBACK`. + If it supports dynamic credentials, add a header builder to + `DYNAMIC_HEADERS_BY_CALLBACK`. +- **A new span kind**: add a role to `spans.py` (registry entry + name builder), + a payload dataclass in `payloads.py`, and a branch in the relevant mapper(s). diff --git a/litellm/integrations/otel/__init__.py b/litellm/integrations/otel/__init__.py new file mode 100644 index 00000000000..42a84a85fbd --- /dev/null +++ b/litellm/integrations/otel/__init__.py @@ -0,0 +1,102 @@ +"""Typed, semconv-aligned OpenTelemetry instrumentation for LiteLLM. + +The three sources of truth — attribute keys (:mod:`semconv`), the span and +hierarchy registry (:mod:`spans`), and the typed span-data inputs +(:mod:`payloads`) — plus :mod:`config` are exported here and are free of any +``opentelemetry`` import. The engine layer (``emitter``, ``providers``, +``context``, ``metrics``) and the ``CustomLogger`` adapter (``logger``) are +reached via their submodule paths so that importing this package never +requires the OTel SDK. + +The ``LITELLM_OTEL_V2`` env var gates whether the factory in +``litellm_core_utils.litellm_logging`` constructs the ``OpenTelemetryV2`` +class (from :mod:`logger`). +""" + +from litellm.integrations.otel.model.config import ( + OTEL_V2_ENV, + OpenTelemetryV2Config, + is_otel_v2_enabled, +) +from litellm.integrations.otel.model.baggage import ( + BAGGAGE_PROMOTED_KEYS, + DEFAULT_BAGGAGE_METADATA_KEYS, + promoted_baggage, +) +from litellm.integrations.otel.model.metadata import ( + RequestContext, + RequestIdentity, +) +from litellm.integrations.otel.model.payloads import ( + GuardrailSpanData, + LLMCallSpanData, + LLMRequestParams, + LLMUsage, + ProxyRequestSpanData, + ServerInfo, + ServiceSpanData, + SpanError, +) +from litellm.integrations.otel.model.semconv import ( + DB, + Error, + GenAI, + GenAIOperation, + GenAIProvider, + HTTP, + LiteLLM, + Metric, + Server, + resolve_operation, + resolve_provider, +) +from litellm.integrations.otel.model.spans import ( + SPAN_REGISTRY, + LiteLLMSpanKind, + SpanRole, + SpanSpec, + db_system, + span_role_for_service, + validate_registry, +) + +__all__ = [ + # config + "OTEL_V2_ENV", + "OpenTelemetryV2Config", + "is_otel_v2_enabled", + # semconv + "BAGGAGE_PROMOTED_KEYS", + "DB", + "DEFAULT_BAGGAGE_METADATA_KEYS", + "Error", + "GenAI", + "GenAIOperation", + "GenAIProvider", + "HTTP", + "LiteLLM", + "Metric", + "Server", + "resolve_operation", + "resolve_provider", + # spans + "SPAN_REGISTRY", + "LiteLLMSpanKind", + "SpanRole", + "SpanSpec", + "db_system", + "span_role_for_service", + "validate_registry", + # payloads + "GuardrailSpanData", + "LLMCallSpanData", + "LLMRequestParams", + "LLMUsage", + "ProxyRequestSpanData", + "RequestContext", + "RequestIdentity", + "ServerInfo", + "ServiceSpanData", + "SpanError", + "promoted_baggage", +] diff --git a/litellm/integrations/otel/emitter.py b/litellm/integrations/otel/emitter.py new file mode 100644 index 00000000000..cae6514efdf --- /dev/null +++ b/litellm/integrations/otel/emitter.py @@ -0,0 +1,175 @@ +"""The span engine: dedup, start, run the mapper chain, set status, end.""" + +from collections import OrderedDict +from typing import Callable, Sequence + +from opentelemetry.context import Context +from opentelemetry.trace import Span, Tracer +from opentelemetry.trace.status import Status, StatusCode + +from litellm.integrations.otel.model.config import OpenTelemetryV2Config +from litellm.integrations.otel.mappers import resolve_mappers +from litellm.integrations.otel.mappers.base import AttributeMapper, SpanData +from litellm.integrations.otel.model.payloads import ( + GuardrailSpanData, + LLMCallSpanData, + ServiceSpanData, +) +from litellm.integrations.otel.plumbing.providers import to_otel_span_kind +from litellm.integrations.otel.model.semconv import Error +from litellm.integrations.otel.model.spans import ( + SPAN_REGISTRY, + SpanRole, + guardrail_span_name, + llm_call_span_name, + service_span_name, +) + +# Roles emit() knows how to name and emit. PROXY_REQUEST and the management +# routes are SERVER spans owned by the mounted FastAPI instrumentor, so they +# have no builder here. +_NAME_BUILDERS: dict[SpanRole, Callable[..., str]] = { + SpanRole.LLM_CALL: llm_call_span_name, + SpanRole.GUARDRAIL: guardrail_span_name, + # DB_CALL and SERVICE are both built from ServiceSpanData; they differ only in + # span kind (CLIENT vs INTERNAL) and attribute vocabulary, not in naming. + SpanRole.DB_CALL: service_span_name, + SpanRole.SERVICE: service_span_name, +} + +# Cap on the dedup cache. It only needs to coalesce the sync+async firing window +# of a single in-flight request, so a bounded LRU keeps memory flat on a +# long-running proxy while still covering every concurrently-open call. +_DEDUP_CACHE_MAX = 10_000 + + +class SpanEmitter: + def __init__( + self, + tracer: Tracer, + config: OpenTelemetryV2Config, + mappers: Sequence[AttributeMapper] | None = None, + ) -> None: + self._tracer = tracer + self._config = config + # The mapper chain is the sole source of span attributes. When not + # passed in, resolve it from the config so there's one source of truth. + self._mappers: list[AttributeMapper] = ( + list(mappers) + if mappers is not None + else resolve_mappers(config.mapper_names) + ) + # Bounded LRU (ordered by insertion / most-recent touch). Storing keys + # only — the value is unused — so it behaves like a capped set. + self._emitted: "OrderedDict[tuple[str, SpanRole], None]" = OrderedDict() + + # -- low-level helpers --------------------------------------------------- # + + def start_span( + self, + role: SpanRole, + name: str, + parent_context: Context | None = None, + start_time_ns: int | None = None, + *, + tracer: Tracer | None = None, + ) -> Span: + """Start a span for ``role`` without dedup or attribute mapping. + + For callers that own and manage their own span lifecycle. ``tracer`` + overrides the bound tracer for this span only, used for per-request + multi-tenant credential routing. + """ + return (tracer or self._tracer).start_span( + name, + context=parent_context, + kind=to_otel_span_kind(SPAN_REGISTRY[role].kind), + start_time=start_time_ns, + ) + + def _seen(self, dedup_key: str | None, role: SpanRole) -> bool: + """Return True once a ``(dedup_key, role)`` pair has been emitted. + + Guards against emitting the same span twice when a streaming call + fires both a sync and an async logging callback. + """ + if not dedup_key: + return False + marker = (dedup_key, role) + if marker in self._emitted: + self._emitted.move_to_end(marker) + return True + self._emitted[marker] = None + if len(self._emitted) > _DEDUP_CACHE_MAX: + self._emitted.popitem(last=False) # evict least-recently-used + return False + + # -- the engine ---------------------------------------------------------- # + + def emit( + self, + role: SpanRole, + data: SpanData, + parent_context: Context | None = None, + *, + start_time_ns: int | None = None, + end_time_ns: int | None = None, + tracer: Tracer | None = None, + ) -> Span | None: + """Emit one complete span: dedup, start, map attributes, status, end. + + Return the span, or ``None`` if it was deduplicated away. ``tracer`` + overrides the bound tracer for this span, used for per-request routing. + """ + # Only LLM-call spans carry a dedup key; LLM-call and service spans + # carry an ``error`` field. ``isinstance`` narrows the type for mypy and + # keeps the engine free of duck-typed attribute reads. + dedup_key = data.identity.call_id if isinstance(data, LLMCallSpanData) else None + if self._seen(dedup_key, role): + return None + span = self.start_span( + role, + _NAME_BUILDERS[role](data), + parent_context=parent_context, + start_time_ns=start_time_ns, + tracer=tracer, + ) + self.finish_span(role, span, data, end_time_ns=end_time_ns) + return span + + def finish_span( + self, + role: SpanRole, + span: Span, + data: SpanData, + *, + end_time_ns: int | None = None, + ) -> None: + """Stamp attributes + status on an already-started ``span`` and end it. + + The counterpart to :meth:`start_span` for callers that own a span's + lifecycle — the LLM-call span is opened at the request's ``pre_call`` + boundary (so it parents to the live server span via real ambient context, + never a span threaded through a metadata dict) and closed here once the + typed payload is available. The span name is (re)built from the now-known + data, since the boundary opener only has a provisional name. + """ + span.update_name(_NAME_BUILDERS[role](data)) + for mapper in self._mappers: + for key, value in mapper.map(data).items(): + span.set_attribute(key, value) + error = ( + data.error + if isinstance(data, (LLMCallSpanData, ServiceSpanData, GuardrailSpanData)) + else None + ) + if error and (error.error_type or error.message): + span.set_attribute(Error.TYPE, error.error_type or "error") + span.set_status( + Status(StatusCode.ERROR, error.message or error.error_type or "error") + ) + # On success leave the status UNSET (the semconv default) rather than + # forcing OK — that matches the FastAPI server span and avoids implying a + # span-level health signal litellm doesn't actually evaluate. Only a + # genuine error sets a status. + span.end(end_time=end_time_ns) diff --git a/litellm/integrations/otel/logger.py b/litellm/integrations/otel/logger.py new file mode 100644 index 00000000000..007b41df0a3 --- /dev/null +++ b/litellm/integrations/otel/logger.py @@ -0,0 +1,495 @@ +"""``CustomLogger`` adapter on the OpenTelemetry span engine.""" + +from collections import OrderedDict +from contextlib import contextmanager +from datetime import datetime +from typing import TYPE_CHECKING, Any, Iterator, Mapping, cast + +from opentelemetry.context import attach, get_current +from opentelemetry.sdk.trace import TracerProvider +from opentelemetry.trace import Span, Tracer, get_current_span, use_span + +import litellm +from litellm.integrations.custom_logger import CustomLogger +from litellm.integrations.otel.model.baggage import promoted_baggage +from litellm.integrations.otel.model.config import OpenTelemetryV2Config +from litellm.integrations.otel.plumbing.context import ( + is_recordable_span, + resolve_parent_context, + resolve_request_span_context, + set_request_baggage, + set_request_root_span, +) +from litellm.integrations.otel.emitter import SpanEmitter +from litellm.integrations.otel.mappers import resolve_mappers +from litellm.integrations.otel.model.metadata import ( + LLMCallEvent, + RequestIdentity, + guardrail_entries_from_request_data, + model_from_request_data, +) +from litellm.integrations.otel.model.payloads import ( + GuardrailSpanData, + LLMCallSpanData, + ServiceSpanData, + SpanError, +) +from litellm.integrations.otel.plumbing.providers import ( + build_tracer_provider, + get_tracer, +) +from litellm.integrations.otel.plumbing.routing import TenantTracerCache +from litellm.integrations.otel.model.spans import SpanRole, span_role_for_service +from litellm.integrations.otel.model.utils import to_ns + +if TYPE_CHECKING: + from litellm.types.utils import StandardLoggingGuardrailInformation + +LITELLM_TRACER_NAME = "litellm" + +# Any callback whose class belongs to one of these modules is "the OTel +# callback" for proxy-global-registration purposes. +_OTEL_MODULES = ( + "litellm.integrations.otel", + "litellm.integrations.opentelemetry", +) + + +# Cap on the open-call carrier map. A span opened at ``pre_call`` that never +# reaches a success/failure callback (e.g. a stream that only fires stream +# events) would otherwise linger; bounding the map evicts the oldest so memory +# stays flat on a long-running proxy while covering every concurrent in-flight +# call. +_OPEN_CALLS_MAX = 10_000 + + +class _LLMCallSpan: + """The state carried from the ``pre_call`` boundary to span close. + + ``span`` is the live span when it could be opened at the boundary (the server + span was ambient), or ``None`` when creation was deferred because no ambient + parent was visible — in which case the async callback creates it against its + own (worker-copied) ambient context using ``start_time_ns``. The presence of + a carrier for a call at all is the proof that ``pre_call`` ran, i.e. that an + upstream call was actually attempted. + """ + + __slots__ = ("span", "start_time_ns") + + def __init__(self, span: "Span | None", start_time_ns: int | None) -> None: + self.span = span + self.start_time_ns = start_time_ns + + +class OpenTelemetryV2(CustomLogger): + """The ``CustomLogger`` for OpenTelemetry.""" + + def __init__( + self, + config: OpenTelemetryV2Config | None = None, + callback_name: str | None = None, + tracer_provider: TracerProvider | None = None, + logger_provider: Any | None = None, # reserved for OTel logs + meter_provider: Any | None = None, # reserved for metrics + **kwargs: Any, + ) -> None: + super().__init__(**kwargs) + self.config: OpenTelemetryV2Config = config or OpenTelemetryV2Config(**kwargs) + self.callback_name = callback_name + self._tracer_provider: TracerProvider = ( + tracer_provider + if tracer_provider is not None + else build_tracer_provider(self.config) + ) + self.tracer: Tracer = get_tracer(self._tracer_provider, LITELLM_TRACER_NAME) + self._emitter = SpanEmitter( + self.tracer, self.config, mappers=resolve_mappers(self.config.mapper_names) + ) + self._tenant_tracers = TenantTracerCache( + self.config, callback_name, LITELLM_TRACER_NAME + ) + self._open_llm_calls: "OrderedDict[str, _LLMCallSpan]" = OrderedDict() + self._init_otel_logger_on_litellm_proxy() + + # ====================================================================== # + # Proxy global registration + # ====================================================================== # + + def _register_in_callback_list(self, callbacks: list) -> None: + already_otel = any( + cb.__class__.__module__.startswith(_OTEL_MODULES) + for cb in callbacks + if hasattr(cb, "__class__") + ) + if not already_otel: + callbacks.append(self) + + def _init_otel_logger_on_litellm_proxy(self) -> None: + try: + from litellm.proxy import proxy_server + except Exception: + return + try: + self._register_in_callback_list(litellm.service_callback) + self._register_in_callback_list(litellm.input_callback) + self._register_in_callback_list(litellm._async_success_callback) + self._register_in_callback_list(litellm._async_failure_callback) + except Exception: + pass + if getattr(proxy_server, "open_telemetry_logger", None) is None: + setattr(proxy_server, "open_telemetry_logger", self) + + # ====================================================================== # + # LLM-call callbacks — the span is opened at the ``pre_call`` boundary and + # closed here. See ``log_pre_api_call``. + # ====================================================================== # + + def log_pre_api_call(self, model, messages, kwargs): + """Open the LLM-call span at the call boundary. + + Runs synchronously inside the request task, before the upstream call — + the one place where the live server span is genuinely the ambient OTel + context — so the span parents to it natively, with no span threaded + through a metadata dict. The open span is stashed on the per-request + ``LiteLLMLoggingObj`` (a typed object) and closed in the async callback. + + When no recordable parent is visible (``pre_call`` was driven from a thread + pool for a sync-only provider, where contextvars — and so the anchor — + don't follow), creation is deferred: only the start time is recorded, and + the async callback — whose worker context was copied from the request task + and so still carries the anchor — creates the span then. + + Synthetic proxy-gate error logs (auth/rate-limit rejections) also fire this + hook but never made an upstream call; they are tagged and skipped so no + phantom LLM-call span is produced. + """ + call = LLMCallEvent.from_dict(kwargs) + if call.is_no_upstream_call: + return + call_id = call.call_id + if call_id is None: + return + # Idempotent: a retried call may re-enter ``pre_call`` with the same + # call id; keep the first span so its start time is the true one. + if call_id in self._open_llm_calls: + return + start_time_ns = to_ns(datetime.now()) + span: Span | None = None + # Parent to the request's anchored root span (stable across the request), + # falling back to ambient on the SDK path. Open the span live only when + # that resolves to a recordable parent; otherwise defer to the close + # callback (the thread-pool case, where the anchor isn't visible here). + parent_context = resolve_request_span_context() + if is_recordable_span(get_current_span(parent_context)): + span = self._emitter.start_span( + SpanRole.LLM_CALL, + call.provisional_span_name, + parent_context=parent_context, + start_time_ns=start_time_ns, + tracer=self._tenant_tracers.tracer_for( + self.tracer, call.dynamic_params + ), + ) + self._open_llm_calls[call_id] = _LLMCallSpan( + span=span, start_time_ns=start_time_ns + ) + # Evict the oldest open call if the map is over budget. A call that opens + # but never closes (a stream that only fires stream events) would linger + # otherwise; the evicted span is simply dropped (never exported). + if len(self._open_llm_calls) > _OPEN_CALLS_MAX: + self._open_llm_calls.popitem(last=False) + + async def async_log_success_event(self, kwargs, response_obj, start_time, end_time): + self._close_llm_call(kwargs, start_time, end_time) + + async def async_log_failure_event(self, kwargs, response_obj, start_time, end_time): + self._close_llm_call(kwargs, start_time, end_time) + + def _close_llm_call( + self, + kwargs: Mapping[str, Any], + start_time: datetime | float | None, + end_time: datetime | float | None, + ) -> Span | None: + """Finish the LLM-call span opened at ``pre_call`` (or create it deferred). + + No carrier for this call id means ``pre_call`` never ran — the request was + rejected at the gate or blocked by a pre-call guardrail before any upstream + call — so there is nothing to record and no phantom span. + """ + call = LLMCallEvent.from_dict(kwargs) + call_id = call.call_id + # ``pop`` is the dedup: this method runs from both the success and failure + # paths, and whichever fires first removes the carrier and closes the span. + carrier = self._open_llm_calls.pop(call_id, None) if call_id else None + if carrier is None: + return None + payload = call.payload + if payload is None: + if carrier.span is not None: + # Opened at the boundary but the payload never materialized — end + # it (named provisionally) so it isn't leaked as an open span. + carrier.span.end(end_time=to_ns(end_time)) + return None + data = LLMCallSpanData.from_standard_logging_payload( + payload, capture_content=self.config.capture_span_content + ) + end_time_ns = to_ns(end_time) + if carrier.span is not None: + # Born at the boundary: stamp attributes from the typed payload, set + # status, and end it. Its parent (the server span) was captured at + # creation from real ambient context. + self._emitter.finish_span( + SpanRole.LLM_CALL, carrier.span, data, end_time_ns=end_time_ns + ) + return carrier.span + # Deferred: ``pre_call`` saw no recordable parent, so create the span now. + # The worker copied the request task's context, which carries the anchored + # root span — parent to it (ambient fallback on the SDK path). Seed identity + # Baggage so the span — and the SDK path, which has none — is labeled + # consistently. + parent_ctx = resolve_request_span_context() + bag = promoted_baggage( + data.identity, + data.request_model, + promoted_keys=tuple(self.config.baggage_promoted_keys), + metadata_keys=tuple(self.config.baggage_metadata_keys), + ) + if bag: + parent_ctx = set_request_baggage(bag, context=parent_ctx) + return self._emitter.emit( + SpanRole.LLM_CALL, + data, + parent_context=parent_ctx, + start_time_ns=carrier.start_time_ns, + end_time_ns=end_time_ns, + tracer=self._tenant_tracers.tracer_for(self.tracer, call.dynamic_params), + ) + + # ====================================================================== # + # Service hooks + # ====================================================================== # + + async def async_service_success_hook( + self, + payload: Any, + parent_otel_span: Span | None = None, + start_time: datetime | float | None = None, + end_time: datetime | float | None = None, + event_metadata: dict | None = None, + ) -> None: + self._emit_service( + payload, + parent_otel_span=parent_otel_span, + start_time=start_time, + end_time=end_time, + event_metadata=event_metadata, + error_override=None, + ) + + async def async_service_failure_hook( + self, + payload: Any, + error: str | None = "", + parent_otel_span: Span | None = None, + start_time: datetime | float | None = None, + end_time: datetime | float | None = None, + event_metadata: dict | None = None, + ) -> None: + self._emit_service( + payload, + parent_otel_span=parent_otel_span, + start_time=start_time, + end_time=end_time, + event_metadata=event_metadata, + error_override=error or "error", + ) + + def _emit_service( + self, + payload: Any, + *, + parent_otel_span: Span | None, + start_time: datetime | float | None, + end_time: datetime | float | None, + event_metadata: dict | None, + error_override: str | None, + ) -> Span | None: + data = ServiceSpanData.from_payload(payload, event_metadata=event_metadata) + # Decide whether this service call is a span at all, and of what kind. + # ``None`` means metrics-only (framework instrumentation that duplicates a + # gen-AI span — ``self``/``router``/``proxy_pre_call`` — or ``auth``, which + # gets a live phase span instead). Those still feed Prometheus/Datadog via + # their own hooks; they just never enter the trace. + role = span_role_for_service(data.service_name) + if role is None: + return None + # A metrics-only ping with neither timing nor a parent (in-memory queue + # gauges) is not a traceable operation; a span for it would be a + # zero-duration root with no context, so skip it. Real background work + # (budget/reset jobs, spend flush) passes start/end times and still emits + # as a root; anything with a parent emits regardless. + if ( + error_override is None + and start_time is None + and end_time is None + and parent_otel_span is None + ): + return None + if error_override is not None and data.error is None: + data = ServiceSpanData( + service_name=data.service_name, + call_type=data.call_type, + error=SpanError(message=error_override), + event_metadata=data.event_metadata, + ) + # Parent like every other span: ambient context first (so identity Baggage + # rides along and the call nests under whatever request phase is active — + # e.g. a DB lookup under the live ``auth`` span), falling back to the + # server span the proxy threaded as ``parent_otel_span``. A background + # service call has neither, so it starts its own root trace. + parent_context = resolve_parent_context(threaded=parent_otel_span) + return self._emitter.emit( + role, + data, + parent_context=parent_context, + start_time_ns=to_ns(start_time), + end_time_ns=to_ns(end_time), + ) + + # ====================================================================== # + # async_post_call_* hooks — emit guardrail spans. The server span's status + # / errors are the FastAPI instrumentor's job, so we don't touch it here. + # ====================================================================== # + + def seed_request_identity(self, user_api_key_dict: Any, model: Any = None) -> None: + """Attach request-identity Baggage to the current context + server span. + + Seeding identity into Baggage makes **every** span emitted afterwards for + this request — LLM call, guardrail, DB call — inherit it via + ``LiteLLMBaggageSpanProcessor``. Called once at the auth boundary (as soon + as the key resolves) so post-auth spans are labeled consistently; the + Baggage rides the request task's contextvar from there on. Auth-internal + DB lookups that run before the key is known stay unlabeled — identity + isn't determined yet, which is correct. + """ + try: + identity = RequestIdentity.from_user_api_key_auth(user_api_key_dict) + bag = promoted_baggage( + identity, + model, + promoted_keys=tuple(self.config.baggage_promoted_keys), + metadata_keys=tuple(self.config.baggage_metadata_keys), + ) + if bag: + # Attach (no detach): the contextvar is scoped to this request's + # asyncio task and is reclaimed when the task ends. + attach(set_request_baggage(bag, context=get_current())) + # The server span was started by the instrumentor before this ran, + # so the Baggage processor (which only fires at span start) won't + # backfill it — stamp identity on it directly. + server_span = get_current_span() + if is_recordable_span(server_span): + # Re-capture the anchor here too: this runs post-auth with the + # server span active and covers entrypoints that bypass + # ``create_litellm_proxy_request_started_span`` (e.g. the SDK + # path's ``async_pre_call_hook``). Idempotent. + set_request_root_span(server_span) + for key, value in bag.items(): + server_span.set_attribute(key, value) + except Exception: + pass + + @contextmanager + def start_phase_span(self, name: str) -> "Iterator[Span]": + span = self._emitter.start_span(SpanRole.SERVICE, name) + with use_span(span, end_on_exit=True): + yield span + + async def async_pre_call_hook( + self, + user_api_key_dict: Any, + cache: Any, + data: dict, + call_type: Any, + ) -> dict: + self.seed_request_identity( + user_api_key_dict, + model=model_from_request_data(data), + ) + return data + + async def async_post_call_success_hook( + self, + data: Mapping[str, Any], + user_api_key_dict: Any, + response: Any, + ) -> Any: + self._emit_guardrail_spans(data) + return response + + async def async_post_call_failure_hook( + self, + request_data: Mapping[str, Any], + original_exception: BaseException | None, + user_api_key_dict: Any, + traceback_str: str | None = None, + ) -> None: + self._emit_guardrail_spans(request_data) + + def _emit_guardrail_spans(self, request_data: Mapping[str, Any]) -> None: + # A guardrail is a sibling of the LLM call under the request's root span, + # so parent it to the explicit anchor — not the active span, which on the + # failure path can be the live ``auth`` phase span (post-call failure hooks + # run from inside it on an auth rejection). Emit with the guardrail's actual + # execution window so a pre_call guardrail is placed before the LLM call + # rather than at post-call emission time. + guardrails = guardrail_entries_from_request_data(request_data) + if not guardrails: + return + parent_ctx = resolve_request_span_context() + for entry in guardrails: + data = GuardrailSpanData.from_logging_entry( + cast("StandardLoggingGuardrailInformation", entry) + ) + self._emitter.emit( + SpanRole.GUARDRAIL, + data, + parent_context=parent_ctx, + start_time_ns=to_ns(data.start_time), + end_time_ns=to_ns(data.end_time), + ) + + def create_litellm_proxy_request_started_span( + self, start_time: datetime, headers: Mapping[str, str] | None + ) -> Span | None: + span = get_current_span() + if not is_recordable_span(span): + return None + set_request_root_span(span) + return span + + +def _registered_v2_logger() -> "OpenTelemetryV2 | None": + try: + from litellm.proxy import proxy_server + except Exception: + return None + logger = getattr(proxy_server, "open_telemetry_logger", None) + return logger if isinstance(logger, OpenTelemetryV2) else None + + +def seed_request_identity(user_api_key_dict: Any, model: Any = None) -> None: + logger = _registered_v2_logger() + if logger is not None: + logger.seed_request_identity(user_api_key_dict, model=model) + + +@contextmanager +def phase_span(name: str) -> "Iterator[Span | None]": + logger = _registered_v2_logger() + if logger is None: + yield None + return + with logger.start_phase_span(name) as span: + yield span diff --git a/litellm/integrations/otel/mappers/__init__.py b/litellm/integrations/otel/mappers/__init__.py new file mode 100644 index 00000000000..012e63f1bee --- /dev/null +++ b/litellm/integrations/otel/mappers/__init__.py @@ -0,0 +1,58 @@ +"""Attribute mappers: pure ``LLMCallSpanData -> {attribute key: value}`` functions. + +Composition over inheritance: vocabularies layer onto the same span. Listing +``["genai", "openinference"]`` in ``config.mapper_names`` makes every span +carry both the canonical ``gen_ai.*`` keys and the OpenInference (Arize + +Phoenix) keys. Add ``"langfuse"`` and it works for all three backends at once. +""" + +from typing import Callable, Iterable + +from litellm.integrations.otel.mappers.base import ( + AttributeMap, + AttributeMapper, + AttrValue, +) +from litellm.integrations.otel.mappers.genai import GenAIMapper +from litellm.integrations.otel.mappers.langfuse import LangfuseMapper +from litellm.integrations.otel.mappers.langtrace import LangtraceMapper +from litellm.integrations.otel.mappers.legacy import LegacyMapper +from litellm.integrations.otel.mappers.openinference import OpenInferenceMapper +from litellm.integrations.otel.mappers.weave import WeaveMapper + +# Registry keyed by ``config.mapper_names`` entries. +_MAPPER_BY_NAME: dict[str, Callable[[], AttributeMapper]] = { + "genai": GenAIMapper, + "legacy": LegacyMapper, + "openinference": OpenInferenceMapper, + "langfuse": LangfuseMapper, + "weave": WeaveMapper, + "langtrace": LangtraceMapper, +} + + +def resolve_mappers(names: Iterable[str]) -> list[AttributeMapper]: + """Resolve mapper names to instances. Unknown names raise ``ValueError``.""" + out: list[AttributeMapper] = [] + for name in names: + factory = _MAPPER_BY_NAME.get(name) + if factory is None: + raise ValueError( + f"unknown mapper name {name!r}; known: " f"{sorted(_MAPPER_BY_NAME)}" + ) + out.append(factory()) + return out + + +__all__ = [ + "AttributeMap", + "AttributeMapper", + "AttrValue", + "GenAIMapper", + "LangfuseMapper", + "LangtraceMapper", + "LegacyMapper", + "OpenInferenceMapper", + "WeaveMapper", + "resolve_mappers", +] diff --git a/litellm/integrations/otel/mappers/base.py b/litellm/integrations/otel/mappers/base.py new file mode 100644 index 00000000000..e8fb5af9797 --- /dev/null +++ b/litellm/integrations/otel/mappers/base.py @@ -0,0 +1,36 @@ +"""Mapper protocol and attribute value types.""" + +from typing import Sequence + +from typing_extensions import Protocol, runtime_checkable + +from litellm.integrations.otel.model.payloads import ( + GuardrailSpanData, + LLMCallSpanData, + ServiceSpanData, +) + +AttrScalar = str | bool | int | float +# Mirrors ``opentelemetry.util.types.AttributeValue`` (homogeneous sequences) +# without importing the SDK, so mappers stay OTel-free. +AttrValue = ( + AttrScalar | Sequence[str] | Sequence[bool] | Sequence[int] | Sequence[float] +) +AttributeMap = dict[str, AttrValue] + +# The closed set of span-data types the engine routes through the mapper chain. +# Server spans (PROXY_REQUEST + management routes) belong to the mounted FastAPI +# instrumentor, not the mapper chain. +SpanData = LLMCallSpanData | GuardrailSpanData | ServiceSpanData + + +@runtime_checkable +class AttributeMapper(Protocol): + """Maps a typed span input to a flat dict of OTel span attributes. + + One method per mapper, dispatched internally on the ``data`` type. The + engine calls this uniformly for every span kind — mappers that don't speak + a given type return ``{}``. This is why the engine contains no attribute keys. + """ + + def map(self, data: SpanData) -> AttributeMap: ... diff --git a/litellm/integrations/otel/mappers/genai.py b/litellm/integrations/otel/mappers/genai.py new file mode 100644 index 00000000000..57fa51ea1fb --- /dev/null +++ b/litellm/integrations/otel/mappers/genai.py @@ -0,0 +1,137 @@ +"""Canonical OpenTelemetry GenAI semantic-convention mapper (always active). + +Owns the attribute schema for every span kind the engine emits — LLM call, +guardrail, and service — so the engine itself never references attribute keys. + +Each span kind declares its schema as a flat ``attribute key -> extractor`` +table: one lambda per mapping operation, applied against the typed span data. +""" + +from typing import Callable + +from litellm.integrations.otel.mappers.base import AttributeMap, AttrValue, SpanData +from litellm.integrations.otel.mappers.utils import collect, drop_none +from litellm.integrations.otel.model.payloads import ( + GuardrailSpanData, + LLMCallSpanData, + ServiceSpanData, + ToolDefinition, +) +from litellm.integrations.otel.model.semconv import DB, Error, GenAI, LiteLLM, Server +from litellm.integrations.otel.model.spans import db_system + + +class GenAIMapper: + + _LLM_CALL_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = { + GenAI.OPERATION_NAME: lambda d: d.operation.value, + GenAI.PROVIDER_NAME: lambda d: d.provider or None, + GenAI.REQUEST_MODEL: lambda d: d.request_model or None, + GenAI.REQUEST_TEMPERATURE: lambda d: d.request_params.temperature, + GenAI.REQUEST_TOP_P: lambda d: d.request_params.top_p, + GenAI.REQUEST_TOP_K: lambda d: d.request_params.top_k, + GenAI.REQUEST_MAX_TOKENS: lambda d: d.request_params.max_tokens, + GenAI.REQUEST_FREQUENCY_PENALTY: lambda d: d.request_params.frequency_penalty, + GenAI.REQUEST_PRESENCE_PENALTY: lambda d: d.request_params.presence_penalty, + GenAI.REQUEST_STOP_SEQUENCES: lambda d: ( + list(d.request_params.stop_sequences) + if d.request_params.stop_sequences + else None + ), + GenAI.REQUEST_SEED: lambda d: d.request_params.seed, + GenAI.RESPONSE_MODEL: lambda d: d.response_model, + GenAI.RESPONSE_ID: lambda d: d.response_id, + GenAI.RESPONSE_FINISH_REASONS: lambda d: ( + list(d.finish_reasons) if d.finish_reasons else None + ), + GenAI.USAGE_INPUT_TOKENS: lambda d: d.usage.input_tokens, + GenAI.USAGE_OUTPUT_TOKENS: lambda d: d.usage.output_tokens, + Error.TYPE: lambda d: d.error.error_type if d.error else None, + Server.ADDRESS: lambda d: d.server.address if d.server else None, + Server.PORT: lambda d: d.server.port if d.server else None, + LiteLLM.CALL_ID: lambda d: d.identity.call_id or None, + # The provider/underlying model is only known once routing has picked a + # deployment, so it can't ride identity Baggage (seeded at auth, before + # routing) onto the boundary-born LLM span — stamp it directly here. + LiteLLM.PROVIDER_MODEL: lambda d: d.identity.provider_model or None, + f"{LiteLLM.COST_PREFIX}total": lambda d: d.response_cost, + LiteLLM.REQUEST_STREAMING: lambda d: d.is_streaming, + } + + _TOOL_ATTRS: dict[str, Callable[[ToolDefinition], AttrValue | None]] = { + "name": lambda t: t.name, + "description": lambda t: t.description or None, + "parameters": lambda t: t.parameters_json or None, + } + + _GUARDRAIL_ATTRS: dict[str, Callable[[GuardrailSpanData], AttrValue | None]] = { + LiteLLM.GUARDRAIL_NAME: lambda d: d.guardrail_name, + LiteLLM.GUARDRAIL_MODE: lambda d: d.mode, + LiteLLM.GUARDRAIL_STATUS: lambda d: d.status, + LiteLLM.GUARDRAIL_PROVIDER: lambda d: d.provider, + LiteLLM.GUARDRAIL_ACTION: lambda d: d.action, + LiteLLM.GUARDRAIL_RESPONSE: lambda d: d.response_json, + LiteLLM.GUARDRAIL_VIOLATION_CATEGORIES: lambda d: ( + list(d.violation_categories) if d.violation_categories else None + ), + LiteLLM.GUARDRAIL_CONFIDENCE_SCORE: lambda d: d.confidence_score, + LiteLLM.GUARDRAIL_RISK_SCORE: lambda d: d.risk_score, + LiteLLM.GUARDRAIL_MASKED_ENTITY_COUNT: lambda d: d.masked_entity_count, + LiteLLM.GUARDRAIL_DURATION: lambda d: d.duration, + LiteLLM.GUARDRAIL_ID: lambda d: d.guardrail_id, + LiteLLM.GUARDRAIL_POLICY_TEMPLATE: lambda d: d.policy_template, + LiteLLM.GUARDRAIL_DETECTION_METHOD: lambda d: d.detection_method, + } + + _SERVICE_ATTRS: dict[str, Callable[[ServiceSpanData], AttrValue | None]] = { + LiteLLM.SERVICE_NAME: lambda d: d.service_name, + LiteLLM.SERVICE_CALL_TYPE: lambda d: d.call_type, + } + + def map(self, data: SpanData) -> AttributeMap: + match data: + case LLMCallSpanData(): + return self._llm_call(data) + case GuardrailSpanData(): + return self._guardrail(data) + case ServiceSpanData(): + return self._service(data) + case _: + return {} + + @classmethod + def _llm_call(cls, data: LLMCallSpanData) -> AttributeMap: + attrs = collect(cls._LLM_CALL_ATTRS, data) + attrs.update( + drop_none( + { + f"gen_ai.tool.{idx}.{suffix}": extract(tool) + for idx, tool in enumerate(data.tools) + for suffix, extract in cls._TOOL_ATTRS.items() + } + ) + ) + return attrs + + @classmethod + def _guardrail(cls, data: GuardrailSpanData) -> AttributeMap: + return collect(cls._GUARDRAIL_ATTRS, data) + + @classmethod + def _service(cls, data: ServiceSpanData) -> AttributeMap: + attrs = collect(cls._SERVICE_ATTRS, data) + # An outbound datastore call (DB_CALL / CLIENT span) also carries db.* + # semconv. Internal services (router, budget jobs, …) have no db.system, + # so they get only the litellm.service.* keys above. + system = db_system(data.service_name) + if system is not None: + attrs[DB.SYSTEM_NAME] = system + if data.call_type: + attrs[DB.OPERATION_NAME] = data.call_type + attrs.update( + { + f"{LiteLLM.METADATA_PREFIX}{key}": value + for key, value in data.event_metadata.items() + } + ) + return attrs diff --git a/litellm/integrations/otel/mappers/langfuse.py b/litellm/integrations/otel/mappers/langfuse.py new file mode 100644 index 00000000000..14c9fd01d05 --- /dev/null +++ b/litellm/integrations/otel/mappers/langfuse.py @@ -0,0 +1,84 @@ +"""Langfuse OTLP attribute mapper. + +Langfuse ingests OTLP spans and reads from its own vendor namespace +(``langfuse.observation.*``, ``langfuse.trace.*``). Compose this mapper after +``GenAIMapper`` to send canonical + Langfuse-flavored spans simultaneously. + +Every attribute is declared as a ``key -> extractor`` table entry (one callable +per mapping operation): ``_LLM_CALL_ATTRS`` for scalars and ``_BLOB_ATTRS`` for +the JSON-serialized payloads. ``_llm_call`` just applies both tables. +""" + +import json +from typing import Callable + +from litellm.integrations.otel.mappers.base import AttributeMap, AttrValue, SpanData +from litellm.integrations.otel.mappers.utils import ( + collect, + json_if, + output_messages, + serialize_messages, +) +from litellm.integrations.otel.model.payloads import ( + LLMCallSpanData, + LLMRequestParams, + LLMUsage, +) + + +class LangfuseMapper: + + _LLM_CALL_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = { + "langfuse.observation.type": lambda d: "generation", + "langfuse.observation.model.name": lambda d: d.request_model or None, + "langfuse.observation.metadata.provider": lambda d: d.provider or None, + "langfuse.observation.id": lambda d: d.identity.call_id or None, + "langfuse.trace.metadata.team_id": lambda d: d.identity.team_id or None, + "langfuse.trace.metadata.team_alias": lambda d: d.identity.team_alias or None, + } + + # Sub-tables folded into their respective JSON blobs. + _MODEL_PARAMS: dict[str, Callable[[LLMRequestParams], AttrValue | None]] = { + "temperature": lambda rp: rp.temperature, + "top_p": lambda rp: rp.top_p, + "max_tokens": lambda rp: rp.max_tokens, + "frequency_penalty": lambda rp: rp.frequency_penalty, + "presence_penalty": lambda rp: rp.presence_penalty, + "seed": lambda rp: rp.seed, + } + _USAGE_FIELDS: dict[str, Callable[[LLMUsage], AttrValue | None]] = { + "input": lambda u: u.input_tokens, + "output": lambda u: u.output_tokens, + "total": lambda u: u.total_tokens, + } + + # JSON-payload attributes: each builder returns the serialized blob or None. + _BLOB_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = { + "langfuse.observation.model.parameters": lambda d: json_if( + collect(LangfuseMapper._MODEL_PARAMS, d.request_params) + ), + "langfuse.observation.input": lambda d: serialize_messages(d.messages_in), + "langfuse.observation.output": lambda d: serialize_messages(output_messages(d)), + "langfuse.observation.usage_details": lambda d: json_if( + collect(LangfuseMapper._USAGE_FIELDS, d.usage) + ), + "langfuse.observation.cost_details": lambda d: ( + json.dumps({"total": d.response_cost}) + if d.response_cost is not None + else None + ), + } + + def map(self, data: SpanData) -> AttributeMap: + match data: + case LLMCallSpanData(): + return self._llm_call(data) + case _: + return {} + + @classmethod + def _llm_call(cls, data: LLMCallSpanData) -> AttributeMap: + return { + **collect(cls._LLM_CALL_ATTRS, data), + **collect(cls._BLOB_ATTRS, data), + } diff --git a/litellm/integrations/otel/mappers/langtrace.py b/litellm/integrations/otel/mappers/langtrace.py new file mode 100644 index 00000000000..7c0f30e57dd --- /dev/null +++ b/litellm/integrations/otel/mappers/langtrace.py @@ -0,0 +1,64 @@ +"""Langtrace attribute mapper. + +Produces Langtrace's attribute vocabulary so a span can be ingested by a +Langtrace backend. Compose it alongside other mappers like any other +vocabulary. + +Scalar attributes are declared as a flat ``key -> extractor`` table (one lambda +per mapping operation); the prompt/completion blobs are serialized as a tail. +""" + +from typing import Callable + +from litellm.integrations.otel.mappers.base import AttributeMap, AttrValue, SpanData +from litellm.integrations.otel.mappers.utils import ( + collect, + json_or_none, + output_messages, +) +from litellm.integrations.otel.model.payloads import LLMCallSpanData + + +class LangtraceMapper: + + _LLM_CALL_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = { + "gen_ai.operation.name": lambda d: "chat", + "langtrace.service.name": lambda d: d.provider or None, + "llm.model": lambda d: d.request_model or None, + "gen_ai.response.model": lambda d: d.response_model or None, + "gen_ai.response_id": lambda d: d.response_id or None, + "gen_ai.system_fingerprint": lambda d: d.system_fingerprint or None, + "llm.temperature": lambda d: d.request_params.temperature, + "llm.top_p": lambda d: d.request_params.top_p, + "llm.top_k": lambda d: d.request_params.top_k, + "llm.max_tokens": lambda d: d.request_params.max_tokens, + "llm.frequency_penalty": lambda d: d.request_params.frequency_penalty, + "llm.presence_penalty": lambda d: d.request_params.presence_penalty, + "llm.stream": lambda d: d.is_streaming, + "llm.token.counts.prompt": lambda d: d.usage.input_tokens, + "llm.token.counts.completion": lambda d: d.usage.output_tokens, + "llm.token.counts.total": lambda d: d.usage.total_tokens, + } + + _BLOB_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = { + "llm.prompts": lambda d: ( + json_or_none(list(d.messages_in)) if d.messages_in else None + ), + "llm.completions": lambda d: ( + json_or_none(output_messages(d)) if d.choices_out else None + ), + } + + def map(self, data: SpanData) -> AttributeMap: + match data: + case LLMCallSpanData(): + return self._llm_call(data) + case _: + return {} + + @classmethod + def _llm_call(cls, data: LLMCallSpanData) -> AttributeMap: + return { + **collect(cls._LLM_CALL_ATTRS, data), + **collect(cls._BLOB_ATTRS, data), + } diff --git a/litellm/integrations/otel/mappers/legacy.py b/litellm/integrations/otel/mappers/legacy.py new file mode 100644 index 00000000000..20ffe8b0dd8 --- /dev/null +++ b/litellm/integrations/otel/mappers/legacy.py @@ -0,0 +1,97 @@ +"""Mapper for the older semantic-convention attribute vocabulary. + +Emits attributes under the semconv-ai / Traceloop key names (e.g. +``gen_ai.system``, ``gen_ai.usage.prompt_tokens``, ``llm.is_streaming``) plus a +few bare, unprefixed service keys (``service``, ``call_type``, ``error``), for +backends that consume those names. + +Like ``GenAIMapper``, each span kind declares its schema as a flat +``attribute key -> extractor`` table: one lambda per mapping operation. +""" + +from typing import Callable, Final + +from litellm.integrations.otel.mappers.base import AttributeMap, AttrValue, SpanData +from litellm.integrations.otel.mappers.utils import collect, drop_none +from litellm.integrations.otel.model.payloads import ( + LLMCallSpanData, + ServiceSpanData, + ToolDefinition, +) + +# Attribute keys in the semconv-ai / Traceloop vocabulary. +_LEGACY_SYSTEM: Final = "gen_ai.system" +_LEGACY_PROMPT_TOKENS: Final = "gen_ai.usage.prompt_tokens" +_LEGACY_COMPLETION_TOKENS: Final = "gen_ai.usage.completion_tokens" +_LEGACY_TOTAL_TOKENS: Final = "gen_ai.usage.total_tokens" +_LEGACY_IS_STREAMING: Final = "llm.is_streaming" +_LEGACY_TOP_K: Final = "llm.top_k" +_LEGACY_FREQUENCY_PENALTY: Final = "llm.frequency_penalty" +_LEGACY_PRESENCE_PENALTY: Final = "llm.presence_penalty" +_LEGACY_STOP_SEQUENCES: Final = "llm.chat.stop_sequences" +_LEGACY_SERVICE: Final = "service" +_LEGACY_CALL_TYPE: Final = "call_type" +_LEGACY_ERROR: Final = "error" + + +class LegacyMapper: + """Emits LLM-call and service attributes under the older key names.""" + + _LLM_CALL_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = { + _LEGACY_SYSTEM: lambda d: d.provider or None, + _LEGACY_PROMPT_TOKENS: lambda d: d.usage.input_tokens, + _LEGACY_COMPLETION_TOKENS: lambda d: d.usage.output_tokens, + _LEGACY_TOTAL_TOKENS: lambda d: d.usage.total_tokens, + _LEGACY_IS_STREAMING: lambda d: d.is_streaming, + _LEGACY_TOP_K: lambda d: d.request_params.top_k, + _LEGACY_FREQUENCY_PENALTY: lambda d: d.request_params.frequency_penalty, + _LEGACY_PRESENCE_PENALTY: lambda d: d.request_params.presence_penalty, + _LEGACY_STOP_SEQUENCES: lambda d: ( + list(d.request_params.stop_sequences) + if d.request_params.stop_sequences + else None + ), + } + + _TOOL_ATTRS: dict[str, Callable[[ToolDefinition], AttrValue | None]] = { + "name": lambda t: t.name, + "description": lambda t: t.description or None, + "parameters": lambda t: t.parameters_json or None, + } + + _SERVICE_ATTRS: dict[str, Callable[[ServiceSpanData], AttrValue | None]] = { + _LEGACY_SERVICE: lambda d: d.service_name, + _LEGACY_CALL_TYPE: lambda d: d.call_type, + _LEGACY_ERROR: lambda d: ( + d.error.message if d.error is not None and d.error.message else None + ), + } + + def map(self, data: SpanData) -> AttributeMap: + match data: + case LLMCallSpanData(): + return self._llm_call(data) + case ServiceSpanData(): + return self._service(data) + case _: + return {} + + @classmethod + def _llm_call(cls, data: LLMCallSpanData) -> AttributeMap: + attrs = collect(cls._LLM_CALL_ATTRS, data) + attrs.update( + drop_none( + { + f"llm.request.functions.{idx}.{suffix}": extract(tool) + for idx, tool in enumerate(data.tools) + for suffix, extract in cls._TOOL_ATTRS.items() + } + ) + ) + return attrs + + @classmethod + def _service(cls, data: ServiceSpanData) -> AttributeMap: + attrs = collect(cls._SERVICE_ATTRS, data) + attrs.update(dict(data.event_metadata)) + return attrs diff --git a/litellm/integrations/otel/mappers/openinference.py b/litellm/integrations/otel/mappers/openinference.py new file mode 100644 index 00000000000..d8195cbe03d --- /dev/null +++ b/litellm/integrations/otel/mappers/openinference.py @@ -0,0 +1,128 @@ +"""OpenInference attribute mapper (Arize + Arize-Phoenix shared vocabulary). + +Spec: https://github.com/Arize-ai/openinference/tree/main/spec — the standard +both Arize and Phoenix consume. Composing this mapper after ``GenAIMapper`` +gives the same span both vocabularies, so a single trace lights up Arize + +Phoenix + any other OpenInference-aware backend simultaneously. +""" + +import json +from typing import Callable, Sequence + +from litellm.integrations.otel.mappers.base import AttributeMap, AttrValue, SpanData +from litellm.integrations.otel.mappers.utils import ( + collect, + drop_none, + json_if, + message_content, + output_messages, +) +from litellm.integrations.otel.model.payloads import ( + LLMCallSpanData, + LLMRequestParams, + ToolDefinition, +) + + +class OpenInferenceMapper: + """Emits OpenInference attributes for LLM_CALL spans. + + Key families (per the OpenInference spec): + - ``openinference.span.kind`` — discriminator (``"LLM"`` here) + - ``llm.model_name`` / ``llm.provider`` / ``llm.invocation_parameters`` + - ``llm.input_messages.{i}.message.role`` / ``...content`` + - ``llm.output_messages.{i}.message.role`` / ``...content`` + - ``llm.token_count.prompt`` / ``...completion`` / ``...total`` + - ``input.value`` / ``output.value`` — JSON-serialized request / response + """ + + _LLM_CALL_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = { + "openinference.span.kind": lambda d: "LLM", + "llm.model_name": lambda d: d.request_model or None, + "llm.provider": lambda d: d.provider or None, + "llm.token_count.prompt": lambda d: d.usage.input_tokens, + "llm.token_count.completion": lambda d: d.usage.output_tokens, + "llm.token_count.total": lambda d: d.usage.total_tokens, + } + + # Folded into the ``llm.invocation_parameters`` JSON blob. + _INVOCATION_PARAMS: dict[str, Callable[[LLMRequestParams], AttrValue | None]] = { + "temperature": lambda rp: rp.temperature, + "top_p": lambda rp: rp.top_p, + "top_k": lambda rp: rp.top_k, + "max_tokens": lambda rp: rp.max_tokens, + "frequency_penalty": lambda rp: rp.frequency_penalty, + "presence_penalty": lambda rp: rp.presence_penalty, + "seed": lambda rp: rp.seed, + } + + # Per-tool extractors, keyed by the ``llm.tools.{idx}.*`` suffix. + _TOOL_ATTRS: dict[str, Callable[[ToolDefinition], AttrValue | None]] = { + "tool.name": lambda t: t.name, + "tool.description": lambda t: t.description or None, + "tool.json_schema": lambda t: t.parameters_json or None, + } + + # JSON-payload attributes: each builder returns the serialized blob or None. + _BLOB_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = { + "llm.invocation_parameters": lambda d: json_if( + collect(OpenInferenceMapper._INVOCATION_PARAMS, d.request_params) + ), + } + + def map(self, data: SpanData) -> AttributeMap: + match data: + case LLMCallSpanData(): + return self._llm_call(data) + case _: + return {} + + @classmethod + def _llm_call(cls, data: LLMCallSpanData) -> AttributeMap: + return { + **collect(cls._LLM_CALL_ATTRS, data), + **collect(cls._BLOB_ATTRS, data), + **cls._messages("llm.input_messages", "input.value", data.messages_in), + **cls._messages( + "llm.output_messages", "output.value", output_messages(data) + ), + **cls._tools(data), + } + + @staticmethod + def _messages( + prefix: str, value_key: str, messages: Sequence[object] + ) -> AttributeMap: + """Per-message ``{prefix}.{idx}.message.*`` keys + the ``value_key`` blob.""" + parsed = [ + (m.get("role") if isinstance(m, dict) else None, message_content(m)) + for m in messages + ] + attrs = drop_none( + { + key: value + for idx, (role, content) in enumerate(parsed) + for key, value in ( + ( + f"{prefix}.{idx}.message.role", + role if isinstance(role, str) else None, + ), + (f"{prefix}.{idx}.message.content", content), + ) + } + ) + if parsed: + attrs[value_key] = json.dumps( + [{"role": role, "content": content} for role, content in parsed] + ) + return attrs + + @classmethod + def _tools(cls, data: LLMCallSpanData) -> AttributeMap: + return drop_none( + { + f"llm.tools.{idx}.{suffix}": extract(tool) + for idx, tool in enumerate(data.tools) + for suffix, extract in cls._TOOL_ATTRS.items() + } + ) diff --git a/litellm/integrations/otel/mappers/utils.py b/litellm/integrations/otel/mappers/utils.py new file mode 100644 index 00000000000..6228fc8bbe7 --- /dev/null +++ b/litellm/integrations/otel/mappers/utils.py @@ -0,0 +1,76 @@ +"""Shared helpers for the attribute mappers. + +Small, mapper-agnostic utilities — JSON serialization, message extraction, and +extractor-table application — pulled out of the individual mapper modules so +they live in one place. +""" + +import json +from typing import Callable, Mapping, Sequence + +from litellm.integrations.otel.mappers.base import AttributeMap, AttrValue +from litellm.integrations.otel.model.payloads import LLMCallSpanData + + +def drop_none(values: Mapping[str, AttrValue | None]) -> AttributeMap: + """Return ``values`` with ``None``-valued entries removed.""" + return {k: v for k, v in values.items() if v is not None} + + +def collect(table: Mapping[str, Callable], source: object) -> AttributeMap: + """Apply an extractor table to ``source``, dropping ``None`` results.""" + return drop_none({key: extract(source) for key, extract in table.items()}) + + +def json_if(payload: Mapping[str, object]) -> str | None: + """JSON-serialize ``payload`` only when it's non-empty; else ``None``.""" + return json.dumps(payload) if payload else None + + +def json_or_none(value: object) -> str | None: + """JSON-serialize ``value`` (falling back to ``str``); ``None`` on failure.""" + try: + return json.dumps(value, default=str) + except Exception: + return None + + +def stringify_message(message: object) -> str | None: + """JSON-serialize a chat message dict; ``None`` if not a dict or on failure.""" + if not isinstance(message, dict): + return None + try: + return json.dumps(message, default=str) + except Exception: + return None + + +def serialize_messages(messages: Sequence[object]) -> str | None: + """Round-trip a sequence of message dicts through ``stringify_message``.""" + serialized = [ + json.loads(s) for s in (stringify_message(m) for m in messages) if s is not None + ] + return json.dumps(serialized) if serialized else None + + +def message_content(message: object) -> str | None: + """Extract the textual ``content`` from a chat message dict.""" + if not isinstance(message, dict): + return None + content = message.get("content") + if isinstance(content, str): + return content + if isinstance(content, list): + # multimodal: concatenate text parts only + parts = [ + part.get("text", "") + for part in content + if isinstance(part, dict) and part.get("type") == "text" + ] + return "".join(p for p in parts if isinstance(p, str)) or None + return None + + +def output_messages(data: LLMCallSpanData) -> list: + """The ``message`` payload of each response choice.""" + return [c.get("message") for c in data.choices_out if isinstance(c, dict)] diff --git a/litellm/integrations/otel/mappers/weave.py b/litellm/integrations/otel/mappers/weave.py new file mode 100644 index 00000000000..54b07299271 --- /dev/null +++ b/litellm/integrations/otel/mappers/weave.py @@ -0,0 +1,48 @@ +"""Weave (W&B) attribute mapper. + +Weave consumes OpenInference + a small set of Weave-specific keys (display +name, thread id, output value). This mapper layers the latter on top of +OpenInference's vocabulary — compose ``["genai", "openinference", "weave"]`` +to feed a Weave backend. +""" + +from typing import Callable + +from litellm.integrations.otel.mappers.base import AttributeMap, AttrValue, SpanData +from litellm.integrations.otel.mappers.utils import collect, json_or_none +from litellm.integrations.otel.model.payloads import LLMCallSpanData + + +class WeaveMapper: + """Maps ``LLMCallSpanData`` to Weave's vendor attributes.""" + + _LLM_CALL_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = { + # ``display_name`` has the form ``"{operation} {model}"``. The span + # name already covers that, but Weave reads this attribute too. + "weave.display_name": lambda d: ( + f"{d.operation.value} {d.request_model}" if d.request_model else None + ), + "weave.call_id": lambda d: d.identity.call_id or None, + } + + # JSON-payload attributes: each builder returns the serialized blob or None. + _BLOB_ATTRS: dict[str, Callable[[LLMCallSpanData], AttrValue | None]] = { + # Weave treats the response choices as the "output" payload. + "weave.output": lambda d: ( + json_or_none(list(d.choices_out)) if d.choices_out else None + ), + } + + def map(self, data: SpanData) -> AttributeMap: + match data: + case LLMCallSpanData(): + return self._llm_call(data) + case _: + return {} + + @classmethod + def _llm_call(cls, data: LLMCallSpanData) -> AttributeMap: + return { + **collect(cls._LLM_CALL_ATTRS, data), + **collect(cls._BLOB_ATTRS, data), + } diff --git a/litellm/integrations/otel/model/__init__.py b/litellm/integrations/otel/model/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/litellm/integrations/otel/model/baggage.py b/litellm/integrations/otel/model/baggage.py new file mode 100644 index 00000000000..67dd64e3914 --- /dev/null +++ b/litellm/integrations/otel/model/baggage.py @@ -0,0 +1,76 @@ +"""Baggage promotion: request-identity values carried across child spans. + +A bounded set of identity values is written into OpenTelemetry Baggage on the +LLM-call span so that child spans (guardrail, service) inherit them. +``providers.LiteLLMBaggageSpanProcessor`` reads Baggage at span start and stamps +the allowlisted keys onto every span. + +This module is the single place baggage is defined: ``_PROMOTABLE`` maps each +promotable attribute key to how its value is read, and the two ``*_KEYS`` +defaults select what is promoted unless the config overrides them. +""" + +from collections.abc import Callable +from typing import Final + +from litellm.integrations.otel.model.metadata import RequestIdentity +from litellm.integrations.otel.model.semconv import GenAI, LiteLLM + +# Attribute key -> value extractor over (identity, request_model). The single +# definition of what may be promoted and under which key. +_PROMOTABLE: Final[dict[str, Callable[[RequestIdentity, str | None], str | None]]] = { + LiteLLM.TEAM_ID: lambda identity, model: identity.team_id, + LiteLLM.TEAM_ALIAS: lambda identity, model: identity.team_alias, + LiteLLM.TEAM_METADATA: lambda identity, model: identity.team_metadata, + LiteLLM.KEY_HASH: lambda identity, model: identity.key_hash, + LiteLLM.END_USER: lambda identity, model: identity.end_user, + GenAI.REQUEST_MODEL: lambda identity, model: model, + LiteLLM.PROVIDER_MODEL: lambda identity, model: identity.provider_model, +} + +# Keys promoted by default (a subset of ``_PROMOTABLE``). ``END_USER`` is +# promotable but off by default — it identifies an individual user, so stamping +# it onto every span is opt-in via ``config.baggage_promoted_keys``. +BAGGAGE_PROMOTED_KEYS: Final[tuple[str, ...]] = ( + LiteLLM.TEAM_ID, + LiteLLM.TEAM_ALIAS, + LiteLLM.TEAM_METADATA, + LiteLLM.KEY_HASH, + GenAI.REQUEST_MODEL, + LiteLLM.PROVIDER_MODEL, +) + +# Metadata sub-keys eligible for promotion under the ``litellm.metadata.*`` +# namespace. The full metadata blob is never promoted; only this allowlist is. +DEFAULT_BAGGAGE_METADATA_KEYS: Final[tuple[str, ...]] = ( + "user_api_key_org_id", + "user_api_key_user_id", + "user_api_key_alias", + "user_api_key_end_user_id", + "requester_ip_address", +) + + +def promoted_baggage( + identity: RequestIdentity, + request_model: str | None, + promoted_keys: tuple[str, ...], + metadata_keys: tuple[str, ...] = DEFAULT_BAGGAGE_METADATA_KEYS, +) -> dict[str, str]: + """Identity values to write into Baggage, filtered to ``promoted_keys``. + + ``promoted_keys`` selects from ``_PROMOTABLE``; ``metadata_keys`` selects + sub-keys of ``identity.metadata`` to promote under ``litellm.metadata.*``. + Empty values are dropped. + """ + out: dict[str, str] = {} + for key, extract in _PROMOTABLE.items(): + if key in promoted_keys: + value = extract(identity, request_model) + if value: + out[key] = value + for meta_key in metadata_keys: + value = identity.metadata.get(meta_key) + if value: + out[f"{LiteLLM.METADATA_PREFIX}{meta_key}"] = value + return out diff --git a/litellm/integrations/otel/model/config.py b/litellm/integrations/otel/model/config.py new file mode 100644 index 00000000000..f78e1515ea1 --- /dev/null +++ b/litellm/integrations/otel/model/config.py @@ -0,0 +1,236 @@ +"""Typed configuration for the OpenTelemetry instrumentation.""" + +from typing import Any, List + +from pydantic import AliasChoices, BaseModel, Field, field_validator, model_validator +from pydantic_settings import BaseSettings, NoDecode, SettingsConfigDict +from typing_extensions import Annotated + +from litellm.integrations.otel.model.baggage import ( + BAGGAGE_PROMOTED_KEYS, + DEFAULT_BAGGAGE_METADATA_KEYS, +) + +#: Master feature-flag env var. The logger is inert until this is truthy. +OTEL_V2_ENV = "LITELLM_OTEL_V2" + + +class CaptureMessageContent(str): + NO_CONTENT = "no_content" + SPAN_ONLY = "span_only" + EVENT_ONLY = "event_only" + SPAN_AND_EVENT = "span_and_event" + + +class _OTelV2Flag(BaseSettings): + model_config = SettingsConfigDict(extra="ignore") + + enabled: bool = Field(default=False, validation_alias=AliasChoices(OTEL_V2_ENV)) + + +def is_otel_v2_enabled() -> bool: + return _OTelV2Flag().enabled + + +class ExporterSpec(BaseModel): + """One span-export destination. + + The shared ``TracerProvider`` attaches one ``SpanProcessor`` per spec, so + listing several specs sends every span to all of them at once (e.g. Arize + + Phoenix + your own Honeycomb). + """ + + model_config = {"extra": "forbid"} + + kind: str = Field( + default="console", + description="console | in_memory | otlp_http | otlp_grpc | ", + ) + endpoint: str | None = None + headers: str | None = None + options: dict[str, str] | None = Field( + default=None, + description=( + "Factory-specific configuration for a custom exporter ``kind`` " + "registered via ``providers.register_exporter_factory`` (e.g. an " + "API key a lazy-auth exporter fetches a token with). Ignored by the " + "built-in console/in_memory/otlp exporters." + ), + ) + use_simple_processor: bool | None = Field( + default=None, + description=( + "Force SimpleSpanProcessor regardless of exporter kind. Default: " + "auto (Simple for console/in_memory, Batch otherwise)." + ), + ) + + +class OpenTelemetryV2Config(BaseSettings): + model_config = SettingsConfigDict(populate_by_name=True, extra="ignore") + + # ----- single-destination shorthand, read from standard OTEL_* envs ----- # + exporter: str = Field( + default="console", + validation_alias=AliasChoices("OTEL_EXPORTER", "OTEL_EXPORTER_OTLP_PROTOCOL"), + description=( + "Exporter kind for the single-destination shorthand. The model " + "validator folds this (with ``endpoint`` / ``headers``) into a " + "one-entry ``exporters`` list when ``exporters`` is empty; set " + "``exporters`` directly for multiple destinations." + ), + ) + endpoint: str | None = Field( + default=None, + validation_alias=AliasChoices("OTEL_ENDPOINT", "OTEL_EXPORTER_OTLP_ENDPOINT"), + ) + headers: str | None = Field( + default=None, + validation_alias=AliasChoices("OTEL_HEADERS", "OTEL_EXPORTER_OTLP_HEADERS"), + ) + service_name: str = Field( + default="litellm", validation_alias=AliasChoices("OTEL_SERVICE_NAME") + ) + deployment_environment: str | None = Field( + default=None, validation_alias=AliasChoices("OTEL_ENVIRONMENT_NAME") + ) + + enable_metrics: bool = Field( + default=False, + validation_alias=AliasChoices("LITELLM_OTEL_INTEGRATION_ENABLE_METRICS"), + ) + enable_events: bool = Field( + default=False, + validation_alias=AliasChoices("LITELLM_OTEL_INTEGRATION_ENABLE_EVENTS"), + ) + capture_message_content: str = Field( + default=CaptureMessageContent.NO_CONTENT, + validation_alias=AliasChoices( + "OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT" + ), + ) + legacy_compat: bool = Field( + default=True, validation_alias=AliasChoices("LITELLM_OTEL_LEGACY_COMPAT") + ) + + # ----- explicit multi-destination / vocabulary configuration ------------ # + + exporters: list[ExporterSpec] = Field( + default_factory=list, + description=( + "One destination per spec. The shared TracerProvider attaches a " + "SpanProcessor per entry. When empty, the model validator folds " + "the ``exporter`` / ``endpoint`` / ``headers`` shorthand into a " + "single spec so there is always at least one destination." + ), + ) + + mapper_names: Annotated[List[str], NoDecode] = Field( + default_factory=lambda: ["genai"], + description=( + "Ordered attribute vocabularies to emit. ``genai`` is the " + "canonical OTel GenAI vocabulary and is always placed first. " + "Vendor names: ``openinference`` (Arize + Phoenix), ``langfuse``, " + "``weave``, ``langtrace``." + ), + ) + + resource_attributes: dict[str, str] = Field( + default_factory=dict, + description=( + "Extra Resource attributes beyond ``service.name`` and " + "``deployment.environment`` (e.g. integration-specific markers)." + ), + ) + + baggage_promoted_keys: Annotated[List[str], NoDecode] = Field( + default_factory=lambda: list(BAGGAGE_PROMOTED_KEYS), + validation_alias=AliasChoices( + "baggage_promoted_keys", "LITELLM_OTEL_BAGGAGE_PROMOTED_KEYS" + ), + description=( + "Identity attribute keys written into Baggage and stamped on every " + "child span (e.g. ``litellm.team.id``). Configure via the " + "``LITELLM_OTEL_BAGGAGE_PROMOTED_KEYS`` env var (comma-separated) or " + "``callback_settings.otel.baggage_promoted_keys`` in config.yaml (a " + "YAML list)." + ), + ) + baggage_metadata_keys: Annotated[List[str], NoDecode] = Field( + default_factory=lambda: list(DEFAULT_BAGGAGE_METADATA_KEYS), + validation_alias=AliasChoices( + "baggage_metadata_keys", "LITELLM_OTEL_BAGGAGE_METADATA_KEYS" + ), + description=( + "Metadata sub-keys promoted under the ``litellm.metadata.*`` " + "namespace. Configure via the ``LITELLM_OTEL_BAGGAGE_METADATA_KEYS`` " + "env var (comma-separated) or " + "``callback_settings.otel.baggage_metadata_keys`` in config.yaml." + ), + ) + + @field_validator( + "baggage_promoted_keys", + "baggage_metadata_keys", + "mapper_names", + mode="before", + ) + @classmethod + def _split_csv(cls, value: Any) -> Any: + """Accept a comma-separated string for list fields. + + Env vars are strings, but these fields are lists. Pydantic-settings would + otherwise require JSON for a list env var; splitting on commas here lets + an operator write ``LITELLM_OTEL_BAGGAGE_PROMOTED_KEYS=litellm.team.id,litellm.api_key.hash``. + YAML lists (from ``callback_settings.otel.*``) and real lists pass through + unchanged. + """ + if isinstance(value, str): + return [item.strip() for item in value.split(",") if item.strip()] + return value + + @model_validator(mode="after") + def _normalize(self) -> "OpenTelemetryV2Config": + # An endpoint with the default exporter kind implies OTLP/HTTP. + if self.endpoint and self.exporter == "console": + self.exporter = "otlp_http" + # When no explicit destinations are given, fold the single-destination + # shorthand into one spec so the provider always has a destination. + if not self.exporters: + self.exporters = [ + ExporterSpec( + kind=self.exporter, + endpoint=self.endpoint, + headers=self.headers, + ) + ] + # Ensure ``genai`` is always present and first. + names = list(self.mapper_names) + if "genai" in names: + names = ["genai"] + [n for n in names if n != "genai"] + else: + names = ["genai"] + names + # When enabled, also emit attribute keys under their semconv-ai / + # Traceloop names via the ``legacy`` mapper. Append it at the tail so + # the canonical ``genai`` keys win on any conflict. + if self.legacy_compat and "legacy" not in names: + names.append("legacy") + self.mapper_names = names + return self + + @property + def capture_span_content(self) -> bool: + """Whether prompt/response content may be stamped as span attributes. + + Defaults off (``no_content``): an operator must opt in before message + bodies leave the process, so a user request can never force its prompt + or completion into the configured backend while capture is disabled. + """ + return self.capture_message_content in ( + CaptureMessageContent.SPAN_ONLY, + CaptureMessageContent.SPAN_AND_EVENT, + ) + + @classmethod + def from_env(cls) -> "OpenTelemetryV2Config": + return cls() diff --git a/litellm/integrations/otel/model/metadata.py b/litellm/integrations/otel/model/metadata.py new file mode 100644 index 00000000000..f0ea0a608c6 --- /dev/null +++ b/litellm/integrations/otel/model/metadata.py @@ -0,0 +1,315 @@ +"""The single translation layer between a request's metadata and the spans. + +Every relevant field litellm exposes about a request — the user-facing model, +the model actually dispatched to the provider, the deployment, and the caller's +identity (team, key, end-user) — is parsed **once**, here, out of the +``StandardLoggingPayload`` (or a ``UserAPIKeyAuth`` at the auth boundary). Span +data, baggage promotion, and the mappers then read these typed fields instead of +each digging into the raw ``metadata`` / ``hidden_params`` dicts. + +Two models live here because a request's identity is known *before* its model +resolution is: + +* :class:`RequestIdentity` — team / key / end-user, seeded into Baggage at the + auth boundary (``from_user_api_key_auth``), before routing has picked a + deployment. ``provider_model`` is therefore absent from that early seed and is + only filled in from the payload once the call closes. +* :class:`RequestContext` — the full picture available at close: the resolved + request vs. provider model split, plus the response model, model group, model + id, and api base, wrapping the :class:`RequestIdentity`. + +The request-vs-provider model split is the subtle part. On the proxy a caller +asks for a *model group* (e.g. ``gpt-4o``) that routes to a concrete deployment +(e.g. ``azure/my-deployment``); the two are distinct and both worth recording. +``StandardLoggingPayload`` exposes them as: + +* ``model_group`` — the user-facing name the caller requested. +* ``model`` — already reconstructed (see ``reconstruct_model_name``) to the name + litellm dispatched to the provider (the deployment, provider-prefixed). +* ``hidden_params.litellm_model_name`` — a secondary source for the dispatched + model (populated only on some call paths, e.g. files). + +So ``gen_ai.request.model`` is the *group* (falling back to the call model on the +SDK path, which has no group), and ``litellm.provider.model`` is the *dispatched* +model. They coincide on the SDK path, which is correct. +""" + +from __future__ import annotations + +import json +from dataclasses import dataclass, field +from typing import TYPE_CHECKING, Any, Mapping, cast + +from litellm.constants import LITELLM_LOGGING_NO_UPSTREAM_LLM_CALL +from litellm.integrations.otel.model.semconv import resolve_operation +from litellm.integrations.otel.model.utils import as_str + +if TYPE_CHECKING: + from litellm.types.utils import StandardLoggingPayload + + +@dataclass(frozen=True) +class RequestIdentity: + call_id: str | None = None + team_id: str | None = None + team_alias: str | None = None + # The team's free-form metadata dict, JSON-serialized (empty/missing -> None). + team_metadata: str | None = None + key_hash: str | None = None + end_user: str | None = None + # The model litellm dispatched to the provider. Only known once the call + # completes (routing has picked a deployment), so it's absent from the + # auth-time seed and filled only from the payload. + provider_model: str | None = None + metadata: Mapping[str, str] = field(default_factory=dict) + + @classmethod + def from_payload(cls, payload: "StandardLoggingPayload") -> "RequestIdentity": + """Parse caller identity out of a closed request's payload metadata. + + ``provider_model`` is resolved here too (see :func:`resolve_provider_model`) + so the identity carried into Baggage labels every span with the dispatched + model, not just the user-facing one. + """ + raw_meta = cast(Mapping[str, object], payload.get("metadata") or {}) + metadata = { + key: str(value) + for key, value in raw_meta.items() + if isinstance(value, (str, bool, int, float)) + } + return cls( + call_id=as_str(payload.get("litellm_call_id")) or as_str(payload.get("id")), + # StandardLoggingMetadata's canonical key is ``user_api_key_team_id``; + # the bare ``team_id`` is a legacy alias and is often empty, so prefer + # the canonical key and fall back to the alias. + team_id=as_str(raw_meta.get("user_api_key_team_id")) + or as_str(raw_meta.get("team_id")), + team_alias=as_str(raw_meta.get("user_api_key_team_alias")) + or as_str(raw_meta.get("team_alias")), + team_metadata=_team_metadata_json( + raw_meta.get("user_api_key_team_metadata") + ), + key_hash=as_str(raw_meta.get("user_api_key_hash")), + end_user=as_str(payload.get("end_user")) + or as_str(raw_meta.get("user_api_key_end_user_id")), + provider_model=resolve_provider_model(payload), + metadata=metadata, + ) + + @classmethod + def from_user_api_key_auth(cls, auth: object) -> "RequestIdentity": + """Identity from a ``UserAPIKeyAuth`` (duck-typed to keep this module + free of a proxy import). + + Used in the pre-call hook to seed Baggage early — before any LLM, + guardrail, or service span is created — so the whole request's spans + inherit identity, not just the LLM-call span. Metadata sub-keys use the + ``user_api_key_*`` names that ``baggage.DEFAULT_BAGGAGE_METADATA_KEYS`` + promotes. + """ + get = lambda name: getattr(auth, name, None) # noqa: E731 + metadata = { + meta_key: str(value) + for meta_key, attr in ( + ("user_api_key_user_id", "user_id"), + ("user_api_key_org_id", "org_id"), + ("user_api_key_alias", "key_alias"), + ("user_api_key_end_user_id", "end_user_id"), + ) + if (value := get(attr)) + } + return cls( + team_id=as_str(get("team_id")), + team_alias=as_str(get("team_alias")), + team_metadata=_team_metadata_json(get("team_metadata")), + key_hash=as_str(get("api_key")), + end_user=as_str(get("end_user_id")), + # ``provider_model`` is unknown at the auth boundary — routing hasn't + # picked a deployment yet — so it's only populated from the payload. + metadata=metadata, + ) + + +@dataclass(frozen=True) +class RequestContext: + """The fully-resolved view of a closed request, parsed once from the payload. + + ``request_model`` is the user-facing requested model and ``provider_model`` + (on :attr:`identity`) is the model litellm dispatched to the provider; the two + differ on the proxy (group vs. deployment) and coincide on the SDK path. + """ + + request_model: str + response_model: str | None + model_group: str | None + model_id: str | None + api_base: str | None + identity: RequestIdentity + + @property + def provider_model(self) -> str | None: + """The dispatched-model name, carried on the identity for Baggage.""" + return self.identity.provider_model + + @classmethod + def from_standard_logging_payload( + cls, payload: "StandardLoggingPayload" + ) -> "RequestContext": + raw_meta = cast(Mapping[str, object], payload.get("metadata") or {}) + hidden = cast(Mapping[str, object], payload.get("hidden_params") or {}) + raw_response = payload.get("response") + response = cast( + Mapping[str, object], raw_response if isinstance(raw_response, dict) else {} + ) + model_group = as_str(payload.get("model_group")) or as_str( + raw_meta.get("model_group") + ) + return cls( + # The user asked for the group; fall back to the call model on the SDK + # path, which has no group. Empty string (never None) so the span name + # builder and the mapper see a plain string. + request_model=model_group or as_str(payload.get("model")) or "", + response_model=as_str(response.get("model")), + model_group=model_group, + model_id=as_str(payload.get("model_id")) + or _model_info_id(raw_meta.get("model_info")), + api_base=as_str(payload.get("api_base")) or as_str(hidden.get("api_base")), + identity=RequestIdentity.from_payload(payload), + ) + + +# --- live-callback kwargs parsing ------------------------------------------- # +# +# The model and helpers below parse the *live* callback ``kwargs`` god object (and +# the raw pre/post-call ``data`` dicts) — the untyped request state that reaches a +# ``CustomLogger`` before, or instead of, a ``StandardLoggingPayload``. They live +# here, with the payload/auth parsers, so every read out of a request's raw dicts +# is in one place rather than scattered across the ``CustomLogger``. + + +@dataclass(frozen=True) +class LLMCallEvent: + """The typed view of the live callback ``kwargs`` (``model_call_details``). + + litellm hands every callback an untyped ``kwargs`` god object. The fields the + OTel logger needs out of it are parsed **once**, here, so the ``CustomLogger`` + reads typed attributes instead of digging into the dict at each boundary. + """ + + # The ``litellm_call_id`` correlating ``pre_call`` with the close callback. + # Present in ``model_call_details`` at ``pre_call`` and in both the kwargs and + # the ``standard_logging_object`` at success/failure, so it's a stable key for + # the open-call carrier — no back-reference to the logging object required (the + # object isn't reachable from the callback kwargs at ``pre_call`` time). + call_id: str | None + # The ``StandardLoggingPayload`` carried on a success/failure callback; ``None`` + # at ``pre_call``, or when the call closed before any payload materialized (so + # there is nothing to stamp on the span). + payload: "StandardLoggingPayload | None" + # The ``standard_callback_dynamic_params`` routing the call to a per-tenant + # tracer (its own exporter/endpoint), or ``None`` when the call isn't scoped. + dynamic_params: Any + # True for synthetic proxy-gate logs (auth / rate-limit rejections): they fire + # the ``pre_call`` hook but never made an upstream call, so they get no span. + is_no_upstream_call: bool + # A best-effort ``"{operation} {model}"`` name known at ``pre_call`` time. The + # span is renamed from the typed payload at close (``finish_span``); this only + # needs to be reasonable for a span that never gets closed (a leak). + provisional_span_name: str + + @classmethod + def from_dict(cls, kwargs: Mapping[str, Any]) -> "LLMCallEvent": + raw_payload = kwargs.get("standard_logging_object") + payload = cast("StandardLoggingPayload", raw_payload) if raw_payload else None + operation = resolve_operation(as_str(kwargs.get("call_type"))) + model = as_str(kwargs.get("model")) or "" + return cls( + call_id=_call_id(payload, kwargs), + payload=payload, + dynamic_params=kwargs.get("standard_callback_dynamic_params"), + is_no_upstream_call=bool(kwargs.get(LITELLM_LOGGING_NO_UPSTREAM_LLM_CALL)), + provisional_span_name=f"{operation.value} {model}".strip(), + ) + + +def _call_id( + payload: "StandardLoggingPayload | None", kwargs: Mapping[str, Any] +) -> str | None: + """The call id from the payload (when closed) or the bare kwargs (at pre_call).""" + if payload is not None: + call_id = as_str(payload.get("litellm_call_id")) or as_str(payload.get("id")) + if call_id: + return call_id + return as_str(kwargs.get("litellm_call_id")) + + +def model_from_request_data(data: object) -> str | None: + """The user-facing ``model`` from a pre-call ``data`` dict (``None`` if absent). + + Read at the auth boundary to label early Baggage before routing has resolved + a deployment; ``data`` is duck-typed since it arrives untyped from the proxy. + """ + if isinstance(data, Mapping): + return as_str(data.get("model")) + return None + + +def guardrail_entries_from_request_data( + request_data: Mapping[str, Any], +) -> list[dict]: + """The guardrail-information dicts buried in ``metadata`` of a post-call dict. + + ``standard_logging_guardrail_information`` is stored as either a single dict + or a list of them; normalize to a list of dicts (dropping non-dict noise) so + the caller just iterates. Empty list when none are present. + """ + metadata = request_data.get("metadata") + if not isinstance(metadata, Mapping): + return [] + info = metadata.get("standard_logging_guardrail_information") + if isinstance(info, Mapping): + return [cast(dict, info)] + if isinstance(info, list): + return [entry for entry in info if isinstance(entry, dict)] + return [] + + +def resolve_provider_model(payload: "StandardLoggingPayload") -> str | None: + """The model litellm dispatched to the provider, from the payload. + + Prefers the explicit ``hidden_params.litellm_model_name`` (set on call paths + that know it, e.g. files), then the top-level ``model`` — which + ``reconstruct_model_name`` has already resolved to the deployment's + provider-prefixed name. Returns ``None`` only when neither is present. + """ + raw_meta = cast(Mapping[str, object], payload.get("metadata") or {}) + hidden = cast(Mapping[str, object], payload.get("hidden_params") or {}) + return ( + # ``deployment`` survives only on paths that don't strip it from metadata; + # harmless (and most precise) to prefer it when present. + as_str(raw_meta.get("deployment")) + or as_str(hidden.get("litellm_model_name")) + or as_str(payload.get("model")) + ) + + +def _model_info_id(model_info: object) -> str | None: + """The deployment id from a ``metadata.model_info`` sub-dict, if present.""" + if isinstance(model_info, Mapping): + return as_str(model_info.get("id")) + return None + + +def _team_metadata_json(value: object) -> str | None: + """JSON-serialize a team's metadata dict for a single Baggage value. + + Returns ``None`` for a missing, non-dict, or empty mapping so the empty case + is dropped rather than promoting a useless ``"{}"``. Keys are sorted for a + stable, diff-friendly serialization. + """ + if not isinstance(value, Mapping) or not value: + return None + try: + return json.dumps(value, default=str, sort_keys=True) + except Exception: + return None diff --git a/litellm/integrations/otel/model/payloads.py b/litellm/integrations/otel/model/payloads.py new file mode 100644 index 00000000000..65b50d0fc12 --- /dev/null +++ b/litellm/integrations/otel/model/payloads.py @@ -0,0 +1,468 @@ +"""Typed span-data inputs: frozen dataclasses the engine and mappers consume.""" + +from __future__ import annotations + +import json +from dataclasses import dataclass, field +from enum import Enum +from typing import TYPE_CHECKING, ClassVar, Mapping, cast +from urllib.parse import urlsplit + +from litellm.integrations.otel.model.metadata import ( + RequestContext, + RequestIdentity, +) +from litellm.integrations.otel.model.semconv import ( + GenAIOperation, + resolve_operation, + resolve_provider, +) +from litellm.integrations.otel.model.utils import ( + as_bool, + as_float, + as_int, + as_str, + as_str_tuple, +) + +# ``RequestIdentity`` and the request-metadata translation now live in +# :mod:`metadata`; re-exported here so existing ``model.payloads`` imports keep +# resolving it. +__all__ = [ + "RequestContext", + "RequestIdentity", + "GuardrailSpanData", + "LLMCallSpanData", + "LLMRequestParams", + "LLMUsage", + "ProxyRequestSpanData", + "ServerInfo", + "ServiceSpanData", + "SpanError", + "ToolDefinition", +] + +if TYPE_CHECKING: + from litellm.types.services import ServiceLoggerPayload + from litellm.types.utils import ( + StandardLoggingGuardrailInformation, + StandardLoggingPayload, + ) + + +# --- typed sub-structures ---------------------------------------------------- # + + +@dataclass(frozen=True) +class LLMRequestParams: + temperature: float | None = None + top_p: float | None = None + top_k: int | None = None + max_tokens: int | None = None + frequency_penalty: float | None = None + presence_penalty: float | None = None + stop_sequences: tuple[str, ...] | None = None + seed: int | None = None + + @classmethod + def from_model_parameters(cls, params: Mapping[str, object]) -> "LLMRequestParams": + max_tokens = as_int(params.get("max_tokens")) + if max_tokens is None: + max_tokens = as_int(params.get("max_completion_tokens")) + return cls( + temperature=as_float(params.get("temperature")), + top_p=as_float(params.get("top_p")), + top_k=as_int(params.get("top_k")), + max_tokens=max_tokens, + frequency_penalty=as_float(params.get("frequency_penalty")), + presence_penalty=as_float(params.get("presence_penalty")), + stop_sequences=as_str_tuple(params.get("stop")), + seed=as_int(params.get("seed")), + ) + + +@dataclass(frozen=True) +class LLMUsage: + input_tokens: int | None = None + output_tokens: int | None = None + total_tokens: int | None = None + + +@dataclass(frozen=True) +class SpanError: + error_type: str | None = None + message: str | None = None + + +@dataclass(frozen=True) +class ServerInfo: + address: str | None = None + port: int | None = None + + @classmethod + def from_api_base(cls, api_base: str | None) -> ServerInfo | None: + if not api_base: + return None + parsed = urlsplit(api_base if "://" in api_base else f"//{api_base}") + if not parsed.hostname: + return None + return cls(address=parsed.hostname, port=parsed.port) + + +@dataclass(frozen=True) +class GuardrailSpanData: + guardrail_name: str + mode: str | None = None + status: str | None = None + masked_entity_count: int | None = None + provider: str | None = None + action: str | None = None + # The guardrail verdict / provider response (e.g. the moderation result), + # JSON-serialized. This is the detail that belongs on the guardrail span. + response_json: str | None = None + violation_categories: tuple[str, ...] = () + confidence_score: float | None = None + risk_score: float | None = None + duration: float | None = None + # Actual execution window (epoch seconds) from the logging entry, so the span + # is placed when the guardrail really ran — a pre_call guardrail before the + # LLM call — rather than at post-call emission time. + start_time: float | None = None + end_time: float | None = None + # Provider-agnostic configuration/detection metadata (see + # ``StandardLoggingGuardrailInformation``). Present for any guardrail that + # populates them, not just one provider's shape. + guardrail_id: str | None = None + policy_template: str | None = None + detection_method: str | None = None + # Set when the guardrail intervened/blocked or failed, so the emitter marks + # the span ERROR — a blocking guardrail is an error outcome for that span. + error: SpanError | None = None + + # Guardrail statuses that mean the guardrail did not pass the request through. + _ERROR_STATUSES: ClassVar[frozenset[str]] = frozenset( + {"guardrail_intervened", "guardrail_failed_to_respond"} + ) + + @classmethod + def from_logging_entry( + cls, entry: "StandardLoggingGuardrailInformation" + ) -> "GuardrailSpanData": + """Build from one ``standard_logging_guardrail_information`` entry. + + Reads the canonical, provider-agnostic ``StandardLoggingGuardrailInformation`` + keys only — no guessing at a single provider's field names. Values that are + typed as enums or lists (e.g. ``guardrail_mode``) are normalized to a + stable string rather than assumed to already be plain strings. + """ + get = cast(Mapping[str, object], entry).get + status = as_str(get("guardrail_status")) + response = get("guardrail_response") + error = ( + SpanError(error_type=status, message=as_str(get("guardrail_action"))) + if status in cls._ERROR_STATUSES + else None + ) + return cls( + guardrail_name=as_str(get("guardrail_name")) or "guardrail", + mode=_guardrail_mode_str(get("guardrail_mode")), + status=status, + masked_entity_count=_total_masked_entities(get("masked_entity_count")), + provider=as_str(get("guardrail_provider")), + action=as_str(get("guardrail_action")), + response_json=_json_or_none(response) if response is not None else None, + violation_categories=as_str_tuple(get("violation_categories")) or (), + confidence_score=as_float(get("confidence_score")), + risk_score=as_float(get("risk_score")), + duration=as_float(get("duration")), + start_time=as_float(get("start_time")), + end_time=as_float(get("end_time")), + guardrail_id=as_str(get("guardrail_id")), + policy_template=as_str(get("policy_template")), + detection_method=as_str(get("detection_method")), + error=error, + ) + + +@dataclass(frozen=True) +class ServiceSpanData: + service_name: str + call_type: str | None = None + error: SpanError | None = None + # Caller-supplied attributes to stamp on the service span, passed through + # from ``async_service_*_hook(event_metadata=...)``. The mapper owns how + # these are namespaced: the canonical vocabulary uses ``litellm.metadata.*`` + # keys, the semconv-ai / Traceloop vocabulary uses the bare key names. + event_metadata: Mapping[str, str] = field(default_factory=dict) + + @classmethod + def from_payload( + cls, + payload: "ServiceLoggerPayload", + event_metadata: Mapping[str, object] | None = None, + ) -> "ServiceSpanData": + # ``payload.service`` is a ``ServiceTypes(str, Enum)`` and ``error`` is + # ``Optional[str]`` on the Pydantic model — no defensive reads needed. + # ``event_metadata`` is sanitized: the legacy service decorators pass raw + # call-site data (live objects, full request metadata, response headers), + # none of which belongs on a span. + return cls( + service_name=payload.service.value, + call_type=payload.call_type, + error=SpanError(message=payload.error) if payload.error else None, + event_metadata=sanitize_event_metadata(event_metadata), + ) + + +@dataclass(frozen=True) +class ProxyRequestSpanData: + http_method: str + route: str + url_path: str | None = None + status_code: int | None = None + identity: RequestIdentity | None = None + + +# --- the primary LLM-call model ---------------------------------------------- # + + +@dataclass(frozen=True) +class ToolDefinition: + """A single function/tool declared on a chat-completion request.""" + + name: str + description: str | None = None + parameters_json: str | None = ( + None # JSON-serialized schema (str so it's an AttrValue) + ) + + +@dataclass(frozen=True) +class LLMCallSpanData: + operation: GenAIOperation + provider: str + request_model: str + response_model: str | None + response_id: str | None + request_params: LLMRequestParams + usage: LLMUsage + finish_reasons: tuple[str, ...] + error: SpanError | None + response_cost: float | None + server: ServerInfo | None + identity: RequestIdentity + is_streaming: bool | None = None + tools: tuple[ToolDefinition, ...] = () + # Raw messages and response, needed by vendor mappers (OpenInference, + # Langfuse, Weave) that stamp message-level attributes. ``messages_in`` is + # the request payload; ``choices_out`` mirrors ``response.choices`` from + # the StandardLoggingPayload. Both are tuples of immutable mappings so the + # dataclass stays hashable and frozen. + messages_in: tuple[Mapping[str, object], ...] = () + choices_out: tuple[Mapping[str, object], ...] = () + system_fingerprint: str | None = None + + @classmethod + def from_standard_logging_payload( + cls, payload: "StandardLoggingPayload", capture_content: bool = False + ) -> "LLMCallSpanData": + params = cast(Mapping[str, object], payload.get("model_parameters") or {}) + # The single parse of the request's metadata — the request-vs-provider + # model split, the response model, api base, and identity all come from + # here rather than being re-derived from the raw payload dicts. + context = RequestContext.from_standard_logging_payload(payload) + # Normalize ``response`` to a dict once so the content/id reads below are a + # plain ``.get`` — no repeated ``isinstance`` guards. + raw_response = payload.get("response") + response = cast( + Mapping[str, object], raw_response if isinstance(raw_response, dict) else {} + ) + choices_out = _dicts(response.get("choices")) + # ``finish_reasons`` is metadata, not content, so derive it from + # ``choices_out`` before gating. The raw message/choice bodies are only + # retained when content capture is enabled (see ``capture_span_content``); + # otherwise the content-bearing mappers receive empty sequences and emit + # no prompt/response text. + finish_reasons = _finish_reasons(choices_out) + return cls( + operation=resolve_operation(as_str(payload.get("call_type"))), + provider=resolve_provider(as_str(payload.get("custom_llm_provider"))), + request_model=context.request_model, + response_model=context.response_model, + response_id=as_str(response.get("id")), + request_params=LLMRequestParams.from_model_parameters(params), + usage=LLMUsage( + input_tokens=as_int(payload.get("prompt_tokens")), + output_tokens=as_int(payload.get("completion_tokens")), + total_tokens=as_int(payload.get("total_tokens")), + ), + finish_reasons=finish_reasons, + error=_parse_error(payload), + response_cost=as_float(payload.get("response_cost")), + server=ServerInfo.from_api_base(context.api_base), + identity=context.identity, + is_streaming=as_bool(payload.get("stream")), + tools=_extract_tools(params), + messages_in=_dicts(payload.get("messages")) if capture_content else (), + choices_out=choices_out if capture_content else (), + system_fingerprint=as_str(response.get("system_fingerprint")), + ) + + +# --- service event_metadata sanitization ------------------------------------ # + +# Substrings (case-insensitive) of keys that must never reach a span: secrets, +# tokens, and raw request/response dumps the legacy service decorators pass. +_SENSITIVE_METADATA_SUBSTRINGS: tuple[str, ...] = ( + "api_key", + "token", + "secret", + "password", + "cookie", + "authorization", + "header", + "hidden_params", +) +# Keys that carry raw call-site internals — live objects, full kwargs/args. The +# operation name is already the span's ``call_type``, so ``function_name`` is +# redundant. +_DROP_METADATA_KEYS: frozenset = frozenset( + {"function_kwargs", "function_args", "function_name"} +) +_MAX_METADATA_VALUE_LEN = 1024 +_MAX_METADATA_ITEMS = 32 + + +def sanitize_event_metadata( + event_metadata: Mapping[str, object] | None, +) -> dict[str, str]: + """Reduce caller-supplied ``event_metadata`` to span-safe string attributes. + + Keeps only primitive values (str/int/float/bool) under non-sensitive keys — + never ``repr()``-ing objects, dicts, or lists, never stamping secrets/headers, + and bounding the count and per-value length. This is the single chokepoint: + both the GenAI and legacy mappers read the cleaned result. + """ + if not event_metadata: + return {} + clean: dict[str, str] = {} + for key, value in event_metadata.items(): + if len(clean) >= _MAX_METADATA_ITEMS: + break + if not isinstance(key, str) or key in _DROP_METADATA_KEYS: + continue + lowered = key.lower() + if any(token in lowered for token in _SENSITIVE_METADATA_SUBSTRINGS): + continue + # ``bool`` is a subclass of ``int``, so it's covered. Non-primitive values + # (objects, dicts, lists) are dropped rather than stringified. + if isinstance(value, (str, int, float)): + clean[key] = str(value)[:_MAX_METADATA_VALUE_LEN] + return clean + + +def _json_or_none(value: object) -> str | None: + """JSON-serialize ``value`` (already-string values pass through). ``None`` on failure.""" + if isinstance(value, str): + return value + try: + return json.dumps(value, default=str) + except Exception: + return None + + +def _guardrail_mode_str(value: object) -> str | None: + """Normalize ``guardrail_mode`` to a stable string. + + ``guardrail_mode`` is typed as a ``GuardrailEventHooks`` enum, a list of them, + or a ``GuardrailMode`` — not a plain string. Emit the enum *value* (e.g. + ``"pre_call"``) rather than ``str(enum)`` (``"GuardrailEventHooks.pre_call"``), + and join a list of modes so a guardrail that runs at multiple hooks is + represented faithfully. + """ + if value is None: + return None + if isinstance(value, (list, tuple)): + parts: list[str] = [] + for item in value: + if item is None: + continue + part = as_str(item.value) if isinstance(item, Enum) else as_str(item) + if part: + parts.append(part) + return ",".join(parts) or None + if isinstance(value, Enum): + return as_str(value.value) + return as_str(value) + + +def _total_masked_entities(value: object) -> int | None: + """``masked_entity_count`` is a ``{entity_type: count}`` map — sum to a total.""" + if isinstance(value, Mapping): + total = sum(v for v in value.values() if isinstance(v, int)) + return total or None + return as_int(value) + + +def _dicts(value: object) -> tuple[Mapping[str, object], ...]: + """The dict items of ``value`` (when it's a list), as a tuple. Else empty.""" + if not isinstance(value, list): + return () + return tuple(item for item in value if isinstance(item, dict)) + + +def _finish_reasons(choices: tuple[Mapping[str, object], ...]) -> tuple[str, ...]: + """Non-empty ``finish_reason`` of each response choice.""" + return tuple(r for c in choices if (r := as_str(c.get("finish_reason")))) + + +def _parse_error(payload: "StandardLoggingPayload") -> SpanError | None: + """A ``SpanError`` for a failed request, or ``None`` on success.""" + if payload.get("status") != "failure": + return None + info = cast(Mapping[str, object], payload.get("error_information") or {}) + return SpanError( + error_type=as_str(info.get("error_class")) or as_str(info.get("error_code")), + message=as_str(info.get("error_message")) or as_str(payload.get("error_str")), + ) + + +def _tool_from_entry(entry: object) -> ToolDefinition | None: + """One ``tools``/``functions`` entry → ``ToolDefinition``, or ``None`` if unusable.""" + if not isinstance(entry, dict): + return None + fn = entry.get("function") if "function" in entry else entry + if not isinstance(fn, dict): + return None + name = as_str(fn.get("name")) + if not name: + return None + params = fn.get("parameters") + parameters_json: str | None = None + if params is not None: + try: + parameters_json = json.dumps(params, default=str) + except Exception: + parameters_json = None + return ToolDefinition( + name=name, + description=as_str(fn.get("description")), + parameters_json=parameters_json, + ) + + +def _extract_tools( + model_parameters: Mapping[str, object], +) -> tuple[ToolDefinition, ...]: + """Pull declared tools from request params (OpenAI / Anthropic shape). + + Accepts the chat-completion ``tools=[{"type":"function", "function": + {...}}, ...]`` shape, and falls back to the ``functions=[...]`` shape. + Returns an empty tuple when neither is present. + """ + raw_tools = model_parameters.get("tools") + if not isinstance(raw_tools, list): + raw_tools = model_parameters.get("functions") # ``functions`` shape + if not isinstance(raw_tools, list): + return () + return tuple(t for entry in raw_tools if (t := _tool_from_entry(entry)) is not None) diff --git a/litellm/integrations/otel/model/semconv.py b/litellm/integrations/otel/model/semconv.py new file mode 100644 index 00000000000..1c6c30eda0d --- /dev/null +++ b/litellm/integrations/otel/model/semconv.py @@ -0,0 +1,201 @@ +""" +Keys follow the OpenTelemetry GenAI semantic conventions (experimental). Anything +without a semconv equivalent lives under the ``litellm.*`` vendor namespace. +""" + +from enum import Enum +from typing import Final + + +class GenAIOperation(str, Enum): + """Values for ``gen_ai.operation.name``.""" + + CHAT = "chat" + TEXT_COMPLETION = "text_completion" + EMBEDDINGS = "embeddings" + GENERATE_CONTENT = "generate_content" + CREATE_AGENT = "create_agent" # reserved for future agent spans + INVOKE_AGENT = "invoke_agent" # reserved for future agent spans + EXECUTE_TOOL = "execute_tool" # reserved for future tool spans + + +class GenAIProvider(str, Enum): + """Common values for the ``gen_ai.provider.name`` attribute.""" + + OPENAI = "openai" + ANTHROPIC = "anthropic" + AWS_BEDROCK = "aws.bedrock" + AZURE_AI_OPENAI = "azure.ai.openai" + AZURE_AI_INFERENCE = "azure.ai.inference" + GCP_GEMINI = "gcp.gemini" + GCP_VERTEX_AI = "gcp.vertex_ai" + COHERE = "cohere" + MISTRAL_AI = "mistral_ai" + DEEPSEEK = "deepseek" + GROQ = "groq" + PERPLEXITY = "perplexity" + X_AI = "x_ai" + IBM_WATSONX_AI = "ibm.watsonx.ai" + + +class GenAI: + """Canonical OTel GenAI span-attribute keys.""" + + # request + OPERATION_NAME: Final = "gen_ai.operation.name" + PROVIDER_NAME: Final = "gen_ai.provider.name" + REQUEST_MODEL: Final = "gen_ai.request.model" + REQUEST_TEMPERATURE: Final = "gen_ai.request.temperature" + REQUEST_TOP_P: Final = "gen_ai.request.top_p" + REQUEST_TOP_K: Final = "gen_ai.request.top_k" + REQUEST_MAX_TOKENS: Final = "gen_ai.request.max_tokens" + REQUEST_FREQUENCY_PENALTY: Final = "gen_ai.request.frequency_penalty" + REQUEST_PRESENCE_PENALTY: Final = "gen_ai.request.presence_penalty" + REQUEST_STOP_SEQUENCES: Final = "gen_ai.request.stop_sequences" + REQUEST_SEED: Final = "gen_ai.request.seed" + REQUEST_CHOICE_COUNT: Final = "gen_ai.request.choice.count" + REQUEST_ENCODING_FORMATS: Final = "gen_ai.request.encoding_formats" + # response + RESPONSE_ID: Final = "gen_ai.response.id" + RESPONSE_MODEL: Final = "gen_ai.response.model" + RESPONSE_FINISH_REASONS: Final = "gen_ai.response.finish_reasons" + # usage + USAGE_INPUT_TOKENS: Final = "gen_ai.usage.input_tokens" + USAGE_OUTPUT_TOKENS: Final = "gen_ai.usage.output_tokens" + # content (opt-in, gated by capture mode) + INPUT_MESSAGES: Final = "gen_ai.input.messages" + OUTPUT_MESSAGES: Final = "gen_ai.output.messages" + SYSTEM_INSTRUCTIONS: Final = "gen_ai.system_instructions" + OUTPUT_TYPE: Final = "gen_ai.output.type" + CONVERSATION_ID: Final = "gen_ai.conversation.id" + # agent / tool (reserved) + AGENT_ID: Final = "gen_ai.agent.id" + AGENT_NAME: Final = "gen_ai.agent.name" + TOOL_NAME: Final = "gen_ai.tool.name" + TOOL_CALL_ID: Final = "gen_ai.tool.call.id" + + +class Error: + TYPE: Final = "error.type" + + +class Server: + ADDRESS: Final = "server.address" + PORT: Final = "server.port" + + +class DB: + """Database / cache client-span keys (OTel ``db.*`` semconv). + + Stamped on ``DB_CALL`` spans (redis / postgres), which are CLIENT spans for + outbound datastore calls — not on the INTERNAL ``SERVICE`` spans. + """ + + SYSTEM_NAME: Final = "db.system.name" + OPERATION_NAME: Final = "db.operation.name" + + +class HTTP: + """HTTP server-span keys. Belong on the SERVER span only (never promoted).""" + + REQUEST_METHOD: Final = "http.request.method" + ROUTE: Final = "http.route" + RESPONSE_STATUS_CODE: Final = "http.response.status_code" + URL_PATH: Final = "url.path" + + +class LiteLLM: + """Vendor-extension keys (no semconv equivalent). Always ``litellm.*``.""" + + CALL_ID: Final = "litellm.call_id" + COST_PREFIX: Final = "litellm.cost." + METADATA_PREFIX: Final = "litellm.metadata." + TEAM_ID: Final = "litellm.team.id" + TEAM_ALIAS: Final = "litellm.team.alias" + # The team's free-form metadata dict, JSON-serialized into a single value. + TEAM_METADATA: Final = "litellm.team.metadata" + KEY_HASH: Final = "litellm.api_key.hash" + END_USER: Final = "litellm.end_user.id" + # The model string litellm actually sent to the provider (the deployment's + # ``litellm_params.model``), distinct from the user-facing ``gen_ai.request.model``. + PROVIDER_MODEL: Final = "litellm.provider.model" + REQUEST_STREAMING: Final = "litellm.request.streaming" + GUARDRAIL_NAME: Final = "litellm.guardrail.name" + GUARDRAIL_MODE: Final = "litellm.guardrail.mode" + GUARDRAIL_STATUS: Final = "litellm.guardrail.status" + GUARDRAIL_PROVIDER: Final = "litellm.guardrail.provider" + GUARDRAIL_ACTION: Final = "litellm.guardrail.action" + GUARDRAIL_RESPONSE: Final = "litellm.guardrail.response" + GUARDRAIL_VIOLATION_CATEGORIES: Final = "litellm.guardrail.violation_categories" + GUARDRAIL_CONFIDENCE_SCORE: Final = "litellm.guardrail.confidence_score" + GUARDRAIL_RISK_SCORE: Final = "litellm.guardrail.risk_score" + GUARDRAIL_MASKED_ENTITY_COUNT: Final = "litellm.guardrail.masked_entity_count" + GUARDRAIL_DURATION: Final = "litellm.guardrail.duration" + GUARDRAIL_ID: Final = "litellm.guardrail.id" + GUARDRAIL_POLICY_TEMPLATE: Final = "litellm.guardrail.policy_template" + GUARDRAIL_DETECTION_METHOD: Final = "litellm.guardrail.detection_method" + SERVICE_NAME: Final = "litellm.service.name" + SERVICE_CALL_TYPE: Final = "litellm.service.call_type" + PREPROCESSING_MS: Final = "litellm.preprocessing.duration_ms" + + +class Metric: + """GenAI metric instrument names.""" + + TOKEN_USAGE: Final = "gen_ai.client.token.usage" + OPERATION_DURATION: Final = "gen_ai.client.operation.duration" + + +# litellm ``custom_llm_provider`` -> ``gen_ai.provider.name`` value. +_PROVIDER_BY_LITELLM: dict[str, GenAIProvider] = { + "openai": GenAIProvider.OPENAI, + "text-completion-openai": GenAIProvider.OPENAI, + "azure": GenAIProvider.AZURE_AI_OPENAI, + "azure_ai": GenAIProvider.AZURE_AI_INFERENCE, + "anthropic": GenAIProvider.ANTHROPIC, + "bedrock": GenAIProvider.AWS_BEDROCK, + "bedrock_converse": GenAIProvider.AWS_BEDROCK, + "vertex_ai": GenAIProvider.GCP_VERTEX_AI, + "vertex_ai_beta": GenAIProvider.GCP_VERTEX_AI, + "gemini": GenAIProvider.GCP_GEMINI, + "cohere": GenAIProvider.COHERE, + "cohere_chat": GenAIProvider.COHERE, + "mistral": GenAIProvider.MISTRAL_AI, + "deepseek": GenAIProvider.DEEPSEEK, + "groq": GenAIProvider.GROQ, + "perplexity": GenAIProvider.PERPLEXITY, + "xai": GenAIProvider.X_AI, + "watsonx": GenAIProvider.IBM_WATSONX_AI, +} + +# litellm ``call_type`` -> ``gen_ai.operation.name``. +_OPERATION_BY_CALL_TYPE: dict[str, GenAIOperation] = { + "completion": GenAIOperation.CHAT, + "acompletion": GenAIOperation.CHAT, + "completion_with_retries": GenAIOperation.CHAT, + "text_completion": GenAIOperation.TEXT_COMPLETION, + "atext_completion": GenAIOperation.TEXT_COMPLETION, + "embedding": GenAIOperation.EMBEDDINGS, + "aembedding": GenAIOperation.EMBEDDINGS, + "responses": GenAIOperation.CHAT, + "aresponses": GenAIOperation.CHAT, +} + + +def resolve_provider(custom_llm_provider: str | None) -> str: + """Map a litellm provider string to a ``gen_ai.provider.name`` value. + + Unknown providers pass through verbatim — the convention explicitly allows + provider-specific values, so an unmapped name is still valid. + """ + if not custom_llm_provider: + return "" + mapped = _PROVIDER_BY_LITELLM.get(custom_llm_provider.lower()) + return mapped.value if mapped is not None else custom_llm_provider + + +def resolve_operation(call_type: str | None) -> GenAIOperation: + """Map a litellm ``call_type`` to a ``gen_ai.operation.name`` value.""" + if not call_type: + return GenAIOperation.CHAT + return _OPERATION_BY_CALL_TYPE.get(call_type.lower(), GenAIOperation.CHAT) diff --git a/litellm/integrations/otel/model/spans.py b/litellm/integrations/otel/model/spans.py new file mode 100644 index 00000000000..e4876f4ee58 --- /dev/null +++ b/litellm/integrations/otel/model/spans.py @@ -0,0 +1,203 @@ +""" +This module declares every span the instrumentation can emit and the hierarchy. + +Span-name patterns live here as typed builder functions. + +Canonical hierarchy:: + + PROXY_REQUEST (SERVER, root) # owned by the FastAPI instrumentor + ├── SERVICE (INTERNAL) # auth phase span (live; see logger.phase_span) + │ └── DB_CALL (CLIENT) # its key/user/team lookups nest here + ├── GUARDRAIL (INTERNAL) # request-lifecycle hook, sibling of LLM_CALL + ├── LLM_CALL (CLIENT) + └── DB_CALL (CLIENT) # e.g. the spend-log write + +Guardrails parent to PROXY_REQUEST, not LLM_CALL: pre/during/post-call guardrail +hooks are orchestrated by the request lifecycle (a pre-call guardrail runs +before the LLM call even starts), so a guardrail is a sibling of the LLM call, +not a child of it. The emitter parents every span to the ambient OTel context +(the active server span), which matches this. + +Not every service call becomes a span — :func:`span_role_for_service` decides: + +- ``DB_CALL`` (CLIENT) — outbound datastores (redis, postgres, + ``batch_write_to_db``), carrying ``db.*`` semconv. +- ``SERVICE`` (INTERNAL) — genuine internal work worth a span (background + budget/reset jobs, pod-lock manager). +- ``None`` (metrics-only) — framework instrumentation that duplicates a gen-AI + span (``self`` = the ``track_llm_api_timing`` wrapper, ``router``, + ``proxy_pre_call``) or ``auth`` (which gets a live phase span instead). These + still feed Prometheus/Datadog; they just never enter the trace. + +``DB_CALL`` and ``SERVICE`` are built from the same ``ServiceSpanData``; only the +role (hence span kind and attribute vocabulary) differs. A service call can fire +outside any request (a background job), in which case it parents to no server +span and starts its own root trace rather than being dropped. + +Management/admin endpoints are ordinary FastAPI routes — their SERVER spans are +owned by the instrumentor too, so they don't appear as a role here. +""" + +from dataclasses import dataclass +from enum import Enum +from typing import TYPE_CHECKING + +if TYPE_CHECKING: + from litellm.integrations.otel.model.payloads import ( + GuardrailSpanData, + LLMCallSpanData, + ProxyRequestSpanData, + ServiceSpanData, + ) + + +class SpanRole(str, Enum): + PROXY_REQUEST = "proxy_request" + LLM_CALL = "llm_call" + GUARDRAIL = "guardrail" + DB_CALL = "db_call" + SERVICE = "service" + + +class LiteLLMSpanKind(str, Enum): + SERVER = "server" + CLIENT = "client" + INTERNAL = "internal" + PRODUCER = "producer" + CONSUMER = "consumer" + + +@dataclass(frozen=True) +class SpanSpec: + role: SpanRole + kind: LiteLLMSpanKind + parent: SpanRole | None + + +SPAN_REGISTRY: dict[SpanRole, SpanSpec] = { + SpanRole.PROXY_REQUEST: SpanSpec( + SpanRole.PROXY_REQUEST, LiteLLMSpanKind.SERVER, parent=None + ), + SpanRole.LLM_CALL: SpanSpec( + SpanRole.LLM_CALL, LiteLLMSpanKind.CLIENT, parent=SpanRole.PROXY_REQUEST + ), + SpanRole.GUARDRAIL: SpanSpec( + SpanRole.GUARDRAIL, LiteLLMSpanKind.INTERNAL, parent=SpanRole.PROXY_REQUEST + ), + SpanRole.DB_CALL: SpanSpec( + SpanRole.DB_CALL, LiteLLMSpanKind.CLIENT, parent=SpanRole.PROXY_REQUEST + ), + SpanRole.SERVICE: SpanSpec( + SpanRole.SERVICE, LiteLLMSpanKind.INTERNAL, parent=SpanRole.PROXY_REQUEST + ), +} + + +# ``ServiceTypes`` value -> ``db.system.name``. These are outbound datastore +# calls and become CLIENT ``DB_CALL`` spans; ``redis_``-prefixed names cover the +# redis-backed spend queues. Any service not mapped here is litellm-internal work +# and stays an INTERNAL ``SERVICE`` span. This table is the single source of +# datastore knowledge — both the role classifier and the mapper read it. +_DB_SYSTEM_BY_SERVICE: dict[str, str] = { + "redis": "redis", + "postgres": "postgresql", + "batch_write_to_db": "postgresql", +} + + +def db_system(service_name: str) -> str | None: + """The ``db.system.name`` for a datastore service, else ``None``. + + ``None`` means the service is not an outbound datastore call. Redis-backed + spend queues (``redis_*``) map to ``redis``. + """ + if service_name in _DB_SYSTEM_BY_SERVICE: + return _DB_SYSTEM_BY_SERVICE[service_name] + if service_name.startswith("redis_"): + return "redis" + return None + + +# ``ServiceTypes`` values that are NOT emitted as spans — they are framework +# instrumentation that either duplicates a gen-AI span or has a better home as a +# Prometheus/Datadog metric. They still flow to those metric backends via their +# own hooks; the v2 logger just does not put them in the trace: +# +# - ``self`` — ``track_llm_api_timing`` wraps the LLM call; the +# ``chat {model}`` CLIENT span already represents it. +# - ``router`` — wraps the whole request; duplicates the server span. +# - ``proxy_pre_call`` — per-callback pre-call timing; a guardrail's real span +# is ``execute_guardrail {name}``. +# - ``auth`` — emitted instead as a live phase span (see +# ``logger.phase_span``) so its DB lookups nest under it, +# not as a flat post-hoc service span. +_METRICS_ONLY_SERVICES: frozenset[str] = frozenset( + {"self", "router", "proxy_pre_call", "auth"} +) + + +def span_role_for_service(service_name: str) -> SpanRole | None: + """The span role for a service call, or ``None`` when it must not be a span. + + ``DB_CALL`` for outbound datastores, ``SERVICE`` for genuine internal work + worth a span (background jobs), and ``None`` for framework instrumentation + that duplicates a gen-AI span or belongs in metrics only + (see ``_METRICS_ONLY_SERVICES``). + """ + if service_name in _METRICS_ONLY_SERVICES: + return None + return SpanRole.DB_CALL if db_system(service_name) is not None else SpanRole.SERVICE + + +# --- span name builders (the naming convention, per role) ------------------- # + + +# The name the FastAPI instrumentor gives the root server span. V2 never creates +# this span (the instrumentor owns it), but it anchors request-level spans to it +# and tests assert against it by name, so the literal lives here with the rest of +# the span vocabulary rather than being duplicated at each call site. +LITELLM_PROXY_REQUEST_SPAN_NAME = "Received Proxy Server Request" + + +def llm_call_span_name(data: "LLMCallSpanData") -> str: + """``"{operation} {model}"`` e.g. ``"chat gpt-4o"`` (GenAI semconv).""" + model = data.request_model or "" + return f"{data.operation.value} {model}".strip() + + +def proxy_request_span_name(data: "ProxyRequestSpanData") -> str: + """``"{method} {route}"`` (HTTP semconv).""" + return f"{data.http_method} {data.route}".strip() + + +def guardrail_span_name(data: "GuardrailSpanData") -> str: + return f"execute_guardrail {data.guardrail_name}".strip() + + +def service_span_name(data: "ServiceSpanData") -> str: + """``"{service} {call_type}"`` e.g. ``"redis set"`` — service name alone when + no call type is known, so identically-named calls stay distinguishable.""" + return f"{data.service_name} {data.call_type or ''}".strip() + + +def root_roles() -> list[SpanRole]: + """Roles that start a new trace (no in-process parent).""" + return [role for role, spec in SPAN_REGISTRY.items() if spec.parent is None] + + +def child_roles(parent: SpanRole) -> list[SpanRole]: + return [role for role, spec in SPAN_REGISTRY.items() if spec.parent == parent] + + +def validate_registry( + registry: dict[SpanRole, SpanSpec] | None = None, +) -> None: + reg = registry if registry is not None else SPAN_REGISTRY + for role, spec in reg.items(): + if spec.role is not role: + raise ValueError(f"SPAN_REGISTRY[{role}] has mismatched role {spec.role}") + if spec.parent is not None and spec.parent not in reg: + raise ValueError(f"span role {role} declares unknown parent {spec.parent}") + missing = [role for role in SpanRole if role not in reg] + if missing: + raise ValueError(f"SPAN_REGISTRY is missing roles: {missing}") diff --git a/litellm/integrations/otel/model/utils.py b/litellm/integrations/otel/model/utils.py new file mode 100644 index 00000000000..f37afc97879 --- /dev/null +++ b/litellm/integrations/otel/model/utils.py @@ -0,0 +1,103 @@ +"""Shared, OpenTelemetry-free helpers for the otel integration. + +Generic value coercion (for reading heterogeneous logging-payload dicts), time +conversion, and header parsing — pulled out of the individual modules so they +live in one place. Deliberately free of any ``opentelemetry`` import so the +OTel-free sources of truth (payloads, semconv, spans, config) can use it too. +""" + +from datetime import datetime + + +def as_str(value: object) -> str | None: + if value is None: + return None + if isinstance(value, str): + return value + return str(value) + + +def as_int(value: object) -> int | None: + if isinstance(value, bool): + return int(value) + if isinstance(value, int): + return value + if isinstance(value, float): + return int(value) + if isinstance(value, str): + try: + return int(value) + except ValueError: + return None + return None + + +def as_float(value: object) -> float | None: + if isinstance(value, bool): + return float(value) + if isinstance(value, (int, float)): + return float(value) + if isinstance(value, str): + try: + return float(value) + except ValueError: + return None + return None + + +def as_bool(value: object) -> bool | None: + if value is None: + return None + if isinstance(value, bool): + return value + return bool(value) + + +def as_str_tuple(value: object) -> tuple[str, ...] | None: + if value is None: + return None + if isinstance(value, str): + return (value,) + if isinstance(value, (list, tuple)): + return tuple(str(v) for v in value) + return None + + +def to_ns(value: datetime | float | int | None) -> int | None: + """Coerce a datetime / epoch value to integer nanoseconds.""" + if value is None: + return None + if isinstance(value, datetime): + return int(value.timestamp() * 1e9) + if isinstance(value, (int, float)) and not isinstance(value, bool): + return int(float(value) * 1e9) + return None + + +def to_seconds(value: datetime | float | int | str | None) -> float | None: + """Coerce a datetime / epoch / formatted-string value to epoch seconds.""" + if value is None: + return None + if isinstance(value, datetime): + return value.timestamp() + if isinstance(value, (int, float)) and not isinstance(value, bool): + return float(value) + if isinstance(value, str): + for fmt in ("%Y-%m-%d %H:%M:%S.%f", "%Y-%m-%d %H:%M:%S"): + try: + return datetime.strptime(value, fmt).timestamp() + except ValueError: + continue + return None + + +def parse_headers(raw: str | None) -> dict[str, str]: + """Parse an OTLP ``"k=v,k=v"`` header string into a dict.""" + headers: dict[str, str] = {} + if not raw: + return headers + for pair in raw.split(","): + if "=" in pair: + key, _, value = pair.partition("=") + headers[key.strip()] = value.strip() + return headers diff --git a/litellm/integrations/otel/mount.py b/litellm/integrations/otel/mount.py new file mode 100644 index 00000000000..ebcf4aa35af --- /dev/null +++ b/litellm/integrations/otel/mount.py @@ -0,0 +1,130 @@ +"""FastAPI server-span instrumentation — the proxy mounts this at app creation. + +``opentelemetry-instrumentation-fastapi`` creates the SERVER span for each HTTP +route and extracts inbound ``traceparent`` headers. This module owns the one call +site that attaches it to the proxy app, plus the passthrough span-naming hook, so +``proxy_server`` stays free of OTel details. + +The ``FastAPIInstrumentor`` import is kept lazy (inside :func:`instrument_fastapi_app`, +after the gate check) so importing this module never requires the optional +``opentelemetry-instrumentation-fastapi`` package and pulls in nothing OTel-related +when the feature gate is off. +""" + +import os +from typing import Any + +from litellm._logging import verbose_logger +from litellm.integrations.otel.model.config import is_otel_v2_enabled + +# Routes excluded from server-span tracing by default: high-frequency pollers and +# static UI/docs assets, none of which are LLM traffic. Entries are substring-matched +# against the request path (unanchored, so they survive a ``server_root_path`` prefix +# and each entry also covers everything beneath it — e.g. ``/health`` covers +# ``/health/readiness``). Operators override the whole set via the standard +# ``OTEL_PYTHON_FASTAPI_EXCLUDED_URLS`` env var (set "" to trace everything). +_DEFAULT_EXCLUDED_ROUTES = ( + "/health", # load-balancer liveness/readiness polling + "/metrics", # Prometheus scrape (also drops the /model/metrics admin analytics) + "/litellm-asset-prefix", # hashed UI asset bundles + "/_next", # Next.js static JS/CSS chunks (root-level mount) + "/ui", # admin UI single-page app + "/swagger", # static Swagger UI assets + "/docs", # FastAPI Swagger docs page + "/redoc", # FastAPI ReDoc docs page + "/openapi.json", # OpenAPI schema + "favicon", # /favicon.ico + /get_favicon + "/.well-known", # UI config discovery +) +_DEFAULT_EXCLUDED_URLS = ",".join(_DEFAULT_EXCLUDED_ROUTES) + +# Passthrough routes are catch-alls (e.g. "/openai/{endpoint:path}"), so the +# default OTel server-span name "{method} {route}" collapses every upstream +# endpoint into "POST /openai/{endpoint:path}". The hook below renames those spans +# to the real request path so each endpoint is distinguishable. Non-catch-all +# routes keep their low-cardinality template name. +PASSTHROUGH_PREFIXES = frozenset( + { + "openai", + "openai_passthrough", + "anthropic", + "azure", + "azure_ai", + "bedrock", + "cohere", + "cursor", + "gemini", + "mistral", + "vllm", + "vertex_ai", + "vertex-ai", + "assemblyai", + "eu.assemblyai", + "milvus", + } +) + + +def _passthrough_span_name_hook(span: Any, scope: dict) -> None: + """FastAPI ``server_request_hook``: give passthrough server spans a useful name. + + The instrumentation matches the route at span creation, so both the span name + and ``http.route`` are set to the catch-all template (``/openai/{endpoint:path}``) + before this hook runs. Rewrite both to the real request path so each upstream + endpoint is distinguishable. (The ASGI ``http receive``/``http send`` sub-spans + can't be renamed from here — their name is captured at creation — so they are + dropped via ``exclude_spans`` at instrumentation time.) + """ + try: + if span is None or not span.is_recording(): + return + path = scope.get("path") or "" + method = scope.get("method") or "" + first_segment = path.lstrip("/").split("/", 1)[0] + if first_segment in PASSTHROUGH_PREFIXES: + span.update_name(f"{method} {path}".strip()) + span.set_attribute("http.route", path) + except Exception: + pass + + +def instrument_fastapi_app(app: Any) -> None: + """Attach OTel server-span instrumentation to the proxy FastAPI app. + + Safe no-op when the V2 gate is off or ``opentelemetry-instrumentation-fastapi`` + is unavailable. This MUST be called at app-creation time — once the lifespan + runs, the middleware stack is frozen and ``instrument_app`` raises "Cannot add + middleware after an application has started". + + No ``TracerProvider`` is passed, so the instrumentation binds to the OTel global + ``ProxyTracerProvider``; the proxy publishes the real provider as the global + after config load (see ``proxy_startup_event``), and the proxy delegates to it. + That way server spans and gen-ai spans share one provider and the same trace. + """ + try: + if not is_otel_v2_enabled(): + return + + # Lazy: only the V2-enabled path needs the optional + # ``opentelemetry-instrumentation-fastapi`` package, which is not part of the + # base ``litellm[proxy]`` install. Importing it at module top would make + # ``proxy_server``'s unconditional ``import`` of this module crash when the + # package is absent, even with the gate off. + from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor + + excluded_urls = ( + os.environ.get("OTEL_PYTHON_FASTAPI_EXCLUDED_URLS") + if "OTEL_PYTHON_FASTAPI_EXCLUDED_URLS" in os.environ + else _DEFAULT_EXCLUDED_URLS + ) + FastAPIInstrumentor.instrument_app( + app, + excluded_urls=excluded_urls, + server_request_hook=_passthrough_span_name_hook, + # Drop the ASGI "http receive"/"http send" lifecycle sub-spans: they + # are low-value noise and (for passthrough) carry the catch-all route + # template in their name, which can't be rewritten from a hook. + exclude_spans=["receive", "send"], + ) + except Exception as e: + verbose_logger.debug("Skipping OTel V2 FastAPI instrumentation: %s", e) diff --git a/litellm/integrations/otel/plumbing/__init__.py b/litellm/integrations/otel/plumbing/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/litellm/integrations/otel/plumbing/context.py b/litellm/integrations/otel/plumbing/context.py new file mode 100644 index 00000000000..64790da814b --- /dev/null +++ b/litellm/integrations/otel/plumbing/context.py @@ -0,0 +1,127 @@ +"""Trace-context + Baggage helpers.""" + +from contextvars import ContextVar +from typing import Mapping + +from opentelemetry import baggage +from opentelemetry.context import Context, get_current +from opentelemetry.trace import Span, get_current_span, set_span_in_context +from opentelemetry.trace.propagation.tracecontext import ( + TraceContextTextMapPropagator, +) + +_PROPAGATOR = TraceContextTextMapPropagator() + +# The request's root span — the FastAPI-owned SERVER span — captured ONCE when the +# proxy first resolves it, so request-level spans (the LLM call, guardrails) can +# parent to it EXPLICITLY instead of to whatever span happens to be active at the +# instant they are emitted. Ambient-only parenting (``get_current_span()``) is +# wrong at two boundaries: +# * inside the ``auth`` phase span the active span is the auth span, so an LLM / +# guardrail span emitted there would nest under auth instead of being its +# sibling; and +# * in a detached success task (pass-through logs success from a fire-and-forget +# ``asyncio.create_task``) the server span may not be active at all, orphaning +# the span into a brand-new trace. +# A ``ContextVar`` (not a request attribute) so it rides the request task's context +# and is inherited by ``asyncio.create_task`` children — i.e. the async logging +# callbacks that close the span. It is never reset: the contextvar dies with the +# request task, so there is nothing to leak. +_request_root_span: "ContextVar[Span | None]" = ContextVar( + "litellm_otel_request_root_span", default=None +) + + +def set_request_root_span(span: Span) -> None: + """Anchor the request's root (server) span for explicit child parenting. + + No-ops for a non-recordable span so a bad capture can never replace a good one + with a phantom parent. Idempotent — the proxy captures the same server span at + more than one entry point. + """ + if is_recordable_span(span): + _request_root_span.set(span) + + +def request_root_span() -> "Span | None": + """The anchored request root span, or ``None`` outside a proxy request.""" + span = _request_root_span.get() + return span if is_recordable_span(span) else None + + +def set_request_baggage( + values: Mapping[str, str], context: Context | None = None +) -> Context: + """Return a context with ``values`` written into Baggage.""" + ctx = context + for key, value in values.items(): + ctx = baggage.set_baggage(key, value, context=ctx) + return ctx if ctx is not None else (context or get_current()) + + +def get_baggage_attributes(context: Context | None = None) -> dict[str, str]: + """All Baggage entries on ``context`` as strings.""" + return {key: str(value) for key, value in baggage.get_all(context).items()} + + +def context_from_span(span: Span, context: Context | None = None) -> Context: + """A context with ``span`` as the active span (for explicit parenting).""" + return set_span_in_context(span, context=context) + + +def resolve_parent_context(threaded: Span | None = None) -> Context: + """The context a child span should parent under. + + Ambient-first: parent to the active OTel context (the server span, restored + by the logging worker or active in the request task), falling back to a span + passed explicitly (``threaded``) only when the ambient context has no + recordable span — e.g. a background service call with no request on the + stack. When neither is recordable the ambient context is returned unchanged, + so the span starts a new root trace. + + Only service/DB spans pass ``threaded`` (the ``parent_otel_span`` handed to + the service hook). Request-level spans — the LLM call and guardrails — are + created where the server span is genuinely ambient, so they never need it. + """ + ctx = get_current() + if is_recordable_span(threaded) and not is_recordable_span(get_current_span(ctx)): + ctx = context_from_span(threaded, context=ctx) # type: ignore[arg-type] + return ctx + + +def resolve_request_span_context() -> Context: + """The parent context for a request-level span (the LLM call, a guardrail). + + These are direct children of the request's root server span — siblings of the + ``auth`` phase span and of each other, never nested under whatever span is + momentarily active. So prefer the explicitly anchored root span; fall back to + ambient context only when there is no anchor (the SDK / no-proxy path), where + the span legitimately starts its own root trace. + + Unlike :func:`resolve_parent_context` (used by DB/service spans, which DO want + to nest under the active phase span, e.g. an auth DB lookup under ``auth``), + this never returns the active span when an anchor exists. + """ + root = request_root_span() + if root is not None: + return context_from_span(root) + return get_current() + + +def is_recordable_span(obj: object) -> bool: + """True if ``obj`` is a live span with a valid context (safe to parent under).""" + if not isinstance(obj, Span): + return False + try: + ctx = obj.get_span_context() + except Exception: + return False + return ctx is not None and ctx.is_valid + + +def extract_traceparent(headers: Mapping[str, str]) -> Context | None: + """Extract a remote parent context from incoming HTTP headers, if present.""" + if not any(key.lower() == "traceparent" for key in headers): + return None + carrier = {str(key).lower(): value for key, value in headers.items()} + return _PROPAGATOR.extract(carrier) diff --git a/litellm/integrations/otel/plumbing/metrics.py b/litellm/integrations/otel/plumbing/metrics.py new file mode 100644 index 00000000000..edd120f91e6 --- /dev/null +++ b/litellm/integrations/otel/plumbing/metrics.py @@ -0,0 +1,28 @@ +"""GenAI client metrics (token usage + operation duration histograms).""" + +from dataclasses import dataclass + +from opentelemetry.metrics import Histogram, Meter + +from litellm.integrations.otel.model.semconv import Metric + + +@dataclass(frozen=True) +class GenAIMetrics: + token_usage: Histogram + operation_duration: Histogram + + +def create_genai_metrics(meter: Meter) -> GenAIMetrics: + return GenAIMetrics( + token_usage=meter.create_histogram( + name=Metric.TOKEN_USAGE, + unit="{token}", + description="Number of tokens used per GenAI request.", + ), + operation_duration=meter.create_histogram( + name=Metric.OPERATION_DURATION, + unit="s", + description="GenAI operation duration.", + ), + ) diff --git a/litellm/integrations/otel/plumbing/providers.py b/litellm/integrations/otel/plumbing/providers.py new file mode 100644 index 00000000000..40a0e41b905 --- /dev/null +++ b/litellm/integrations/otel/plumbing/providers.py @@ -0,0 +1,220 @@ +"""Provider / exporter factory + the Baggage span processor.""" + +from typing import Callable, Iterable + +from opentelemetry import baggage +from opentelemetry.context import Context +from opentelemetry.sdk.resources import Resource +from opentelemetry.sdk.trace import ReadableSpan, SpanProcessor, TracerProvider +from opentelemetry.sdk.trace.export import ( + BatchSpanProcessor, + ConsoleSpanExporter, + SimpleSpanProcessor, + SpanExporter, +) +from opentelemetry.sdk.trace.export.in_memory_span_exporter import ( + InMemorySpanExporter, +) +from opentelemetry.trace import Span, SpanKind, Tracer + +from litellm.integrations.otel.model.config import ExporterSpec, OpenTelemetryV2Config +from litellm.integrations.otel.model.semconv import LiteLLM +from litellm.integrations.otel.model.spans import LiteLLMSpanKind + +# Re-exported so ``providers.parse_headers`` remains a stable entry point. +from litellm.integrations.otel.model.utils import parse_headers as parse_headers + +_SPAN_KIND_BY_ROLE_KIND: dict[LiteLLMSpanKind, SpanKind] = { + LiteLLMSpanKind.SERVER: SpanKind.SERVER, + LiteLLMSpanKind.CLIENT: SpanKind.CLIENT, + LiteLLMSpanKind.INTERNAL: SpanKind.INTERNAL, + LiteLLMSpanKind.PRODUCER: SpanKind.PRODUCER, + LiteLLMSpanKind.CONSUMER: SpanKind.CONSUMER, +} + + +def to_otel_span_kind(kind: LiteLLMSpanKind) -> SpanKind: + return _SPAN_KIND_BY_ROLE_KIND[kind] + + +# Custom exporter factories keyed by ``ExporterSpec.kind``. A preset registers +# one here when its destination needs construction logic the built-in kinds +# can't express — e.g. an exporter that fetches an auth token lazily on its +# first export (off the event loop) instead of blocking at config-build time. +# Keeping the registry here lets this module stay vendor-agnostic: the factory +# lives with the integration that needs it. +_EXPORTER_FACTORIES: dict[str, Callable[[ExporterSpec], SpanExporter]] = {} + + +def register_exporter_factory( + kind: str, factory: Callable[[ExporterSpec], SpanExporter] +) -> None: + """Register a custom exporter ``factory`` for the exporter ``kind``.""" + _EXPORTER_FACTORIES[kind.lower()] = factory + + +class LiteLLMBaggageSpanProcessor(SpanProcessor): + """Stamps an allowlisted set of Baggage entries onto every span at start.""" + + def __init__( + self, + allowed_keys: Iterable[str], + allowed_prefixes: tuple[str, ...] = (LiteLLM.METADATA_PREFIX,), + ) -> None: + self._allowed_keys = frozenset(allowed_keys) + self._allowed_prefixes = tuple(allowed_prefixes) + + def _is_allowed(self, key: str) -> bool: + return key in self._allowed_keys or any( + key.startswith(prefix) for prefix in self._allowed_prefixes + ) + + def on_start(self, span: Span, parent_context: Context | None = None) -> None: + for key, value in baggage.get_all(parent_context).items(): + if self._is_allowed(key) and isinstance(value, (str, bool, int, float)): + span.set_attribute(key, value) + + def on_end(self, span: ReadableSpan) -> None: # noqa: D401 - no-op + return None + + def shutdown(self) -> None: + return None + + def force_flush(self, timeout_millis: int = 30000) -> bool: + return True + + +def _otlp_traces_endpoint(endpoint: str | None) -> str | None: + """Point an OTLP/HTTP base endpoint at the ``/v1/traces`` signal path. + + ``OTEL_EXPORTER_OTLP_ENDPOINT`` is a base URL (e.g. ``http://host:4318``). + The OTLP/HTTP exporter only appends the ``/v1/traces`` path when it reads + that env var itself; when an endpoint is passed explicitly it is used + verbatim, so a base URL would POST to the root and the collector returns + 404. Append the signal path here (leaving an already-correct path intact). + """ + if not endpoint: + return endpoint + endpoint = endpoint.rstrip("/") + # Splunk Observability uses ``/v2/trace/otlp``; never rewrite it. + if endpoint.endswith("/v1/traces") or "/v2/trace/otlp" in endpoint: + return endpoint + for other_signal in ("/v1/logs", "/v1/metrics"): + if endpoint.endswith(other_signal): + return endpoint[: -len(other_signal)] + "/v1/traces" + return endpoint + "/v1/traces" + + +def _exporter_from_spec(spec: ExporterSpec) -> SpanExporter: + kind = (spec.kind or "console").lower() + factory = _EXPORTER_FACTORIES.get(kind) + if factory is not None: + return factory(spec) + if kind in ("in_memory", "inmemory", "memory"): + return InMemorySpanExporter() + if kind in ("otlp_http", "http", "http/protobuf", "http/json"): + from opentelemetry.exporter.otlp.proto.http.trace_exporter import ( + OTLPSpanExporter as HTTPExporter, + ) + + return HTTPExporter( + endpoint=_otlp_traces_endpoint(spec.endpoint), + headers=parse_headers(spec.headers), + ) + if kind in ("otlp_grpc", "grpc"): + from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import ( + OTLPSpanExporter as GRPCExporter, + ) + + return GRPCExporter(endpoint=spec.endpoint, headers=parse_headers(spec.headers)) + return ConsoleSpanExporter() + + +def _processor_for(exporter: SpanExporter, use_simple: bool | None) -> SpanProcessor: + """Pick a Simple or Batch span processor for ``exporter``. + + When ``use_simple`` is unset, default to Simple for console and in-memory + exporters (spans export synchronously, which tests rely on) and Batch for + everything else (the right export semantics for production). + """ + if use_simple is None: + use_simple = isinstance(exporter, (ConsoleSpanExporter, InMemorySpanExporter)) + return SimpleSpanProcessor(exporter) if use_simple else BatchSpanProcessor(exporter) + + +def build_span_exporter(config: OpenTelemetryV2Config) -> SpanExporter: + """Build a single exporter from the top-level config fields. + + Convenience for the common single-exporter case (and for tests): reads the + ``exporter`` / ``endpoint`` / ``headers`` fields. To configure multiple + exporters, populate ``config.exporters`` directly. + """ + return _exporter_from_spec( + ExporterSpec( + kind=config.exporter, endpoint=config.endpoint, headers=config.headers + ) + ) + + +def build_resource(config: OpenTelemetryV2Config) -> Resource: + attributes: dict[str, str] = {"service.name": config.service_name} + if config.deployment_environment: + attributes["deployment.environment"] = config.deployment_environment + attributes.update(config.resource_attributes) + return Resource.create(attributes) + + +def build_tracer_provider( + config: OpenTelemetryV2Config, + exporter: SpanExporter | None = None, + baggage_processor: SpanProcessor | None = None, + use_simple_processor: bool | None = None, +) -> TracerProvider: + """Build the shared :class:`TracerProvider`. + + Attach the Baggage processor first (so identity attributes land on each + span before any export decision), then add one ``SpanProcessor`` per + ``config.exporters`` entry — this is what fans spans out to multiple + backends. ``exporter`` and ``use_simple_processor`` are explicit overrides: + pass a single exporter to attach exactly that one (used by tests). + """ + provider = TracerProvider(resource=build_resource(config)) + if baggage_processor is None: + baggage_processor = LiteLLMBaggageSpanProcessor( + allowed_keys=config.baggage_promoted_keys + ) + provider.add_span_processor(baggage_processor) + + if exporter is not None: + provider.add_span_processor(_processor_for(exporter, use_simple_processor)) + return provider + + # ``config._normalize`` guarantees at least one spec (it folds the top-level + # ``exporter``/``endpoint``/``headers`` fields in when ``exporters`` is empty). + for spec in config.exporters: + exp = _exporter_from_spec(spec) + provider.add_span_processor( + _processor_for( + exp, + ( + spec.use_simple_processor + if spec.use_simple_processor is not None + else use_simple_processor + ), + ) + ) + return provider + + +def get_tracer(provider: TracerProvider, name: str = "litellm") -> Tracer: + return provider.get_tracer(name) + + +def in_memory_provider( + config: OpenTelemetryV2Config | None = None, +) -> tuple[TracerProvider, InMemorySpanExporter]: + """Convenience for tests: a provider exporting to an in-memory buffer.""" + cfg = config or OpenTelemetryV2Config(exporter="in_memory") + exporter = InMemorySpanExporter() + provider = build_tracer_provider(cfg, exporter=exporter) + return provider, exporter diff --git a/litellm/integrations/otel/plumbing/routing.py b/litellm/integrations/otel/plumbing/routing.py new file mode 100644 index 00000000000..4d0943a263a --- /dev/null +++ b/litellm/integrations/otel/plumbing/routing.py @@ -0,0 +1,101 @@ +"""Per-request multi-tenant tracer routing. + +When a request carries team/key vendor credentials in +``standard_callback_dynamic_params``, its spans must export through a +``TracerProvider`` whose OTLP headers carry those credentials. +``TenantTracerCache`` builds and caches one provider per distinct credential +set, and otherwise hands back the logger's default tracer. This lets a single +logger fan requests out to many tenants without needing a logger per tenant. +""" + +from collections import OrderedDict +from typing import Any, Mapping + +from opentelemetry.sdk.trace import TracerProvider +from opentelemetry.trace import Tracer + +from litellm._logging import verbose_logger +from litellm.integrations.otel.model.config import OpenTelemetryV2Config +from litellm.integrations.otel.presets import dynamic_otlp_headers +from litellm.integrations.otel.plumbing.providers import ( + build_tracer_provider, + get_tracer, +) + +# Exporter kinds that ignore headers — never rewritten with dynamic credentials. +_NON_OTLP_KINDS = ("console", "in_memory", "inmemory", "memory") + +# Cap on distinct credential-scoped providers held at once. ``dynamic_params`` +# can be populated from request metadata, so an unbounded cache lets a caller +# spawn one ``TracerProvider`` (plus its ``BatchSpanProcessor`` background +# thread) per unique credential set and exhaust the proxy. The LRU bound keeps +# the working set of active tenants resident while flushing and shutting down +# evicted providers so their threads are reclaimed. +_MAX_CACHED_PROVIDERS = 256 + + +def _shutdown_provider(provider: TracerProvider) -> None: + """Flush + stop an evicted provider's processors (reclaims their threads). + + ``TracerProvider.shutdown`` force-flushes each ``SpanProcessor`` before + stopping it, so any spans already handed to a ``BatchSpanProcessor`` are + exported rather than dropped. Best-effort: a shutdown failure must not break + the request that triggered the eviction. + """ + try: + provider.shutdown() + except Exception as e: # pragma: no cover - defensive + verbose_logger.debug("OTel V2: error shutting down evicted provider: %s", e) + + +class TenantTracerCache: + """Credential-scoped ``TracerProvider`` cache keyed by the dynamic headers.""" + + def __init__( + self, + config: OpenTelemetryV2Config, + callback_name: str | None, + tracer_name: str, + ) -> None: + self._config = config + self._callback_name = callback_name + self._tracer_name = tracer_name + self._providers: "OrderedDict[tuple[tuple[str, str], ...], TracerProvider]" = ( + OrderedDict() + ) + + def tracer_for(self, default: Tracer, dynamic_params: Any) -> Tracer: + """Return the tracer for this request. + + Use ``default`` unless the request's dynamic credentials require a + credential-scoped tracer, in which case build (or reuse) one. The cache + is a bounded LRU: the least-recently-used provider is flushed and shut + down on overflow so its exporter threads don't accumulate. + """ + headers = dynamic_otlp_headers(self._callback_name, dynamic_params) + if not headers: + return default + cache_key = tuple(sorted(headers.items())) + provider = self._providers.get(cache_key) + if provider is not None: + self._providers.move_to_end(cache_key) + else: + provider = build_tracer_provider(self._config_with_headers(headers)) + self._providers[cache_key] = provider + if len(self._providers) > _MAX_CACHED_PROVIDERS: + _, evicted = self._providers.popitem(last=False) + _shutdown_provider(evicted) + return get_tracer(provider, self._tracer_name) + + def _config_with_headers(self, headers: Mapping[str, str]) -> OpenTelemetryV2Config: + """Clone the config, replacing OTLP exporter headers with ``headers``.""" + header_str = ",".join(f"{key}={value}" for key, value in headers.items()) + exporters = [ + ( + spec + if spec.kind.lower() in _NON_OTLP_KINDS + else spec.model_copy(update={"headers": header_str}) + ) + for spec in self._config.exporters + ] + return self._config.model_copy(update={"exporters": exporters}) diff --git a/litellm/integrations/otel/presets/__init__.py b/litellm/integrations/otel/presets/__init__.py new file mode 100644 index 00000000000..c69d257ab52 --- /dev/null +++ b/litellm/integrations/otel/presets/__init__.py @@ -0,0 +1,78 @@ +"""Integration presets — each one returns an :class:`OpenTelemetryV2Config`. + +A preset is a callable that reads an integration's env vars and returns an +``OpenTelemetryV2Config`` describing the exporter destination, the mapper +vocabularies to apply, and any resource attributes. ``PRESET_BY_CALLBACK`` +maps a callback name (``"arize"``, ``"langfuse_otel"``, ...) to its preset so +the factory in ``litellm_logging`` can resolve a name and build a single +``OpenTelemetryV2`` instance from the result. +""" + +from typing import Callable + +from litellm.integrations.otel.presets.agentops import agentops_preset +from litellm.integrations.otel.presets.arize import arize_dynamic_headers, arize_preset +from litellm.integrations.otel.presets.base import Preset +from litellm.integrations.otel.presets.langfuse import ( + langfuse_dynamic_headers, + langfuse_preset, +) +from litellm.integrations.otel.presets.langtrace import langtrace_preset +from litellm.integrations.otel.presets.levo import levo_preset +from litellm.integrations.otel.presets.phoenix import phoenix_preset +from litellm.integrations.otel.presets.weave import weave_dynamic_headers, weave_preset +from litellm.types.utils import StandardCallbackDynamicParams + +#: Callback name → preset. The ``Preset`` annotation makes mypy verify every +#: registered value matches the preset interface. +PRESET_BY_CALLBACK: dict[str, Preset] = { + "agentops": agentops_preset, + "arize": arize_preset, + "arize_phoenix": phoenix_preset, + "langfuse_otel": langfuse_preset, + "langtrace": langtrace_preset, + "levo": levo_preset, + "weave_otel": weave_preset, +} + +#: Callback name → per-request OTLP header builder (team/key multi-tenant +#: routing). Only integrations that support dynamic credentials appear here — +#: Arize-Phoenix/Langtrace/Levo/AgentOps don't, so they use the logger's +#: default tracer. +DYNAMIC_HEADERS_BY_CALLBACK: dict[ + str, Callable[[StandardCallbackDynamicParams], dict[str, str]] +] = { + "arize": arize_dynamic_headers, + "langfuse_otel": langfuse_dynamic_headers, + "weave_otel": weave_dynamic_headers, +} + + +def dynamic_otlp_headers( + callback_name: str | None, + dynamic_params: StandardCallbackDynamicParams | None, +) -> dict[str, str] | None: + """Per-request OTLP headers for ``callback_name``, or ``None`` if N/A. + + ``None`` means "no per-request routing" — the caller uses its default tracer. + """ + builder = DYNAMIC_HEADERS_BY_CALLBACK.get(callback_name or "") + if builder is None or not dynamic_params: + return None + headers = builder(dynamic_params) + return headers or None + + +__all__ = [ + "PRESET_BY_CALLBACK", + "DYNAMIC_HEADERS_BY_CALLBACK", + "Preset", + "dynamic_otlp_headers", + "agentops_preset", + "arize_preset", + "langfuse_preset", + "langtrace_preset", + "levo_preset", + "phoenix_preset", + "weave_preset", +] diff --git a/litellm/integrations/otel/presets/agentops.py b/litellm/integrations/otel/presets/agentops.py new file mode 100644 index 00000000000..5a12818fd99 --- /dev/null +++ b/litellm/integrations/otel/presets/agentops.py @@ -0,0 +1,139 @@ +"""AgentOps preset — OTLP/HTTP to AgentOps' endpoint with a lazily-fetched JWT. + +AgentOps authenticates with a short-lived JWT minted from the API key. Fetching +it is blocking network I/O, so it must never run on the event loop: callback +construction (where presets are built) can run inside the proxy's async startup +or, in the SDK, on the first request. Instead of fetching at config-build time, +this preset registers a custom exporter (``kind="agentops"``) that mints the JWT +**on its first export** — which the ``BatchSpanProcessor`` runs in its own +worker thread, off any event loop — and caches it for the process lifetime. +""" + +from typing import Any + +import httpx +from pydantic import Field +from pydantic_settings import BaseSettings, SettingsConfigDict + +from litellm._logging import verbose_logger +from litellm.integrations.otel.model.config import ExporterSpec, OpenTelemetryV2Config +from litellm.integrations.otel.plumbing.providers import register_exporter_factory + +_AGENTOPS_ENDPOINT = "https://otlp.agentops.cloud/v1/traces" +_AGENTOPS_AUTH_ENDPOINT = "https://api.agentops.ai/v3/auth/token" +_AGENTOPS_EXPORTER_KIND = "agentops" + + +class _AgentOpsSettings(BaseSettings): + model_config = SettingsConfigDict(case_sensitive=False, extra="ignore") + + api_key: str | None = Field(default=None, validation_alias="AGENTOPS_API_KEY") + service_name: str = Field( + default="agentops", validation_alias="AGENTOPS_SERVICE_NAME" + ) + environment: str | None = Field( + default=None, validation_alias="AGENTOPS_ENVIRONMENT" + ) + + +def agentops_preset( + *, + config_overrides: OpenTelemetryV2Config | None = None, +) -> OpenTelemetryV2Config: + """Build the AgentOps config without any network I/O. + + The ``agentops`` exporter mints (and caches) the JWT lazily on its first + export, so this stays non-blocking. ``project.id`` is therefore not a + resource attribute — it is encoded in the JWT, which AgentOps uses to route + the trace to the right project. + """ + settings = _AgentOpsSettings() + base = config_overrides or OpenTelemetryV2Config() + return base.model_copy( + update={ + "exporters": [ + *base.exporters, + ExporterSpec( + kind=_AGENTOPS_EXPORTER_KIND, + endpoint=_AGENTOPS_ENDPOINT, + options=( + {"api_key": settings.api_key} if settings.api_key else None + ), + ), + ], + "resource_attributes": { + **base.resource_attributes, + "service.name": settings.service_name, + "telemetry.sdk.name": "agentops", + **( + {"deployment.environment": settings.environment} + if settings.environment + else {} + ), + }, + } + ) + + +def _build_agentops_exporter(spec: ExporterSpec) -> Any: + """Factory for the ``agentops`` exporter kind: a lazy-auth OTLP/HTTP exporter.""" + from opentelemetry.exporter.otlp.proto.http.trace_exporter import ( + OTLPSpanExporter, + ) + + class _LazyAuthAgentOpsExporter(OTLPSpanExporter): + """OTLP/HTTP exporter that mints the AgentOps JWT on its first export. + + ``export`` runs in the ``BatchSpanProcessor`` worker thread, so the + blocking token fetch never touches an event loop. The result is cached + after the first attempt (success or failure) so it runs at most once. + """ + + def __init__(self, *, endpoint: str | None, api_key: str | None) -> None: + super().__init__(endpoint=endpoint) + self._agentops_api_key = api_key + self._auth_resolved = False + + def _ensure_authenticated(self) -> None: + if self._auth_resolved: + return + self._auth_resolved = True + if not self._agentops_api_key: + return + try: + token = _fetch_agentops_jwt(self._agentops_api_key).get("token") + if token: + # ``_session`` is the requests.Session the base exporter + # POSTs through; updating its Authorization header is how the + # minted JWT reaches every subsequent export. + self._session.headers["Authorization"] = f"Bearer {token}" + except Exception as e: + verbose_logger.debug("AgentOps JWT fetch failed: %s", e) + + def export(self, spans: Any) -> Any: + self._ensure_authenticated() + return super().export(spans) + + options = spec.options or {} + return _LazyAuthAgentOpsExporter( + endpoint=spec.endpoint, api_key=options.get("api_key") + ) + + +def _fetch_agentops_jwt(api_key: str) -> dict[str, Any]: + # Own a short-lived client rather than ``_get_httpx_client()``: that returns + # a process-wide cached ``HTTPHandler`` whose connection pool is shared by + # every caller, so closing it here would break concurrent/subsequent + # requests. This one-shot auth call gets its own client to close. + with httpx.Client(timeout=10) as client: + response = client.post( + url=_AGENTOPS_AUTH_ENDPOINT, + headers={"Content-Type": "application/json", "Connection": "keep-alive"}, + json={"api_key": api_key}, + ) + if response.status_code != 200: + raise RuntimeError(f"Failed to fetch AgentOps token: {response.text}") + return response.json() + + +register_exporter_factory(_AGENTOPS_EXPORTER_KIND, _build_agentops_exporter) diff --git a/litellm/integrations/otel/presets/arize.py b/litellm/integrations/otel/presets/arize.py new file mode 100644 index 00000000000..4df15125f5a --- /dev/null +++ b/litellm/integrations/otel/presets/arize.py @@ -0,0 +1,75 @@ +"""Arize preset — OTLP exporter to Arize + OpenInference vocabulary.""" + +from pydantic import Field +from pydantic_settings import BaseSettings, SettingsConfigDict + +from litellm.integrations.arize.arize import ArizeLogger as _V1ArizeLogger +from litellm.integrations.otel.model.config import ExporterSpec, OpenTelemetryV2Config +from litellm.integrations.otel.presets.utils import ensure_mappers +from litellm.types.utils import StandardCallbackDynamicParams + + +class _ArizeSettings(BaseSettings): + model_config = SettingsConfigDict(case_sensitive=False, extra="ignore") + + # Standard OTLP headers env var, used as the fallback when no Arize + # credentials are configured. + otlp_traces_headers: str | None = Field( + default=None, validation_alias="OTEL_EXPORTER_OTLP_TRACES_HEADERS" + ) + + +def arize_preset( + *, + config_overrides: OpenTelemetryV2Config | None = None, +) -> OpenTelemetryV2Config: + arize_cfg = _V1ArizeLogger.get_arize_config() + headers = _arize_headers(arize_cfg) + base = config_overrides or OpenTelemetryV2Config() + return base.model_copy( + update={ + "exporters": [ + *base.exporters, + ExporterSpec( + kind=arize_cfg.protocol or "otlp_grpc", + endpoint=arize_cfg.endpoint or "https://otlp.arize.com/v1", + headers=headers, + ), + ], + "mapper_names": ensure_mappers(base.mapper_names, "openinference"), + "resource_attributes": { + **base.resource_attributes, + **( + {"model_id": arize_cfg.project_name} + if arize_cfg.project_name + else {} + ), + }, + } + ) + + +def _arize_headers(arize_cfg) -> str | None: + pieces = [] + if arize_cfg.space_id or arize_cfg.space_key: + pieces.append(f"space_id={arize_cfg.space_id or arize_cfg.space_key}") + if arize_cfg.api_key: + pieces.append(f"api_key={arize_cfg.api_key}") + if not pieces: + # Fall back to the standard OTLP headers env var when no Arize + # credentials are configured. + return _ArizeSettings().otlp_traces_headers + return ",".join(pieces) + + +def arize_dynamic_headers(params: StandardCallbackDynamicParams) -> dict[str, str]: + """Per-request Arize OTLP headers from team/key dynamic params.""" + headers: dict[str, str] = {} + # ``arize_space_key`` is the suggested param and wins over ``arize_space_id``. + space = params.get("arize_space_key") or params.get("arize_space_id") + if space: + headers["arize-space-id"] = space + api_key = params.get("arize_api_key") + if api_key: + headers["api_key"] = api_key + return headers diff --git a/litellm/integrations/otel/presets/base.py b/litellm/integrations/otel/presets/base.py new file mode 100644 index 00000000000..b50908e7652 --- /dev/null +++ b/litellm/integrations/otel/presets/base.py @@ -0,0 +1,25 @@ +"""Preset interface. + +A preset is a callable that reads its integration's env vars and produces an +:class:`OpenTelemetryV2Config` (exporter list + mapper-name list + resource +attributes). This ``Protocol`` pins that contract so ``PRESET_BY_CALLBACK`` and +the factory in ``litellm_logging`` are type-checked structurally against it, +matching the ``AttributeMapper`` protocol the mappers use. +""" + +from typing import Protocol, runtime_checkable + +from litellm.integrations.otel.model.config import OpenTelemetryV2Config + + +@runtime_checkable +class Preset(Protocol): + """Reads an integration's env config and returns an ``OpenTelemetryV2Config``. + + ``config_overrides`` lets one preset layer onto another's config (or onto + test-supplied defaults); the factory calls presets with no arguments. + """ + + def __call__( + self, *, config_overrides: OpenTelemetryV2Config | None = None + ) -> OpenTelemetryV2Config: ... diff --git a/litellm/integrations/otel/presets/langfuse.py b/litellm/integrations/otel/presets/langfuse.py new file mode 100644 index 00000000000..011545384b9 --- /dev/null +++ b/litellm/integrations/otel/presets/langfuse.py @@ -0,0 +1,43 @@ +"""Langfuse-OTEL preset.""" + +from litellm.integrations.langfuse.langfuse_otel import ( + LangfuseOtelLogger as _V1Langfuse, +) +from litellm.integrations.otel.model.config import ExporterSpec, OpenTelemetryV2Config +from litellm.integrations.otel.presets.utils import ensure_mappers +from litellm.types.utils import StandardCallbackDynamicParams + + +def langfuse_preset( + *, + config_overrides: OpenTelemetryV2Config | None = None, +) -> OpenTelemetryV2Config: + cfg = _V1Langfuse.get_langfuse_otel_config() + kind = cfg.exporter if isinstance(cfg.exporter, str) else "otlp_http" + base = config_overrides or OpenTelemetryV2Config() + return base.model_copy( + update={ + "exporters": [ + *base.exporters, + ExporterSpec( + kind=kind, + endpoint=cfg.endpoint, + headers=cfg.headers, + ), + ], + "mapper_names": ensure_mappers(base.mapper_names, "langfuse"), + } + ) + + +def langfuse_dynamic_headers(params: StandardCallbackDynamicParams) -> dict[str, str]: + """Per-request Langfuse OTLP headers from team/key dynamic params.""" + public_key = params.get("langfuse_public_key") + secret_key = params.get("langfuse_secret_key") + if public_key and secret_key: + return { + "Authorization": _V1Langfuse._get_langfuse_authorization_header( + public_key=public_key, secret_key=secret_key + ) + } + return {} diff --git a/litellm/integrations/otel/presets/langtrace.py b/litellm/integrations/otel/presets/langtrace.py new file mode 100644 index 00000000000..acdbaf870d3 --- /dev/null +++ b/litellm/integrations/otel/presets/langtrace.py @@ -0,0 +1,22 @@ +"""Langtrace preset — Langtrace consumes generic OTLP + a vendor mapper.""" + +from litellm.integrations.otel.model.config import OpenTelemetryV2Config +from litellm.integrations.otel.presets.utils import ensure_mappers + + +def langtrace_preset( + *, + config_overrides: OpenTelemetryV2Config | None = None, +) -> OpenTelemetryV2Config: + """Compose the Langtrace mapper on top of the customer's OTLP destination. + + Unlike Arize / Phoenix / Langfuse, Langtrace doesn't ship its own endpoint + — users point their existing OTLP collector at Langtrace and just + need the vendor attribute schema applied to outgoing spans. + """ + base = config_overrides or OpenTelemetryV2Config() + return base.model_copy( + update={ + "mapper_names": ensure_mappers(base.mapper_names, "langtrace"), + } + ) diff --git a/litellm/integrations/otel/presets/levo.py b/litellm/integrations/otel/presets/levo.py new file mode 100644 index 00000000000..4c4cba982a4 --- /dev/null +++ b/litellm/integrations/otel/presets/levo.py @@ -0,0 +1,24 @@ +"""Levo preset — OTLP/HTTP to a Levo collector with org+workspace headers.""" + +from litellm.integrations.levo.levo import LevoLogger as _V1Levo +from litellm.integrations.otel.model.config import ExporterSpec, OpenTelemetryV2Config + + +def levo_preset( + *, + config_overrides: OpenTelemetryV2Config | None = None, +) -> OpenTelemetryV2Config: + cfg = _V1Levo.get_levo_config() + base = config_overrides or OpenTelemetryV2Config() + return base.model_copy( + update={ + "exporters": [ + *base.exporters, + ExporterSpec( + kind="otlp_http", + endpoint=cfg.endpoint, + headers=cfg.otlp_auth_headers, + ), + ], + } + ) diff --git a/litellm/integrations/otel/presets/phoenix.py b/litellm/integrations/otel/presets/phoenix.py new file mode 100644 index 00000000000..4c2b165ffca --- /dev/null +++ b/litellm/integrations/otel/presets/phoenix.py @@ -0,0 +1,48 @@ +"""Arize-Phoenix preset.""" + +from pydantic import AliasChoices, Field +from pydantic_settings import BaseSettings, SettingsConfigDict + +from litellm.integrations.arize.arize_phoenix import ( + ArizePhoenixLogger as _V1Phoenix, +) +from litellm.integrations.otel.model.config import ExporterSpec, OpenTelemetryV2Config +from litellm.integrations.otel.presets.utils import ensure_mappers + + +class _PhoenixSettings(BaseSettings): + model_config = SettingsConfigDict(case_sensitive=False, extra="ignore") + + project_name: str = Field( + default="default", + validation_alias=AliasChoices( + "PHOENIX_PROJECT_NAME", "PHOENIX_COLLECTOR_PROJECT_NAME" + ), + ) + + +def phoenix_preset( + *, + config_overrides: OpenTelemetryV2Config | None = None, +) -> OpenTelemetryV2Config: + cfg = _V1Phoenix.get_arize_phoenix_config() + headers = cfg.otlp_auth_headers if hasattr(cfg, "otlp_auth_headers") else None + project_name = _PhoenixSettings().project_name + base = config_overrides or OpenTelemetryV2Config() + return base.model_copy( + update={ + "exporters": [ + *base.exporters, + ExporterSpec( + kind=cfg.protocol if hasattr(cfg, "protocol") else "otlp_http", + endpoint=cfg.endpoint, + headers=headers, + ), + ], + "mapper_names": ensure_mappers(base.mapper_names, "openinference"), + "resource_attributes": { + **base.resource_attributes, + "openinference.project.name": project_name, + }, + } + ) diff --git a/litellm/integrations/otel/presets/utils.py b/litellm/integrations/otel/presets/utils.py new file mode 100644 index 00000000000..fdf8184441d --- /dev/null +++ b/litellm/integrations/otel/presets/utils.py @@ -0,0 +1,16 @@ +"""Shared helpers for the integration presets.""" + +from typing import Iterable + + +def ensure_mappers(mapper_names: Iterable[str], *names: str) -> list[str]: + """Return ``mapper_names`` with each of ``names`` appended if not already present. + + Order is preserved and duplicates are skipped, so composing several presets + (or re-applying one) never double-adds a vocabulary. + """ + result = list(mapper_names) + for name in names: + if name not in result: + result.append(name) + return result diff --git a/litellm/integrations/otel/presets/weave.py b/litellm/integrations/otel/presets/weave.py new file mode 100644 index 00000000000..9fc03c84a6d --- /dev/null +++ b/litellm/integrations/otel/presets/weave.py @@ -0,0 +1,43 @@ +"""Weave (W&B) preset.""" + +from litellm.integrations.otel.model.config import ExporterSpec, OpenTelemetryV2Config +from litellm.integrations.otel.presets.utils import ensure_mappers +from litellm.integrations.weave.weave_otel import ( + _get_weave_authorization_header, + get_weave_otel_config, +) +from litellm.types.utils import StandardCallbackDynamicParams + + +def weave_preset( + *, + config_overrides: OpenTelemetryV2Config | None = None, +) -> OpenTelemetryV2Config: + weave_cfg = get_weave_otel_config() + base = config_overrides or OpenTelemetryV2Config() + return base.model_copy( + update={ + "exporters": [ + *base.exporters, + ExporterSpec( + kind=weave_cfg.protocol or "otlp_http", + endpoint=weave_cfg.endpoint, + headers=weave_cfg.otlp_auth_headers, + ), + ], + # Weave consumes OpenInference + a small Weave-specific overlay. + "mapper_names": ensure_mappers(base.mapper_names, "openinference", "weave"), + } + ) + + +def weave_dynamic_headers(params: StandardCallbackDynamicParams) -> dict[str, str]: + """Per-request Weave OTLP headers from team/key dynamic params.""" + headers: dict[str, str] = {} + api_key = params.get("wandb_api_key") + if api_key: + headers["Authorization"] = _get_weave_authorization_header(api_key=api_key) + project_id = params.get("weave_project_id") + if project_id: + headers["project_id"] = project_id + return headers diff --git a/litellm/integrations/otel/runtime.py b/litellm/integrations/otel/runtime.py new file mode 100644 index 00000000000..ac3b991c971 --- /dev/null +++ b/litellm/integrations/otel/runtime.py @@ -0,0 +1,38 @@ +"""SDK-free entrypoints for proxy-core call sites (auth, …). + +Proxy code may run without the OpenTelemetry SDK installed, so it must not import +``litellm.integrations.otel.logger`` (which imports the SDK at module scope) at +module load. These wrappers import it lazily and no-op when the SDK is absent or +V2 is not the active logger — so a call site can wrap a request phase or seed +identity unconditionally. +""" + +from contextlib import contextmanager +from typing import Any, Iterator + + +@contextmanager +def phase_span(name: str) -> "Iterator[Any]": + """Run a request phase inside a live active span so its DB/service calls nest. + + Yields ``None`` (a plain no-op) when the OTel SDK is unavailable or V2 is not + the active logger. + """ + try: + from litellm.integrations.otel.logger import phase_span as _phase_span + except Exception: + yield None + return + with _phase_span(name) as span: + yield span + + +def seed_request_identity(user_api_key_dict: Any, model: Any = None) -> None: + """Seed request-identity Baggage at the auth boundary (no-op without V2).""" + try: + from litellm.integrations.otel.logger import ( + seed_request_identity as _seed_request_identity, + ) + except Exception: + return + _seed_request_identity(user_api_key_dict, model=model) diff --git a/litellm/integrations/prometheus.py b/litellm/integrations/prometheus.py index f9b1c666439..5f052842122 100644 --- a/litellm/integrations/prometheus.py +++ b/litellm/integrations/prometheus.py @@ -166,6 +166,53 @@ class PrometheusLogger(CustomLogger): labelnames=self.get_labels_for_metric("litellm_output_tokens_metric"), ) + # Token-type detail metrics. These break out cached, cache-creation, + # audio and reasoning tokens that providers report inside + # prompt_tokens_details / completion_tokens_details on the usage + # object. They are sparse (only incremented when the provider + # reports a non-zero value) and are additive to the existing + # input/output token totals — no breaking change for existing + # dashboards built on the totals. + self.litellm_input_cached_tokens_metric = self._counter_factory( + "litellm_input_cached_tokens_metric", + "Provider-side cached input tokens (e.g. OpenAI prompt_tokens_details.cached_tokens, Anthropic cache_read_input_tokens)", + labelnames=self.get_labels_for_metric( + "litellm_input_cached_tokens_metric" + ), + ) + + self.litellm_input_cache_creation_tokens_metric = self._counter_factory( + "litellm_input_cache_creation_tokens_metric", + "Provider-side input tokens written to prompt cache (e.g. Anthropic cache_creation_input_tokens)", + labelnames=self.get_labels_for_metric( + "litellm_input_cache_creation_tokens_metric" + ), + ) + + self.litellm_input_audio_tokens_metric = self._counter_factory( + "litellm_input_audio_tokens_metric", + "Audio input tokens reported in prompt_tokens_details.audio_tokens", + labelnames=self.get_labels_for_metric( + "litellm_input_audio_tokens_metric" + ), + ) + + self.litellm_output_reasoning_tokens_metric = self._counter_factory( + "litellm_output_reasoning_tokens_metric", + "Reasoning tokens reported in completion_tokens_details.reasoning_tokens", + labelnames=self.get_labels_for_metric( + "litellm_output_reasoning_tokens_metric" + ), + ) + + self.litellm_output_audio_tokens_metric = self._counter_factory( + "litellm_output_audio_tokens_metric", + "Audio output tokens reported in completion_tokens_details.audio_tokens", + labelnames=self.get_labels_for_metric( + "litellm_output_audio_tokens_metric" + ), + ) + # Remaining Budget for Team self.litellm_remaining_team_budget_metric = self._gauge_factory( "litellm_remaining_team_budget_metric", @@ -1226,6 +1273,17 @@ class PrometheusLogger(CustomLogger): label_context=label_context, ) + # Provider-agnostic fallback: providers like Bedrock and Vertex don't return + # x-ratelimit-remaining-* headers, so the gauges above only fire for OpenAI / + # Anthropic / Azure. When the proxy router has tpm/rpm configured for the + # model_group, derive remaining from configured-limit minus current usage so + # the same metric is populated for any provider. + await self._async_set_router_remaining_metrics( + standard_logging_payload=standard_logging_payload, # type: ignore + enum_values=enum_values, + label_context=label_context, + ) + # cache metrics self._increment_cache_metrics( standard_logging_payload=standard_logging_payload, # type: ignore @@ -1290,6 +1348,101 @@ class PrometheusLogger(CustomLogger): amount=float(standard_logging_payload["completion_tokens"]), ) + # Token-type detail metrics — sparse, only emitted when the provider + # reports a non-zero value in usage.prompt_tokens_details / + # usage.completion_tokens_details. + self._increment_token_detail_metrics( + standard_logging_payload=standard_logging_payload, + enum_values=enum_values, + label_context=label_context, + ) + + def _increment_token_detail_metrics( + self, + standard_logging_payload: StandardLoggingPayload, + enum_values: UserAPIKeyLabelValues, + label_context: Optional[PrometheusLabelFactoryContext] = None, + ) -> None: + """ + Increment per-token-type counters from the Usage object that providers + attach to the request. The Usage dict is plumbed onto + ``standard_logging_payload["metadata"]["usage_object"]`` by + ``get_standard_logging_object_payload``. + + Each counter is only incremented when the underlying value is > 0, so + scrape output stays sparse for providers that don't report these + details (most non-OpenAI/Anthropic models). + """ + metadata = standard_logging_payload.get("metadata") or {} + usage_object = ( + metadata.get("usage_object") if isinstance(metadata, dict) else None + ) + if not isinstance(usage_object, dict): + return + + prompt_details = usage_object.get("prompt_tokens_details") or {} + completion_details = usage_object.get("completion_tokens_details") or {} + + detail_metrics: List[Tuple[Any, DEFINED_PROMETHEUS_METRICS, Any]] = [ + ( + self.litellm_input_cached_tokens_metric, + "litellm_input_cached_tokens_metric", + ( + prompt_details.get("cached_tokens") + if isinstance(prompt_details, dict) + else None + ), + ), + ( + self.litellm_input_cache_creation_tokens_metric, + "litellm_input_cache_creation_tokens_metric", + ( + prompt_details.get("cache_creation_tokens") + if isinstance(prompt_details, dict) + else None + ), + ), + ( + self.litellm_input_audio_tokens_metric, + "litellm_input_audio_tokens_metric", + ( + prompt_details.get("audio_tokens") + if isinstance(prompt_details, dict) + else None + ), + ), + ( + self.litellm_output_reasoning_tokens_metric, + "litellm_output_reasoning_tokens_metric", + ( + completion_details.get("reasoning_tokens") + if isinstance(completion_details, dict) + else None + ), + ), + ( + self.litellm_output_audio_tokens_metric, + "litellm_output_audio_tokens_metric", + ( + completion_details.get("audio_tokens") + if isinstance(completion_details, dict) + else None + ), + ), + ] + + for counter, metric_name, value in detail_metrics: + if not isinstance(value, (int, float)) or value <= 0: + continue + PrometheusLogger._inc_labeled_counter( + self, + counter, + metric_name, + enum_values, + label_context=label_context, + amount=float(value), + ) + def _increment_cache_metrics( self, standard_logging_payload: StandardLoggingPayload, @@ -2199,6 +2352,99 @@ class PrometheusLogger(CustomLogger): ) self.litellm_deployment_rpm_limit.labels(**_labels).set(rpm) + async def _async_set_router_remaining_metrics( + self, + standard_logging_payload: StandardLoggingPayload, + enum_values: UserAPIKeyLabelValues, + label_context: Optional[PrometheusLabelFactoryContext] = None, + ) -> None: + """ + Populate ``litellm_remaining_tokens_metric`` / + ``litellm_remaining_requests_metric`` from the router's internal usage + counters when the upstream provider did not return + ``x-ratelimit-remaining-*`` response headers. + + OpenAI / Anthropic / Azure return remaining tokens/requests in response + headers, but Bedrock and Vertex AI do not. This fallback computes + ``configured_limit - current_usage`` via + ``Router.get_remaining_model_group_usage`` so the same gauges are + emitted for every provider when tpm/rpm is configured on the + deployment. + """ + try: + additional_headers = ( + standard_logging_payload.get("hidden_params", {}) or {} + ).get("additional_headers") or {} + + already_have_tokens = ( + additional_headers.get("x_ratelimit_remaining_tokens") is not None + ) + already_have_requests = ( + additional_headers.get("x_ratelimit_remaining_requests") is not None + ) + if already_have_tokens and already_have_requests: + return + + model_group = standard_logging_payload.get("model_group") + if not model_group: + return + + try: + from litellm.proxy.proxy_server import llm_router + except ImportError: + llm_router = None + + if llm_router is None: + return + + try: + remaining_usage = await llm_router.get_remaining_model_group_usage( + model_group + ) + except Exception as e: + verbose_logger.exception( + "Prometheus: get_remaining_model_group_usage failed for " + "model_group=%s: %s", + model_group, + e, + ) + return + + if not remaining_usage: + return + + remaining_tokens = remaining_usage.get("x-ratelimit-remaining-tokens") + remaining_requests = remaining_usage.get("x-ratelimit-remaining-requests") + + if not already_have_tokens and remaining_tokens is not None: + _labels = prometheus_label_factory( + supported_enum_labels=self.get_labels_for_metric( + metric_name="litellm_remaining_tokens_metric" + ), + enum_values=enum_values, + label_context=label_context, + ) + self.litellm_remaining_tokens_metric.labels(**_labels).set( + remaining_tokens + ) + + if not already_have_requests and remaining_requests is not None: + _labels = prometheus_label_factory( + supported_enum_labels=self.get_labels_for_metric( + metric_name="litellm_remaining_requests_metric" + ), + enum_values=enum_values, + label_context=label_context, + ) + self.litellm_remaining_requests_metric.labels(**_labels).set( + remaining_requests + ) + except Exception as e: + verbose_logger.exception( + "Prometheus Error: _async_set_router_remaining_metrics. " + "Exception occured - {}".format(str(e)) + ) + def set_llm_deployment_success_metrics( self, request_kwargs: dict, @@ -3436,6 +3682,10 @@ class PrometheusLogger(CustomLogger): user_object.budget_reset_at = user_info.budget_reset_at if user_object.max_budget is None and user_info.max_budget is not None: user_object.max_budget = user_info.max_budget + if user_info.user_email is not None: + user_object.user_email = user_info.user_email + if user_info.user_alias is not None: + user_object.user_alias = user_info.user_alias return user_object @@ -3452,6 +3702,8 @@ class PrometheusLogger(CustomLogger): """ enum_values = UserAPIKeyLabelValues( user=user.user_id, + user_email=user.user_email or "", + user_alias=user.user_alias or "", ) _labels = prometheus_label_factory( diff --git a/litellm/integrations/rubrik.py b/litellm/integrations/rubrik.py new file mode 100644 index 00000000000..af396ecdc73 --- /dev/null +++ b/litellm/integrations/rubrik.py @@ -0,0 +1,605 @@ +"""Rubrik LiteLLM Plugin for tool blocking and batch logging.""" + +import asyncio +import os +import random +import time +import urllib.parse +import uuid +from collections import Counter +from typing import TYPE_CHECKING, Any, Literal, Optional + +import httpx +from litellm._logging import verbose_logger +from litellm.integrations.custom_batch_logger import CustomBatchLogger +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + ModifyResponseException, +) +from litellm.litellm_core_utils.core_helpers import safe_deep_copy +from litellm.llms.custom_httpx.http_handler import ( + get_async_httpx_client, + httpxSpecialProvider, +) +from litellm.types.guardrails import GuardrailEventHooks +from litellm.types.utils import ( + ChatCompletionMessageToolCall, + Function, + GenericGuardrailAPIInputs, + StandardLoggingPayload, +) + +if TYPE_CHECKING: + from litellm.litellm_core_utils.litellm_logging import ( + Logging as LiteLLMLoggingObj, + ) + +_ENDPOINT_ANTHROPIC_MESSAGES = "/v1/messages" +_WEBHOOK_PATH_TOOL_BLOCKING = "/v1/after_completion/openai/v1" +_WEBHOOK_PATH_LOGGING_BATCH = "/v1/litellm/batch" +_MAX_QUEUE_SIZE = 10_000 +_DROP_WARNING_INTERVAL_SECONDS = 60.0 + + +class _MalformedToolBlockingResponseError(Exception): + """Raised when the tool blocking service returns a structurally invalid + response (e.g. empty ``choices``). + + Distinct from transient network/HTTP errors so callers can surface a + louder, misconfiguration-style log instead of treating it as a routine + fail-open. + """ + + +class RubrikLogger(CustomGuardrail, CustomBatchLogger): + def __init__( + self, + api_key: str | None = None, + api_base: str | None = None, + **kwargs, + ): + self.flush_lock = asyncio.Lock() + kwargs.setdefault("guardrail_name", "rubrik") + # `initialize_guardrail` always passes these kwargs explicitly, with + # value `None` when the user omits `mode` / `default_on` from the + # guardrail config. Coerce None (omitted) to the desired default + # while preserving any explicit value the caller did set -- + # in particular `default_on=False` if the user wants the guardrail + # off by default. + kwargs["event_hook"] = kwargs.get("event_hook") or GuardrailEventHooks.post_call + if kwargs.get("default_on") is None: + kwargs["default_on"] = True + super().__init__( + flush_lock=self.flush_lock, + **kwargs, + ) + + verbose_logger.debug("initializing rubrik logger") + + self.sampling_rate = 1.0 + rbrk_sampling_rate = os.getenv("RUBRIK_SAMPLING_RATE") + if rbrk_sampling_rate is not None: + try: + parsed_rate = float(rbrk_sampling_rate.strip()) + self.sampling_rate = max(0.0, min(1.0, parsed_rate)) + if parsed_rate != self.sampling_rate: + verbose_logger.warning( + f"RUBRIK_SAMPLING_RATE={parsed_rate} clamped to " + f"{self.sampling_rate}" + ) + except ValueError: + verbose_logger.warning( + f"Invalid RUBRIK_SAMPLING_RATE: {rbrk_sampling_rate!r}, using 1.0" + ) + + self.key = api_key or os.getenv("RUBRIK_API_KEY") + if not self.key: + verbose_logger.warning( + "Rubrik: No API key configured. Requests will be unauthenticated." + ) + _batch_size = os.getenv("RUBRIK_BATCH_SIZE") + + if _batch_size: + try: + self.batch_size = int(_batch_size) + except ValueError: + verbose_logger.warning( + f"Invalid RUBRIK_BATCH_SIZE: {_batch_size!r}, using default" + ) + + # Cap the in-memory retry queue so a Rubrik webhook outage cannot let + # authenticated traffic accumulate prompt/response payloads until the + # proxy runs out of memory. Once the cap is reached, oldest events are + # dropped to make room for fresh ones (drop-oldest backpressure). + self.max_queue_size = _MAX_QUEUE_SIZE + self._dropped_since_warning = 0 + self._last_drop_warning_time = 0.0 + + _webhook_url = api_base or os.getenv("RUBRIK_WEBHOOK_URL") + + if _webhook_url is None: + raise ValueError( + "Rubrik webhook URL not configured. " + "Set RUBRIK_WEBHOOK_URL or pass api_base." + ) + + _webhook_url = _webhook_url.rstrip("/").removesuffix("/v1") + self.tool_blocking_endpoint = f"{_webhook_url}{_WEBHOOK_PATH_TOOL_BLOCKING}" + self.logging_endpoint = f"{_webhook_url}{_WEBHOOK_PATH_LOGGING_BATCH}" + + self.async_httpx_client = get_async_httpx_client( + llm_provider=httpxSpecialProvider.LoggingCallback + ) + + self.tool_blocking_client = get_async_httpx_client( + llm_provider=httpxSpecialProvider.LoggingCallback, + params={"timeout": httpx.Timeout(5.0, connect=2.0)}, + ) + + self._headers: dict[str, str] = {"Content-Type": "application/json"} + if self.key: + self._headers["Authorization"] = f"Bearer {self.key}" + + # Periodic flush is started lazily on the first log event so that + # low-traffic deployments still get their batches drained even when the + # logger is instantiated outside a running event loop (sync init). + self._flush_task: Optional[asyncio.Task[Any]] = ( + self._start_periodic_flush_task() + ) + + def _start_periodic_flush_task(self) -> Optional[asyncio.Task[Any]]: + """Start the periodic flush task only when an event loop is already running.""" + try: + loop = asyncio.get_running_loop() + except RuntimeError: + verbose_logger.debug( + "Rubrik logger init: no running event loop, " + "periodic flush will start on first log event." + ) + return None + return loop.create_task(self.periodic_flush()) + + def _ensure_periodic_flush_task(self) -> None: + # Synchronous helper: in asyncio's cooperative model there is no await + # between the check and assignment, so two callers cannot race here. + if self._flush_task is None or self._flush_task.done(): + self._flush_task = self._start_periodic_flush_task() + + async def aclose(self): + """Close the dedicated HTTP clients used by this logger.""" + # Cancel the periodic flush task before closing the HTTP clients so + # the loop doesn't wake up and try to POST via a closed client. + if self._flush_task is not None and not self._flush_task.done(): + self._flush_task.cancel() + try: + await self._flush_task + except (asyncio.CancelledError, Exception): + pass + self._flush_task = None + await self.tool_blocking_client.close() + await self.async_httpx_client.close() + + # -- Guardrail hook -------------------------------------------------------- + + async def apply_guardrail( + self, + inputs: GenericGuardrailAPIInputs, + request_data: dict, + input_type: Literal["request", "response"], + logging_obj: Optional["LiteLLMLoggingObj"] = None, + ) -> GenericGuardrailAPIInputs: + """Validate tool calls against the blocking service (fail-open).""" + if input_type != "response": + return inputs + + tool_calls = inputs.get("tool_calls") + if not tool_calls: + return inputs + + try: + return await self._check_tool_calls( + inputs, tool_calls, request_data, logging_obj + ) + except ModifyResponseException: + raise + except _MalformedToolBlockingResponseError as e: + # Distinct from transient errors: the service responded but the + # payload was structurally invalid, which usually indicates a + # misconfigured webhook or a breaking change in its response + # format. Log loudly so operators notice their tool-blocking + # policy is not actually being enforced. + verbose_logger.critical( + "Tool blocking service returned a malformed response: %s. " + "Tool calls are NOT being checked -- verify the webhook " + "configuration. Returning original response unchanged.", + e, + exc_info=True, + ) + return inputs + except Exception as e: + verbose_logger.error( + f"Tool blocking hook failed: {e}. " + "Returning original response unchanged.", + exc_info=True, + ) + return inputs + + async def _check_tool_calls( + self, + inputs: GenericGuardrailAPIInputs, + tool_calls: Any, + request_data: dict, + logging_obj: Optional["LiteLLMLoggingObj"], + ) -> GenericGuardrailAPIInputs: + """Send tool calls to blocking service, raise if any are blocked.""" + message_tool_calls = self._normalize_tool_calls(tool_calls) + + call_details = ( + getattr(logging_obj, "model_call_details", {}) if logging_obj else {} + ) + response = request_data.get("response") + request_id = getattr(response, "id", None) if response else None + if logging_obj and not call_details: + verbose_logger.warning( + "Rubrik: logging_obj present but model_call_details is empty " + "-- request context will be missing" + ) + + response_data = self._build_tool_call_payload(message_tool_calls, request_id) + req_data = self._extract_request_data(call_details) + + service_response = await self._post_to_tool_blocking_service( + response_data, req_data + ) + blocked_explanation = self._extract_blocked_tools( + service_response, message_tool_calls + ) + + if blocked_explanation is not None: + model = self._resolve_model(request_data, call_details) + raise ModifyResponseException( + message=blocked_explanation, + model=model, + request_data=request_data, + guardrail_name=self.guardrail_name, + ) + + return inputs + + @staticmethod + def _normalize_tool_calls(tool_calls: Any) -> list[ChatCompletionMessageToolCall]: + """Convert tool_calls from inputs to ChatCompletionMessageToolCall objects.""" + result = [] + for tc in tool_calls: + if isinstance(tc, ChatCompletionMessageToolCall): + result.append(tc) + elif isinstance(tc, dict): + func = tc.get("function", {}) + result.append( + ChatCompletionMessageToolCall( + id=tc.get("id", ""), + type=tc.get("type", "function"), + function=Function( + name=func.get("name", ""), + arguments=func.get("arguments", ""), + ), + ) + ) + elif hasattr(tc, "id") and hasattr(tc, "function"): + result.append( + ChatCompletionMessageToolCall( + id=tc.id or "", + type=getattr(tc, "type", None) or "function", + function=tc.function, + ) + ) + else: + raise TypeError( + f"Cannot normalize tool_call of type {type(tc).__name__}" + ) + return result + + @staticmethod + def _build_tool_call_payload( + tool_calls: list[ChatCompletionMessageToolCall], + request_id: str | None, + ) -> dict[str, Any]: + """Build a full OpenAI ChatCompletion-format dict for the blocking service.""" + return { + "id": request_id or f"chatcmpl-{uuid.uuid4()}", + "object": "chat.completion", + "created": int(time.time()), + "model": "", + "choices": [ + { + "index": 0, + "message": { + "role": "assistant", + "content": None, + "tool_calls": [ + tc.model_dump(exclude_none=True) for tc in tool_calls + ], + }, + "finish_reason": "tool_calls", + } + ], + } + + @staticmethod + def _extract_request_data(call_details: dict[str, Any]) -> dict[str, Any]: + """Extract original request data from model_call_details.""" + if not call_details: + return {} + litellm_params = call_details.get("litellm_params", {}) or {} + return { + "messages": call_details.get("messages"), + "model": call_details.get("model"), + "proxy_server_request": RubrikLogger._sanitize_proxy_server_request( + litellm_params.get("proxy_server_request") + ), + } + + @staticmethod + def _sanitize_proxy_server_request(proxy_server_request: Any) -> Any: + """Allowlist only routing fields (``url``, ``method``) when forwarding + ``proxy_server_request`` to the external Rubrik webhook, dropping + inbound ``headers`` (Authorization, Cookie, x-api-key, ...) and the raw + request ``body`` so proxy credentials are not exfiltrated.""" + if not isinstance(proxy_server_request, dict): + return proxy_server_request + return { + key: proxy_server_request[key] + for key in ("url", "method") + if key in proxy_server_request + } + + @staticmethod + def _resolve_model( + request_data: dict[str, Any], call_details: dict[str, Any] + ) -> str: + """Get the model name for the ModifyResponseException.""" + response = request_data.get("response") + if response and hasattr(response, "model"): + return response.model or "unknown" + return call_details.get("model", "unknown") + + # -- Logging hooks --------------------------------------------------------- + + async def _prepare_log_payload( + self, kwargs: dict, event_type: str + ) -> StandardLoggingPayload | None: + """Shared logic for success and failure logging.""" + if random.random() > self.sampling_rate: + verbose_logger.debug( + f"Skipping Rubrik {event_type} logging " + f"(sampling_rate={self.sampling_rate})" + ) + return None + + # Deep-copy so mutations don't affect other callbacks sharing this object + standard_logging_payload: StandardLoggingPayload = safe_deep_copy( + kwargs["standard_logging_object"] + ) + + # For Anthropic /v1/messages requests, LiteLLM creates a separate + # ModelResponse (with a generated chatcmpl-* id) for logging, which + # differs from the original Anthropic msg-* id on the response dict. + # Normalize to litellm_call_id so that the logging and tool-blocking + # endpoints see the same request identifier. + litellm_params = kwargs.get("litellm_params", {}) or {} + proxy_request = litellm_params.get("proxy_server_request", {}) or {} + url_path = urllib.parse.urlparse(proxy_request.get("url", "")).path + if url_path.endswith(_ENDPOINT_ANTHROPIC_MESSAGES): + _litellm_call_id = kwargs.get("litellm_call_id") + if _litellm_call_id: + standard_logging_payload["id"] = _litellm_call_id # type: ignore[literal-required] + + if "system" in kwargs: + system_prompt_msg_list = kwargs["system"] + try: + if system_prompt_msg_list: + system_scaffold = { + "role": "system", + "content": system_prompt_msg_list, + } + if isinstance(standard_logging_payload["messages"], list): + standard_logging_payload["messages"].insert(0, system_scaffold) + elif isinstance(standard_logging_payload["messages"], (dict, str)): + standard_logging_payload["messages"] = [ + system_scaffold, + standard_logging_payload["messages"], + ] + except Exception as e: + verbose_logger.warning( + f"Rubrik: failed to prepend system prompt: {e}", + exc_info=True, + ) + + return standard_logging_payload + + async def _enqueue_log_event(self, kwargs: dict, event_type: str): + try: + self._ensure_periodic_flush_task() + payload = await self._prepare_log_payload(kwargs, event_type) + if payload is None: + return + + self.log_queue.append(payload) + self._enforce_max_queue_size() + + if len(self.log_queue) >= self.batch_size: + await self.flush_queue() + except Exception as e: + verbose_logger.error( + f"Rubrik {event_type} logging hook failed: {e}. " + "Skipping logging for this event.", + exc_info=True, + ) + + def _enforce_max_queue_size(self) -> None: + overflow = len(self.log_queue) - self.max_queue_size + if overflow <= 0: + return + del self.log_queue[:overflow] + self._dropped_since_warning += overflow + now = time.time() + if now - self._last_drop_warning_time >= _DROP_WARNING_INTERVAL_SECONDS: + verbose_logger.warning( + "Rubrik: log queue exceeded max_queue_size=%s; dropped %s " + "oldest events since the last warning. The Rubrik webhook may " + "be unhealthy or undersized for current traffic.", + self.max_queue_size, + self._dropped_since_warning, + ) + self._dropped_since_warning = 0 + self._last_drop_warning_time = now + + async def async_log_success_event(self, kwargs, response_obj, start_time, end_time): + await self._enqueue_log_event(kwargs, "success") + + async def async_log_failure_event(self, kwargs, response_obj, start_time, end_time): + await self._enqueue_log_event(kwargs, "failure") + + # -- Batch logging --------------------------------------------------------- + + async def _log_batch_to_rubrik(self, data): + # NOTE: this method intentionally re-raises on failure so the parent + # CustomBatchLogger.flush_queue keeps the unsent events in the queue + # for the next flush attempt instead of silently dropping them. + try: + response = await self.async_httpx_client.post( + url=self.logging_endpoint, + json=data, + headers=self._headers, + ) + response.raise_for_status() + except httpx.HTTPStatusError as e: + verbose_logger.exception( + f"Rubrik HTTP Error: {e.response.status_code} - {e.response.text}" + ) + raise + except Exception: + verbose_logger.exception("Rubrik Layer Error") + raise + + async def async_send_batch(self): + """Handles sending batches of responses to Rubrik. + + Note: the canonical flush path is :meth:`flush_queue`, which takes a + single snapshot used for both sending and queue draining. This method + is kept for direct callers / tests; it intentionally does NOT remove + events from the queue. + """ + if not self.log_queue: + return + + log_queue_snapshot = list(self.log_queue) + verbose_logger.debug( + "Rubrik: Flushing batch of %s events", len(log_queue_snapshot) + ) + await self._log_batch_to_rubrik( + data=log_queue_snapshot, + ) + + async def flush_queue(self): + """Snapshot, send, and drain in one consistent step. + + Overrides the base implementation so the same snapshot drives both + the HTTP send and the queue truncation. This avoids the subtle + coupling where the base class captures `len(self.log_queue)` + separately from the snapshot taken inside `async_send_batch`, + which could otherwise drift in a future refactor and cause + duplicate deliveries to Rubrik. + """ + if self.flush_lock is None: + return + + async with self.flush_lock: + if not self.log_queue: + return + snapshot = list(self.log_queue) + verbose_logger.debug("Rubrik: Flushing batch of %s events", len(snapshot)) + try: + await self._log_batch_to_rubrik(data=snapshot) + except Exception: + # Already logged with traceback inside _log_batch_to_rubrik. + # Preserve the in-flight events for retry on the next flush. + return + del self.log_queue[: len(snapshot)] + self.last_flush_time = time.time() + + # -- Tool blocking service ------------------------------------------------- + + async def _post_to_tool_blocking_service( + self, + response_data: dict[str, Any], + request_data: dict[str, Any], + ) -> dict[str, Any]: + """Post a payload to the tool blocking service and return the response. + + Args: + response_data: The OpenAI-formatted response payload to send. + request_data: Original LLM request data to include alongside + the response for additional context. Empty dict if unavailable. + + Raises: + Exception: If the service is unavailable or returns an error. + """ + envelope = { + "request": request_data, + "response": response_data, + } + verbose_logger.debug( + f"Sending request to tool blocking service: " + f"{self.tool_blocking_endpoint}" + ) + http_response = await self.tool_blocking_client.post( + self.tool_blocking_endpoint, + json=envelope, + headers=self._headers, + ) + http_response.raise_for_status() + result: dict[str, Any] = http_response.json() + return result + + @staticmethod + def _extract_blocked_tools( + service_response: dict[str, Any], + all_tool_calls: list[ChatCompletionMessageToolCall], + ) -> Optional[str]: + """Return the blocking explanation if any tool calls were blocked. + + Compares the service response (which contains only allowed tools) against + the full set of tool calls. Returns ``None`` if all tools are allowed, or + the explanation string (prefixed with newlines) otherwise. + + Expects service_response in OpenAI chat completion format: + {"choices": [{"message": {"tool_calls": [...], "content": "..."}}]} + """ + choices = service_response.get("choices", []) + if not choices: + raise _MalformedToolBlockingResponseError( + "Tool blocking service returned empty response" + ) + + message = choices[0].get("message", {}) + returned_tool_calls = message.get("tool_calls") or [] + blocking_explanation = message.get("content", "") + + allowed_id_counts: Counter = Counter( + tc["id"] + for tc in returned_tool_calls + if isinstance(tc, dict) and tc.get("id") + ) + required_id_counts: Counter = Counter(tc.id for tc in all_tool_calls if tc.id) + + all_allowed = len(returned_tool_calls) >= len(all_tool_calls) and all( + allowed_id_counts.get(tc_id, 0) >= count + for tc_id, count in required_id_counts.items() + ) + + if all_allowed: + return None + + explanation = blocking_explanation or "Tool call blocked by policy." + return f"\n\n{explanation}" diff --git a/litellm/integrations/s3_v2.py b/litellm/integrations/s3_v2.py index 332e84dd07d..4ed8a809a13 100644 --- a/litellm/integrations/s3_v2.py +++ b/litellm/integrations/s3_v2.py @@ -1,8 +1,8 @@ """ s3 Bucket Logging Integration -async_log_success_event: Processes the event, stores it in memory for DEFAULT_S3_FLUSH_INTERVAL_SECONDS seconds or until DEFAULT_S3_BATCH_SIZE and then flushes to s3 -async_log_failure_event: Processes the event, stores it in memory for DEFAULT_S3_FLUSH_INTERVAL_SECONDS seconds or until DEFAULT_S3_BATCH_SIZE and then flushes to s3 +async_log_success_event: Processes the event, stores it in memory for DEFAULT_S3_FLUSH_INTERVAL_SECONDS seconds or until DEFAULT_S3_BATCH_SIZE and then flushes to s3 +async_log_failure_event: Processes the event, stores it in memory for DEFAULT_S3_FLUSH_INTERVAL_SECONDS seconds or until DEFAULT_S3_BATCH_SIZE and then flushes to s3 NOTE 1: S3 does not provide a BATCH PUT API endpoint, so we create tasks to upload each element individually """ diff --git a/litellm/integrations/websearch_interception/handler.py b/litellm/integrations/websearch_interception/handler.py index 41618c72627..37528e7dcd5 100644 --- a/litellm/integrations/websearch_interception/handler.py +++ b/litellm/integrations/websearch_interception/handler.py @@ -19,12 +19,14 @@ from litellm.integrations.custom_logger import CustomLogger from litellm.integrations.websearch_interception.tools import ( get_litellm_web_search_tool, get_litellm_web_search_tool_openai, + is_anthropic_native_web_search_tool, is_web_search_tool, is_web_search_tool_chat_completion, ) from litellm.integrations.websearch_interception.transformation import ( WebSearchTransformation, ) +from litellm.llms.base_llm.search.transformation import SearchResponse from litellm.types.integrations.websearch_interception import ( WebSearchInterceptionConfig, ) @@ -36,6 +38,16 @@ from litellm.types.llms.openai import AllMessageValues from litellm.types.utils import LlmProviders from litellm.utils import ProviderConfigManager +# Key used to flag, on per-request kwargs, that the originating client sent +# an Anthropic-native ``web_search_*`` tool — meaning the final response +# should include ``web_search_tool_result`` content blocks so the client +# (e.g. Claude Desktop's citations panel) can render sources. +WEBSEARCH_EMIT_NATIVE_BLOCKS_KEY = "_websearch_interception_emit_native_blocks" + +# Key on ``AgenticLoopPlan.metadata`` carrying the list of pre-built +# ``web_search_tool_result`` blocks to inject into the final response. +WEBSEARCH_NATIVE_BLOCKS_METADATA_KEY = "websearch_native_blocks" + class WebSearchInterceptionLogger(CustomLogger): """ @@ -152,22 +164,55 @@ class WebSearchInterceptionLogger(CustomLogger): f"(provider={provider_str}, query='{query}')" ) - # Execute search + # Native clients (Claude Desktop / Cowork / Anthropic SDK) make a + # standalone /v1/messages sub-request just for the search, and they + # expect the response in native shape with server_tool_use + + # web_search_tool_result content blocks so the citations panel can + # render. The agentic-loop post-hook never fires on this path because + # there is no model call — emit the native blocks here instead. + native_tool = next( + (t for t in tools if is_anthropic_native_web_search_tool(t)), + None, + ) + + # Execute search — keep the structured SearchResponse so the native + # block can carry per-result url/title/page_age. try: - search_result_text = await self._execute_search(query) + search_result_text, structured = await self._execute_search(query) except Exception as e: verbose_logger.error( f"WebSearchInterception: Short-circuit search failed: {e}" ) - search_result_text = f"Search failed: {e}" + search_result_text, structured = f"Search failed: {e}", None + + content: List[Dict[str, Any]] = [] + if native_tool is not None: + tool_use_id = f"srvtoolu_{uuid.uuid4().hex}" + tool_name = native_tool.get("name") or "web_search" + content.append( + { + "type": "server_tool_use", + "id": tool_use_id, + "name": tool_name, + "input": {"query": query}, + } + ) + content.append( + WebSearchTransformation.build_web_search_tool_result_block( + tool_use_id=tool_use_id, + search_response=structured, + ) + ) + # Keep the text block so non-native short-circuit callers (Claude Code, + # github_copilot, etc.) see the same payload they always have. + content.append({"type": "text", "text": search_result_text}) - # Build synthetic Anthropic response response: Dict[str, Any] = { "id": f"msg_{str(uuid.uuid4())}", "type": "message", "role": "assistant", "model": model, - "content": [{"type": "text", "text": search_result_text}], + "content": content, "stop_reason": "end_turn", "stop_sequence": None, "usage": {"input_tokens": 0, "output_tokens": 0}, @@ -175,7 +220,8 @@ class WebSearchInterceptionLogger(CustomLogger): verbose_logger.debug( "WebSearchInterception: Short-circuit search completed, " - f"returning synthetic response ({len(search_result_text)} chars)" + f"returning synthetic response ({len(search_result_text)} chars, " + f"native_blocks={native_tool is not None})" ) return response @@ -219,6 +265,14 @@ class WebSearchInterceptionLogger(CustomLogger): "WebSearchInterception: Converting native web_search tools to LiteLLM standard" ) + # If the client sent an Anthropic-native web_search_* tool, mark the + # request so the agentic loop emits native web_search_tool_result + # blocks in the final response (matches async_pre_request_hook). This + # deployment hook fires before async_pre_request_hook on some paths, + # so flagging here ensures the signal isn't lost regardless of order. + if any(is_anthropic_native_web_search_tool(t) for t in tools): + kwargs[WEBSEARCH_EMIT_NATIVE_BLOCKS_KEY] = True + # Convert native/custom web_search tools to LiteLLM standard converted_tools = [] for tool in tools: @@ -342,6 +396,14 @@ class WebSearchInterceptionLogger(CustomLogger): f"WebSearchInterception: Pre-request hook triggered for provider={custom_llm_provider}" ) + # If the client sent an Anthropic-native web_search_* tool, mark the + # request so the agentic loop emits native web_search_tool_result + # blocks in the final response (for citations panels, etc.). The flag + # is read by async_build_agentic_loop_plan; the leading underscore + # prefix ensures it is stripped before the follow-up call kwargs. + if any(is_anthropic_native_web_search_tool(t) for t in tools): + kwargs[WEBSEARCH_EMIT_NATIVE_BLOCKS_KEY] = True + # Convert native web search tools to LiteLLM standard converted_tools = [] for tool in tools: @@ -591,7 +653,7 @@ class WebSearchInterceptionLogger(CustomLogger): ) -> AgenticLoopPlan: tool_calls = tools["tool_calls"] thinking_blocks = tools.get("thinking_blocks", []) - request_patch = await self._build_anthropic_request_patch( + request_patch, structured_results = await self._build_anthropic_request_patch( model=model, messages=messages, tool_calls=tool_calls, @@ -600,12 +662,92 @@ class WebSearchInterceptionLogger(CustomLogger): logging_obj=logging_obj, kwargs=kwargs, ) + + metadata: Dict[str, Any] = { + "tool_type": "websearch", + "response_format": "anthropic", + } + + # If the client request originally carried a native web_search_* tool, + # pre-build the Anthropic-native ``web_search_tool_result`` blocks now + # (while we still have the structured SearchResponse list) and stash + # them on plan metadata for the post-hook to inject. + if kwargs.get(WEBSEARCH_EMIT_NATIVE_BLOCKS_KEY): + metadata[WEBSEARCH_NATIVE_BLOCKS_METADATA_KEY] = ( + self._build_native_result_blocks( + tool_calls=tool_calls, + structured_results=structured_results, + ) + ) + return AgenticLoopPlan( run_agentic_loop=True, request_patch=request_patch, - metadata={"tool_type": "websearch", "response_format": "anthropic"}, + metadata=metadata, ) + async def async_post_agentic_loop_response_hook( + self, + response: Any, + plan: AgenticLoopPlan, + kwargs: Dict, + ) -> Any: + """ + Inject Anthropic-native ``web_search_tool_result`` blocks into the + final response when the originating client used a native + ``web_search_*`` tool. + + See ``WebSearchTransformation.build_web_search_tool_result_block`` for + the block shape. The blocks are prepended to ``response.content`` so + Anthropic-native clients (Claude Desktop, the Anthropic SDK) can + render citations / sources alongside the model's textual reply. + """ + native_blocks = plan.metadata.get(WEBSEARCH_NATIVE_BLOCKS_METADATA_KEY) + if not native_blocks: + return response + return self._inject_native_blocks(response, native_blocks) + + @staticmethod + def _build_native_result_blocks( + tool_calls: List[Dict], + structured_results: List[Optional[SearchResponse]], + ) -> List[Dict[str, Any]]: + """Build one ``web_search_tool_result`` block per tool_call.""" + blocks: List[Dict[str, Any]] = [] + for i, tool_call in enumerate(tool_calls): + tool_use_id = tool_call.get("id") or "" + structured = structured_results[i] if i < len(structured_results) else None + blocks.append( + WebSearchTransformation.build_web_search_tool_result_block( + tool_use_id=tool_use_id, + search_response=structured, + ) + ) + return blocks + + @staticmethod + def _inject_native_blocks( + response: Any, native_blocks: List[Dict[str, Any]] + ) -> Any: + """Prepend native blocks to response content, dict or object form.""" + if not native_blocks: + return response + if isinstance(response, dict): + existing = response.get("content") or [] + response["content"] = list(native_blocks) + list(existing) + return response + existing = getattr(response, "content", None) or [] + try: + response.content = list(native_blocks) + list(existing) + except (AttributeError, TypeError): + # Object refused write — fall through and leave the response + # untouched rather than crash the request. + verbose_logger.debug( + "WebSearchInterception: could not inject native blocks into " + f"response of type {type(response).__name__}" + ) + return response + async def async_run_chat_completion_agentic_loop( self, tools: Dict, @@ -733,7 +875,7 @@ class WebSearchInterceptionLogger(CustomLogger): kwargs: Dict, ) -> Any: """Legacy path: execute search + build patch + run follow-up call.""" - request_patch = await self._build_anthropic_request_patch( + request_patch, structured_results = await self._build_anthropic_request_patch( model=model, messages=messages, tool_calls=tool_calls, @@ -755,7 +897,7 @@ class WebSearchInterceptionLogger(CustomLogger): if max_tokens is None: max_tokens = cast(int, kwargs.get("max_tokens", 1024)) - return await anthropic_messages.acreate( + response = await anthropic_messages.acreate( max_tokens=max_tokens, messages=request_patch.messages, model=request_patch.model or model, @@ -763,6 +905,18 @@ class WebSearchInterceptionLogger(CustomLogger): **request_patch.kwargs, ) + # Legacy path: the new path goes through the typed plan + core + # dispatcher which runs the post-hook automatically. Mirror the + # native-block injection here so both paths behave identically. + if kwargs.get(WEBSEARCH_EMIT_NATIVE_BLOCKS_KEY): + native_blocks = self._build_native_result_blocks( + tool_calls=tool_calls, + structured_results=structured_results, + ) + response = self._inject_native_blocks(response, native_blocks) + + return response + async def _build_anthropic_request_patch( self, model: str, @@ -772,8 +926,16 @@ class WebSearchInterceptionLogger(CustomLogger): anthropic_messages_optional_request_params: Dict, logging_obj: Any, kwargs: Dict, - ) -> AgenticLoopRequestPatch: - """Execute litellm.search() and build follow-up request patch.""" + ) -> Tuple[AgenticLoopRequestPatch, List[Optional[SearchResponse]]]: + """ + Execute litellm.search() and build follow-up request patch. + + Returns the patch alongside the parallel list of structured + ``SearchResponse`` objects (one per tool_call, ``None`` when the + search failed or the tool_call had no query). The caller uses these + to optionally build Anthropic-native ``web_search_tool_result`` + content blocks for the final response. + """ # Extract search queries from tool_use blocks search_tasks = [] @@ -797,23 +959,38 @@ class WebSearchInterceptionLogger(CustomLogger): ) search_results = await asyncio.gather(*search_tasks, return_exceptions=True) - # Handle any exceptions in search results + # Split the gathered (text, structured) tuples into two parallel lists. + # The text list feeds the follow-up model call; the structured list + # is returned to the caller for native-block emission. final_search_results: List[str] = [] + structured_results: List[Optional[SearchResponse]] = [] for i, result in enumerate(search_results): if isinstance(result, Exception): verbose_logger.error( f"WebSearchInterception: Search {i} failed with error: {str(result)}" ) final_search_results.append(f"Search failed: {str(result)}") - elif isinstance(result, str): - # Explicitly cast to str for type checker - final_search_results.append(cast(str, result)) + structured_results.append(None) + elif isinstance(result, tuple) and len(result) == 2: + text_value, structured_value = result + final_search_results.append( + cast(str, text_value) + if isinstance(text_value, str) + else str(text_value) + ) + structured_results.append( + structured_value + if isinstance(structured_value, SearchResponse) + else None + ) else: - # Should never happen, but handle for type safety + # Defensive: legacy callers / unexpected shape — preserve text, + # drop structure. verbose_logger.debug( f"WebSearchInterception: Unexpected result type {type(result)} at index {i}" ) final_search_results.append(str(result)) + structured_results.append(None) # Build assistant and user messages using transformation assistant_message, user_message = WebSearchTransformation.transform_response( @@ -859,16 +1036,26 @@ class WebSearchInterceptionLogger(CustomLogger): len(follow_up_messages), len(final_search_results), ) - return AgenticLoopRequestPatch( + patch = AgenticLoopRequestPatch( model=full_model_name, messages=follow_up_messages, max_tokens=max_tokens, optional_params=optional_params_without_max_tokens, kwargs=kwargs_for_followup, ) + return patch, structured_results - async def _execute_search(self, query: str) -> str: - """Execute a single web search using router's search tools""" + async def _execute_search(self, query: str) -> Tuple[str, Optional[SearchResponse]]: + """ + Execute a single web search using router's search tools. + + Returns both the formatted text (fed back to the model in the follow-up + call) and the structured ``SearchResponse`` (preserved so callers can + build Anthropic-native ``web_search_tool_result`` blocks for clients + that requested a native ``web_search_*`` tool). The structured value + is None on the failure path so callers can still emit an empty result + block rather than dropping the search entirely. + """ try: # Import router from proxy_server try: @@ -934,7 +1121,7 @@ class WebSearchInterceptionLogger(CustomLogger): verbose_logger.debug( f"WebSearchInterception: Search completed for '{query}', got {len(search_result_text)} chars" ) - return search_result_text + return search_result_text, result except Exception as e: verbose_logger.error( f"WebSearchInterception: Search failed for '{query}': {str(e)}" @@ -1015,7 +1202,8 @@ class WebSearchInterceptionLogger(CustomLogger): ) search_results = await asyncio.gather(*search_tasks, return_exceptions=True) - # Handle any exceptions in search results + # Chat-completion path only needs text — OpenAI tool_result format + # has no equivalent of Anthropic's web_search_tool_result block. final_search_results: List[str] = [] for i, result in enumerate(search_results): if isinstance(result, Exception): @@ -1023,8 +1211,13 @@ class WebSearchInterceptionLogger(CustomLogger): f"WebSearchInterception: Search {i} failed with error: {str(result)}" ) final_search_results.append(f"Search failed: {str(result)}") - elif isinstance(result, str): - final_search_results.append(cast(str, result)) + elif isinstance(result, tuple) and len(result) == 2: + text_value, _ = result + final_search_results.append( + cast(str, text_value) + if isinstance(text_value, str) + else str(text_value) + ) else: verbose_logger.debug( f"WebSearchInterception: Unexpected result type {type(result)} at index {i}" @@ -1112,9 +1305,11 @@ class WebSearchInterceptionLogger(CustomLogger): kwargs=kwargs_for_followup, ) - async def _create_empty_search_result(self) -> str: + async def _create_empty_search_result( + self, + ) -> Tuple[str, Optional[SearchResponse]]: """Create an empty search result for tool calls without queries""" - return "No search query provided" + return "No search query provided", None @staticmethod def initialize_from_proxy_config( diff --git a/litellm/integrations/websearch_interception/tools.py b/litellm/integrations/websearch_interception/tools.py index e373b64cdda..b29372af9ed 100644 --- a/litellm/integrations/websearch_interception/tools.py +++ b/litellm/integrations/websearch_interception/tools.py @@ -126,6 +126,27 @@ def is_web_search_tool_chat_completion(tool: Dict[str, Any]) -> bool: return False +def is_anthropic_native_web_search_tool(tool: Dict[str, Any]) -> bool: + """ + Check if a tool is an Anthropic-native ``web_search_*`` tool. + + Native clients (Anthropic SDK, Claude Desktop, Anthropic Console) send + tools like ``{"type": "web_search_20250305", "name": "web_search"}`` and + expect the response to contain ``web_search_tool_result`` content blocks + so that citations can be rendered. This helper identifies that contract + so the agentic loop can emit native-format blocks for those clients + without affecting clients that send the LiteLLM standard tool. + + Returns False for the LiteLLM standard tool (``litellm_web_search``), + the OpenAI-shaped variant, the bare ``WebSearch`` legacy name, and the + bare ``web_search`` name (Claude Code style). + """ + tool_type = tool.get("type", "") + if not isinstance(tool_type, str): + return False + return tool_type.startswith("web_search_") and tool_type != "function" + + def is_web_search_tool(tool: Dict[str, Any]) -> bool: """ Check if a tool is a web search tool (native or LiteLLM standard). @@ -135,7 +156,22 @@ def is_web_search_tool(tool: Dict[str, Any]) -> bool: - OpenAI format: type == "function" with function.name == "litellm_web_search" - Anthropic native: type starts with "web_search_" (e.g., "web_search_20250305") - Claude Code: name == "web_search" with a type field - - Custom: name == "WebSearch" (legacy format) + - Custom: name == "WebSearch" (legacy interception marker — only matched + when input_schema is absent; see note below) + + Note on the legacy ``WebSearch`` name: + Clients like Claude Desktop / Cowork ship a *client-side* tool called + ``WebSearch`` (a fully-formed Anthropic client tool with its own + ``input_schema``) that they handle themselves. Treating that as our + interception marker hijacks it server-side and the client's own tool + handler never fires — which means Cowork's separate native + ``web_search_20250305`` sub-request (where citation data actually + flows) never gets made. + + Real Anthropic client tools always carry an ``input_schema`` (the API + rejects them otherwise), so a bare ``{name: "WebSearch"}`` with no + schema is the only thing that could be a legacy interception marker. + Gate the match on schema absence to keep both groups working. Args: tool: Tool dictionary to check @@ -152,6 +188,10 @@ def is_web_search_tool(tool: Dict[str, Any]) -> bool: True >>> is_web_search_tool({"name": "calculator"}) False + >>> is_web_search_tool({"name": "WebSearch"}) # legacy interception marker + True + >>> is_web_search_tool({"name": "WebSearch", "input_schema": {"type": "object"}}) # Cowork client tool + False """ tool_name = tool.get("name", "") tool_type = tool.get("type", "") @@ -175,8 +215,9 @@ def is_web_search_tool(tool: Dict[str, Any]) -> bool: if tool_name == "web_search" and tool_type: return True - # Check for legacy WebSearch format - if tool_name == "WebSearch": + # Legacy "WebSearch" interception marker — only when no schema is + # present, so real client-side WebSearch tools (Cowork) pass through. + if tool_name == "WebSearch" and "input_schema" not in tool: return True return False diff --git a/litellm/integrations/websearch_interception/transformation.py b/litellm/integrations/websearch_interception/transformation.py index 00d4829ad39..9c20a3f6c77 100644 --- a/litellm/integrations/websearch_interception/transformation.py +++ b/litellm/integrations/websearch_interception/transformation.py @@ -100,11 +100,14 @@ class WebSearchTransformation: block_id = getattr(block, "id", None) block_input = getattr(block, "input", {}) - # Check for LiteLLM standard or legacy web search tools - # Handles: litellm_web_search, WebSearch, web_search + # Detect tool_use blocks that came from interception. After + # pre-request conversion the model always sees + # ``litellm_web_search``; the bare ``web_search`` entry handles + # callers that bypass our pre-request hooks (e.g. direct + # litellm.acompletion). "WebSearch" is intentionally omitted — + # see is_web_search_tool for the Cowork rationale. if block_type == "tool_use" and block_name in ( LITELLM_WEB_SEARCH_TOOL_NAME, - "WebSearch", "web_search", ): # Convert to dict for easier handling @@ -190,10 +193,12 @@ class WebSearchTransformation: getattr(function, "arguments", None) if function else None ) - # Check for LiteLLM standard or legacy web search tools + # Detect function-style web search tool_calls. ``WebSearch`` is + # intentionally omitted — see is_web_search_tool for the Cowork + # rationale (clients ship their own client-side ``WebSearch`` and + # we must not hijack it). if tool_type == "function" and function_name in ( LITELLM_WEB_SEARCH_TOOL_NAME, - "WebSearch", "web_search", ): # Parse arguments (might be JSON string) @@ -350,6 +355,57 @@ class WebSearchTransformation: return assistant_message, tool_messages + @staticmethod + def build_web_search_tool_result_block( + tool_use_id: str, + search_response: Optional[SearchResponse], + ) -> Dict[str, Any]: + """ + Build an Anthropic-native ``web_search_tool_result`` content block. + + Native Anthropic clients (Claude Desktop, the Anthropic SDK, the + Anthropic Console) expect search-tool results to be returned as + structured ``web_search_tool_result`` blocks so that citations and + source links can be rendered. The agentic loop currently feeds the + model a flat text blob in the follow-up call (which is correct — the + model needs readable evidence). This helper produces the *additional* + block that should accompany the model's text reply when the original + request used a native ``web_search_*`` tool. + + Spec reference: + https://docs.anthropic.com/en/api/web-search-tool + + Args: + tool_use_id: The ``tool_use_id`` the model emitted on the first + turn. Must match exactly so the client can pair the result + with its tool_use block. + search_response: Structured ``SearchResponse`` from + ``litellm.asearch()``. If None or empty, the block is still + emitted with an empty result list (signals "search ran, no + results" rather than "search did not run"). + """ + items: List[Dict[str, Any]] = [] + if search_response is not None: + results = getattr(search_response, "results", None) or [] + for r in results: + url = getattr(r, "url", "") or "" + title = getattr(r, "title", "") or "" + page_age = getattr(r, "date", None) or getattr(r, "last_updated", None) + items.append( + { + "type": "web_search_result", + "url": url, + "title": title, + "page_age": page_age, + "encrypted_content": "", + } + ) + return { + "type": "web_search_tool_result", + "tool_use_id": tool_use_id, + "content": items, + } + @staticmethod def format_search_response(result: SearchResponse) -> str: """ diff --git a/litellm/interactions/__init__.py b/litellm/interactions/__init__.py index e1125b649a6..ed01462cba6 100644 --- a/litellm/interactions/__init__.py +++ b/litellm/interactions/__init__.py @@ -5,31 +5,40 @@ This module provides SDK methods for Google's Interactions API. Usage: import litellm - + # Create an interaction with a model response = litellm.interactions.create( model="gemini-2.5-flash", input="Hello, how are you?" ) - + # Create an interaction with an agent response = litellm.interactions.create( agent="deep-research-pro-preview-12-2025", input="Research the current state of cancer research" ) - + # Async version response = await litellm.interactions.acreate(...) - + # Get an interaction response = litellm.interactions.get(interaction_id="...") - + # Delete an interaction result = litellm.interactions.delete(interaction_id="...") - + # Cancel an interaction result = litellm.interactions.cancel(interaction_id="...") + # Create a managed agent on the provider side + result = litellm.interactions.agents.create( + name="waverunner", + custom_llm_provider="gemini", + api_key="...", + base_agent="gemini-2.5-flash", + instructions="You are a helpful assistant.", + ) + Methods: - create(): Sync create interaction - acreate(): Async create interaction @@ -39,8 +48,12 @@ Methods: - adelete(): Async delete interaction - cancel(): Sync cancel interaction - acancel(): Async cancel interaction + +Sub-modules: +- agents: Provider-side agent creation (litellm.interactions.agents.create) """ +from litellm.interactions import agents from litellm.interactions.main import ( acancel, acreate, @@ -65,4 +78,6 @@ __all__ = [ # Cancel "cancel", "acancel", + # Sub-modules + "agents", ] diff --git a/litellm/interactions/agents/__init__.py b/litellm/interactions/agents/__init__.py new file mode 100644 index 00000000000..711a54fdcbb --- /dev/null +++ b/litellm/interactions/agents/__init__.py @@ -0,0 +1,39 @@ +""" +litellm.interactions.agents + +Full CRUD SDK for provider-side managed agents (e.g. Gemini v1beta/agents). + + litellm.interactions.agents.create(name=..., ...) + litellm.interactions.agents.list(api_key=...) + litellm.interactions.agents.get(name=..., ...) + litellm.interactions.agents.delete(name=..., ...) + litellm.interactions.agents.list_versions(name=..., ...) + +Async counterparts: acreate, alist, aget, adelete, alist_versions +""" + +from litellm.interactions.agents.main import ( + acreate, + adelete, + aget, + alist, + alist_versions, + create, + delete, + get, + list, + list_versions, +) + +__all__ = [ + "create", + "acreate", + "list", + "alist", + "get", + "aget", + "delete", + "adelete", + "list_versions", + "alist_versions", +] diff --git a/litellm/interactions/agents/http_handler.py b/litellm/interactions/agents/http_handler.py new file mode 100644 index 00000000000..d45ca6f4346 --- /dev/null +++ b/litellm/interactions/agents/http_handler.py @@ -0,0 +1,478 @@ +""" +HTTP handler for the Agents API. + +Extends InteractionsHTTPHandler so that the shared HTTP infrastructure +(_handle_error, _sync_client, _async_client) is reused rather than +duplicated. BaseAgentsAPIConfig stays as pure transform code. +""" + +from typing import Any, Coroutine, Dict, Optional, Union + +import httpx + +from litellm.constants import request_timeout +from litellm.interactions.http_handler import InteractionsHTTPHandler +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.llms.base_llm.agents.transformation import BaseAgentsAPIConfig +from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler +from litellm.types.agents import ( + AgentCreateResponse, + AgentDeleteResult, + AgentListResponse, + AgentVersionsResponse, +) +from litellm.types.router import GenericLiteLLMParams + + +class AgentsHTTPHandler(InteractionsHTTPHandler): + """HTTP handler for Agents API CRUD requests.""" + + # ------------------------------------------------------------------ # + # CREATE # + # ------------------------------------------------------------------ # + + def create_agent( + self, + agents_api_config: BaseAgentsAPIConfig, + name: str, + litellm_params: GenericLiteLLMParams, + logging_obj: LiteLLMLoggingObj, + extra_headers: Optional[Dict[str, Any]] = None, + extra_body: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + client: Optional[HTTPHandler] = None, + _is_async: bool = False, + ) -> Union[AgentCreateResponse, Coroutine[Any, Any, AgentCreateResponse]]: + if _is_async: + return self.async_create_agent( + agents_api_config=agents_api_config, + name=name, + litellm_params=litellm_params, + logging_obj=logging_obj, + extra_headers=extra_headers, + extra_body=extra_body, + timeout=timeout, + ) + + sync_httpx_client = self._sync_client(litellm_params, client) + headers = agents_api_config.validate_environment( + headers=extra_headers or {}, litellm_params=dict(litellm_params) + ) + url = agents_api_config.get_complete_url( + api_base=litellm_params.get("api_base"), + litellm_params=dict(litellm_params), + ) + data = agents_api_config.transform_create_request( + name=name, litellm_params=dict(litellm_params) + ) + if extra_body: + data.update(extra_body) + + logging_obj.pre_call( + input=name, + api_key="", + additional_args={ + "complete_input_dict": data, + "api_base": url, + "headers": headers, + }, + ) + try: + response = sync_httpx_client.post( + url=url, headers=headers, json=data, timeout=timeout or request_timeout + ) + except Exception as e: + raise self._handle_error(e=e, provider_config=agents_api_config) + + logging_obj.post_call( + original_response=response.text, + additional_args={"complete_input_dict": data}, + ) + return agents_api_config.transform_create_response( + raw_response=response, name=name + ) + + async def async_create_agent( + self, + agents_api_config: BaseAgentsAPIConfig, + name: str, + litellm_params: GenericLiteLLMParams, + logging_obj: LiteLLMLoggingObj, + extra_headers: Optional[Dict[str, Any]] = None, + extra_body: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + client: Optional[AsyncHTTPHandler] = None, + ) -> AgentCreateResponse: + async_httpx_client = self._async_client(litellm_params, client) + headers = agents_api_config.validate_environment( + headers=extra_headers or {}, litellm_params=dict(litellm_params) + ) + url = agents_api_config.get_complete_url( + api_base=litellm_params.get("api_base"), + litellm_params=dict(litellm_params), + ) + data = agents_api_config.transform_create_request( + name=name, litellm_params=dict(litellm_params) + ) + if extra_body: + data.update(extra_body) + + logging_obj.pre_call( + input=name, + api_key="", + additional_args={ + "complete_input_dict": data, + "api_base": url, + "headers": headers, + }, + ) + try: + response = await async_httpx_client.post( + url=url, headers=headers, json=data, timeout=timeout or request_timeout + ) + except Exception as e: + raise self._handle_error(e=e, provider_config=agents_api_config) + + logging_obj.post_call( + original_response=response.text, + additional_args={"complete_input_dict": data}, + ) + return agents_api_config.transform_create_response( + raw_response=response, name=name + ) + + # ------------------------------------------------------------------ # + # LIST # + # ------------------------------------------------------------------ # + + def list_agents( + self, + agents_api_config: BaseAgentsAPIConfig, + litellm_params: GenericLiteLLMParams, + logging_obj: LiteLLMLoggingObj, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + client: Optional[HTTPHandler] = None, + _is_async: bool = False, + ) -> Union[AgentListResponse, Coroutine[Any, Any, AgentListResponse]]: + if _is_async: + return self.async_list_agents( + agents_api_config=agents_api_config, + litellm_params=litellm_params, + logging_obj=logging_obj, + extra_headers=extra_headers, + timeout=timeout, + ) + + sync_httpx_client = self._sync_client(litellm_params, client) + headers = agents_api_config.validate_environment( + headers=extra_headers or {}, litellm_params=dict(litellm_params) + ) + url, params = agents_api_config.transform_list_request( + api_base=litellm_params.get("api_base"), + litellm_params=dict(litellm_params), + ) + logging_obj.pre_call( + input="list_agents", + api_key="", + additional_args={"api_base": url, "headers": headers}, + ) + try: + response = sync_httpx_client.get(url=url, headers=headers, params=params) + except Exception as e: + raise self._handle_error(e=e, provider_config=agents_api_config) + + logging_obj.post_call(original_response=response.text, additional_args={}) + return agents_api_config.transform_list_response(raw_response=response) + + async def async_list_agents( + self, + agents_api_config: BaseAgentsAPIConfig, + litellm_params: GenericLiteLLMParams, + logging_obj: LiteLLMLoggingObj, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + client: Optional[AsyncHTTPHandler] = None, + ) -> AgentListResponse: + async_httpx_client = self._async_client(litellm_params, client) + headers = agents_api_config.validate_environment( + headers=extra_headers or {}, litellm_params=dict(litellm_params) + ) + url, params = agents_api_config.transform_list_request( + api_base=litellm_params.get("api_base"), + litellm_params=dict(litellm_params), + ) + logging_obj.pre_call( + input="list_agents", + api_key="", + additional_args={"api_base": url, "headers": headers}, + ) + try: + response = await async_httpx_client.get( + url=url, headers=headers, params=params + ) + except Exception as e: + raise self._handle_error(e=e, provider_config=agents_api_config) + + logging_obj.post_call(original_response=response.text, additional_args={}) + return agents_api_config.transform_list_response(raw_response=response) + + # ------------------------------------------------------------------ # + # GET # + # ------------------------------------------------------------------ # + + def get_agent( + self, + agents_api_config: BaseAgentsAPIConfig, + name: str, + litellm_params: GenericLiteLLMParams, + logging_obj: LiteLLMLoggingObj, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + client: Optional[HTTPHandler] = None, + _is_async: bool = False, + ) -> Union[AgentCreateResponse, Coroutine[Any, Any, AgentCreateResponse]]: + if _is_async: + return self.async_get_agent( + agents_api_config=agents_api_config, + name=name, + litellm_params=litellm_params, + logging_obj=logging_obj, + extra_headers=extra_headers, + timeout=timeout, + ) + + sync_httpx_client = self._sync_client(litellm_params, client) + headers = agents_api_config.validate_environment( + headers=extra_headers or {}, litellm_params=dict(litellm_params) + ) + url, params = agents_api_config.transform_get_request( + name=name, + api_base=litellm_params.get("api_base"), + litellm_params=dict(litellm_params), + ) + logging_obj.pre_call( + input=name, + api_key="", + additional_args={"api_base": url, "headers": headers}, + ) + try: + response = sync_httpx_client.get(url=url, headers=headers, params=params) + except Exception as e: + raise self._handle_error(e=e, provider_config=agents_api_config) + + logging_obj.post_call(original_response=response.text, additional_args={}) + return agents_api_config.transform_get_response( + raw_response=response, name=name + ) + + async def async_get_agent( + self, + agents_api_config: BaseAgentsAPIConfig, + name: str, + litellm_params: GenericLiteLLMParams, + logging_obj: LiteLLMLoggingObj, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + client: Optional[AsyncHTTPHandler] = None, + ) -> AgentCreateResponse: + async_httpx_client = self._async_client(litellm_params, client) + headers = agents_api_config.validate_environment( + headers=extra_headers or {}, litellm_params=dict(litellm_params) + ) + url, params = agents_api_config.transform_get_request( + name=name, + api_base=litellm_params.get("api_base"), + litellm_params=dict(litellm_params), + ) + logging_obj.pre_call( + input=name, + api_key="", + additional_args={"api_base": url, "headers": headers}, + ) + try: + response = await async_httpx_client.get( + url=url, headers=headers, params=params + ) + except Exception as e: + raise self._handle_error(e=e, provider_config=agents_api_config) + + logging_obj.post_call(original_response=response.text, additional_args={}) + return agents_api_config.transform_get_response( + raw_response=response, name=name + ) + + # ------------------------------------------------------------------ # + # DELETE # + # ------------------------------------------------------------------ # + + def delete_agent( + self, + agents_api_config: BaseAgentsAPIConfig, + name: str, + litellm_params: GenericLiteLLMParams, + logging_obj: LiteLLMLoggingObj, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + client: Optional[HTTPHandler] = None, + _is_async: bool = False, + ) -> Union[AgentDeleteResult, Coroutine[Any, Any, AgentDeleteResult]]: + if _is_async: + return self.async_delete_agent( + agents_api_config=agents_api_config, + name=name, + litellm_params=litellm_params, + logging_obj=logging_obj, + extra_headers=extra_headers, + timeout=timeout, + ) + + sync_httpx_client = self._sync_client(litellm_params, client) + headers = agents_api_config.validate_environment( + headers=extra_headers or {}, litellm_params=dict(litellm_params) + ) + url = agents_api_config.transform_delete_request( + name=name, + api_base=litellm_params.get("api_base"), + litellm_params=dict(litellm_params), + ) + logging_obj.pre_call( + input=name, + api_key="", + additional_args={"api_base": url, "headers": headers}, + ) + try: + response = sync_httpx_client.delete( + url=url, headers=headers, timeout=timeout or request_timeout + ) + except Exception as e: + raise self._handle_error(e=e, provider_config=agents_api_config) + + logging_obj.post_call(original_response=response.text, additional_args={}) + return agents_api_config.transform_delete_response( + raw_response=response, name=name + ) + + async def async_delete_agent( + self, + agents_api_config: BaseAgentsAPIConfig, + name: str, + litellm_params: GenericLiteLLMParams, + logging_obj: LiteLLMLoggingObj, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + client: Optional[AsyncHTTPHandler] = None, + ) -> AgentDeleteResult: + async_httpx_client = self._async_client(litellm_params, client) + headers = agents_api_config.validate_environment( + headers=extra_headers or {}, litellm_params=dict(litellm_params) + ) + url = agents_api_config.transform_delete_request( + name=name, + api_base=litellm_params.get("api_base"), + litellm_params=dict(litellm_params), + ) + logging_obj.pre_call( + input=name, + api_key="", + additional_args={"api_base": url, "headers": headers}, + ) + try: + response = await async_httpx_client.delete( + url=url, headers=headers, timeout=timeout or request_timeout + ) + except Exception as e: + raise self._handle_error(e=e, provider_config=agents_api_config) + + logging_obj.post_call(original_response=response.text, additional_args={}) + return agents_api_config.transform_delete_response( + raw_response=response, name=name + ) + + # ------------------------------------------------------------------ # + # LIST VERSIONS # + # ------------------------------------------------------------------ # + + def list_agent_versions( + self, + agents_api_config: BaseAgentsAPIConfig, + name: str, + litellm_params: GenericLiteLLMParams, + logging_obj: LiteLLMLoggingObj, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + client: Optional[HTTPHandler] = None, + _is_async: bool = False, + ) -> Union[AgentVersionsResponse, Coroutine[Any, Any, AgentVersionsResponse]]: + if _is_async: + return self.async_list_agent_versions( + agents_api_config=agents_api_config, + name=name, + litellm_params=litellm_params, + logging_obj=logging_obj, + extra_headers=extra_headers, + timeout=timeout, + ) + + sync_httpx_client = self._sync_client(litellm_params, client) + headers = agents_api_config.validate_environment( + headers=extra_headers or {}, litellm_params=dict(litellm_params) + ) + url, params = agents_api_config.transform_list_versions_request( + name=name, + api_base=litellm_params.get("api_base"), + litellm_params=dict(litellm_params), + ) + logging_obj.pre_call( + input=name, + api_key="", + additional_args={"api_base": url, "headers": headers}, + ) + try: + response = sync_httpx_client.get(url=url, headers=headers, params=params) + except Exception as e: + raise self._handle_error(e=e, provider_config=agents_api_config) + + logging_obj.post_call(original_response=response.text, additional_args={}) + return agents_api_config.transform_list_versions_response( + raw_response=response, name=name + ) + + async def async_list_agent_versions( + self, + agents_api_config: BaseAgentsAPIConfig, + name: str, + litellm_params: GenericLiteLLMParams, + logging_obj: LiteLLMLoggingObj, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + client: Optional[AsyncHTTPHandler] = None, + ) -> AgentVersionsResponse: + async_httpx_client = self._async_client(litellm_params, client) + headers = agents_api_config.validate_environment( + headers=extra_headers or {}, litellm_params=dict(litellm_params) + ) + url, params = agents_api_config.transform_list_versions_request( + name=name, + api_base=litellm_params.get("api_base"), + litellm_params=dict(litellm_params), + ) + logging_obj.pre_call( + input=name, + api_key="", + additional_args={"api_base": url, "headers": headers}, + ) + try: + response = await async_httpx_client.get( + url=url, headers=headers, params=params + ) + except Exception as e: + raise self._handle_error(e=e, provider_config=agents_api_config) + + logging_obj.post_call(original_response=response.text, additional_args={}) + return agents_api_config.transform_list_versions_response( + raw_response=response, name=name + ) + + +agents_http_handler = AgentsHTTPHandler() diff --git a/litellm/interactions/agents/main.py b/litellm/interactions/agents/main.py new file mode 100644 index 00000000000..f56c6f3ed5e --- /dev/null +++ b/litellm/interactions/agents/main.py @@ -0,0 +1,522 @@ +""" +LiteLLM Agents API - Main Module + +Usage: + import litellm + + # Create + response = litellm.interactions.agents.create( + name="waverunner", + custom_llm_provider="gemini", + api_key="...", + base_agent="gemini-2.5-flash", + instructions="You are a helpful assistant.", + ) + + # List + response = litellm.interactions.agents.list(api_key="...", custom_llm_provider="gemini") + + # Get + response = litellm.interactions.agents.get(name="waverunner", api_key="...") + + # Delete + result = litellm.interactions.agents.delete(name="waverunner", api_key="...") + + # List versions + result = litellm.interactions.agents.list_versions(name="waverunner", api_key="...") + + # Async versions: acreate, alist, aget, adelete, alist_versions +""" + +import asyncio +import contextvars +from functools import partial +from typing import Any, Coroutine, Dict, Optional, Union + +import httpx + +import litellm +from litellm.interactions.agents.http_handler import agents_http_handler +from litellm.interactions.agents.utils import get_provider_agents_api_config +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.types.agents import ( + AgentCreateResponse, + AgentDeleteResult, + AgentListResponse, + AgentVersionsResponse, +) +from litellm.types.interactions import InteractionEnvironment +from litellm.types.router import GenericLiteLLMParams +from litellm.utils import client + +# ------------------------------------------------------------------ # +# Shared helpers # +# ------------------------------------------------------------------ # + + +def _get_agents_api_config(custom_llm_provider: str): + config = get_provider_agents_api_config(custom_llm_provider) + if config is None: + raise litellm.BadRequestError( + message=( + f"Provider '{custom_llm_provider}' does not have a native " + "agents API. Use the proxy POST /v1/agents endpoint to store " + "agents locally." + ), + model="", + llm_provider=custom_llm_provider, + ) + return config + + +def _make_logging_obj( + kwargs: Dict[str, Any], + model: str, + custom_llm_provider: str, + call_type: str, + optional_params: Dict[str, Any], +) -> LiteLLMLoggingObj: + litellm_logging_obj: LiteLLMLoggingObj = kwargs.get("litellm_logging_obj") # type: ignore + litellm_call_id: Optional[str] = kwargs.get("litellm_call_id", None) + litellm_logging_obj.update_from_kwargs( + kwargs=kwargs, + model=model, + optional_params=optional_params, + litellm_params={"litellm_call_id": litellm_call_id}, + custom_llm_provider=custom_llm_provider, + ) + return litellm_logging_obj + + +# ================================================================== # +# CREATE # +# ================================================================== # + + +@client +async def acreate( + name: str, + base_agent: Optional[str] = None, + instructions: Optional[str] = None, + base_environment: Optional[InteractionEnvironment] = None, + custom_llm_provider: Optional[str] = None, + extra_headers: Optional[Dict[str, Any]] = None, + extra_body: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + **kwargs, +) -> AgentCreateResponse: + """Async: Create a managed agent on the provider side.""" + local_vars = locals() + try: + loop = asyncio.get_event_loop() + kwargs["acreate_agent"] = True + func = partial( + create, + name=name, + base_agent=base_agent, + instructions=instructions, + base_environment=base_environment, + custom_llm_provider=custom_llm_provider or "gemini", + extra_headers=extra_headers, + extra_body=extra_body, + timeout=timeout, + **kwargs, + ) + ctx = contextvars.copy_context() + init_response = await loop.run_in_executor(None, partial(ctx.run, func)) + if asyncio.iscoroutine(init_response): + return await init_response + return init_response + except Exception as e: + raise litellm.exception_type( + model=name, + custom_llm_provider=custom_llm_provider or "gemini", + original_exception=e, + completion_kwargs=local_vars, + extra_kwargs=kwargs, + ) + + +@client +def create( + name: str, + base_agent: Optional[str] = None, + instructions: Optional[str] = None, + base_environment: Optional[InteractionEnvironment] = None, + custom_llm_provider: Optional[str] = None, + extra_headers: Optional[Dict[str, Any]] = None, + extra_body: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + **kwargs, +) -> Union[AgentCreateResponse, Coroutine[Any, Any, AgentCreateResponse]]: + """ + Sync: Create a managed agent on the provider side. + + Args: + name: Name for the agent (required). + base_agent: Base agent to derive from (e.g. "waverunner"). + instructions: System instructions for the agent. + base_environment: Environment to fork from — an env_id string or a + dict like ``{"type": "remote", "sources": [...]}``. + custom_llm_provider: Provider to use, e.g. "gemini". + extra_headers: Additional HTTP headers. + extra_body: Additional request body fields. + timeout: Request timeout. + **kwargs: Forwarded to GenericLiteLLMParams (api_key, api_base, etc.). + """ + local_vars = locals() + custom_llm_provider = ( + custom_llm_provider or kwargs.get("custom_llm_provider") or "gemini" + ) + try: + _is_async = kwargs.pop("acreate_agent", False) is True + if base_agent is not None: + kwargs["base_agent"] = base_agent + if instructions is not None: + kwargs["instructions"] = instructions + if base_environment is not None: + kwargs["base_environment"] = base_environment + kwargs.setdefault("custom_llm_provider", custom_llm_provider) + litellm_params = GenericLiteLLMParams(**kwargs) + logging_obj = _make_logging_obj( + kwargs, name, custom_llm_provider, "create_agent", {} + ) + config = _get_agents_api_config(custom_llm_provider) + return agents_http_handler.create_agent( + agents_api_config=config, + name=name, + litellm_params=litellm_params, + logging_obj=logging_obj, + extra_headers=extra_headers, + extra_body=extra_body, + timeout=timeout, + _is_async=_is_async, + ) + except Exception as e: + raise litellm.exception_type( + model=name, + custom_llm_provider=custom_llm_provider, + original_exception=e, + completion_kwargs=local_vars, + extra_kwargs=kwargs, + ) + + +# ================================================================== # +# LIST # +# ================================================================== # + + +@client +async def alist( + custom_llm_provider: Optional[str] = None, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + **kwargs, +) -> AgentListResponse: + """Async: List all agents on the provider side.""" + local_vars = locals() + try: + loop = asyncio.get_event_loop() + kwargs["alist_agents"] = True + func = partial( + list, + custom_llm_provider=custom_llm_provider or "gemini", + extra_headers=extra_headers, + timeout=timeout, + **kwargs, + ) + ctx = contextvars.copy_context() + init_response = await loop.run_in_executor(None, partial(ctx.run, func)) + if asyncio.iscoroutine(init_response): + return await init_response + return init_response + except Exception as e: + raise litellm.exception_type( + model="", + custom_llm_provider=custom_llm_provider or "gemini", + original_exception=e, + completion_kwargs=local_vars, + extra_kwargs=kwargs, + ) + + +@client +def list( + custom_llm_provider: Optional[str] = None, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + **kwargs, +) -> Union[AgentListResponse, Coroutine[Any, Any, AgentListResponse]]: + """Sync: List all agents on the provider side.""" + local_vars = locals() + custom_llm_provider = ( + custom_llm_provider or kwargs.get("custom_llm_provider") or "gemini" + ) + try: + _is_async = kwargs.pop("alist_agents", False) is True + kwargs.setdefault("custom_llm_provider", custom_llm_provider) + litellm_params = GenericLiteLLMParams(**kwargs) + logging_obj = _make_logging_obj( + kwargs, "", custom_llm_provider, "list_agents", {} + ) + config = _get_agents_api_config(custom_llm_provider) + return agents_http_handler.list_agents( + agents_api_config=config, + litellm_params=litellm_params, + logging_obj=logging_obj, + extra_headers=extra_headers, + timeout=timeout, + _is_async=_is_async, + ) + except Exception as e: + raise litellm.exception_type( + model="", + custom_llm_provider=custom_llm_provider, + original_exception=e, + completion_kwargs=local_vars, + extra_kwargs=kwargs, + ) + + +# ================================================================== # +# GET # +# ================================================================== # + + +@client +async def aget( + name: str, + custom_llm_provider: Optional[str] = None, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + **kwargs, +) -> AgentCreateResponse: + """Async: Get a specific agent by name.""" + local_vars = locals() + try: + loop = asyncio.get_event_loop() + kwargs["aget_agent"] = True + func = partial( + get, + name=name, + custom_llm_provider=custom_llm_provider or "gemini", + extra_headers=extra_headers, + timeout=timeout, + **kwargs, + ) + ctx = contextvars.copy_context() + init_response = await loop.run_in_executor(None, partial(ctx.run, func)) + if asyncio.iscoroutine(init_response): + return await init_response + return init_response + except Exception as e: + raise litellm.exception_type( + model=name, + custom_llm_provider=custom_llm_provider or "gemini", + original_exception=e, + completion_kwargs=local_vars, + extra_kwargs=kwargs, + ) + + +@client +def get( + name: str, + custom_llm_provider: Optional[str] = None, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + **kwargs, +) -> Union[AgentCreateResponse, Coroutine[Any, Any, AgentCreateResponse]]: + """Sync: Get a specific agent by name.""" + local_vars = locals() + custom_llm_provider = ( + custom_llm_provider or kwargs.get("custom_llm_provider") or "gemini" + ) + try: + _is_async = kwargs.pop("aget_agent", False) is True + kwargs.setdefault("custom_llm_provider", custom_llm_provider) + litellm_params = GenericLiteLLMParams(**kwargs) + logging_obj = _make_logging_obj( + kwargs, name, custom_llm_provider, "get_agent", {"name": name} + ) + config = _get_agents_api_config(custom_llm_provider) + return agents_http_handler.get_agent( + agents_api_config=config, + name=name, + litellm_params=litellm_params, + logging_obj=logging_obj, + extra_headers=extra_headers, + timeout=timeout, + _is_async=_is_async, + ) + except Exception as e: + raise litellm.exception_type( + model=name, + custom_llm_provider=custom_llm_provider, + original_exception=e, + completion_kwargs=local_vars, + extra_kwargs=kwargs, + ) + + +# ================================================================== # +# DELETE # +# ================================================================== # + + +@client +async def adelete( + name: str, + custom_llm_provider: Optional[str] = None, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + **kwargs, +) -> AgentDeleteResult: + """Async: Delete a specific agent by name.""" + local_vars = locals() + try: + loop = asyncio.get_event_loop() + kwargs["adelete_agent"] = True + func = partial( + delete, + name=name, + custom_llm_provider=custom_llm_provider or "gemini", + extra_headers=extra_headers, + timeout=timeout, + **kwargs, + ) + ctx = contextvars.copy_context() + init_response = await loop.run_in_executor(None, partial(ctx.run, func)) + if asyncio.iscoroutine(init_response): + return await init_response + return init_response + except Exception as e: + raise litellm.exception_type( + model=name, + custom_llm_provider=custom_llm_provider or "gemini", + original_exception=e, + completion_kwargs=local_vars, + extra_kwargs=kwargs, + ) + + +@client +def delete( + name: str, + custom_llm_provider: Optional[str] = None, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + **kwargs, +) -> Union[AgentDeleteResult, Coroutine[Any, Any, AgentDeleteResult]]: + """Sync: Delete a specific agent by name.""" + local_vars = locals() + custom_llm_provider = ( + custom_llm_provider or kwargs.get("custom_llm_provider") or "gemini" + ) + try: + _is_async = kwargs.pop("adelete_agent", False) is True + kwargs.setdefault("custom_llm_provider", custom_llm_provider) + litellm_params = GenericLiteLLMParams(**kwargs) + logging_obj = _make_logging_obj( + kwargs, name, custom_llm_provider, "delete_agent", {"name": name} + ) + config = _get_agents_api_config(custom_llm_provider) + return agents_http_handler.delete_agent( + agents_api_config=config, + name=name, + litellm_params=litellm_params, + logging_obj=logging_obj, + extra_headers=extra_headers, + timeout=timeout, + _is_async=_is_async, + ) + except Exception as e: + raise litellm.exception_type( + model=name, + custom_llm_provider=custom_llm_provider, + original_exception=e, + completion_kwargs=local_vars, + extra_kwargs=kwargs, + ) + + +# ================================================================== # +# LIST VERSIONS # +# ================================================================== # + + +@client +async def alist_versions( + name: str, + custom_llm_provider: Optional[str] = None, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + **kwargs, +) -> AgentVersionsResponse: + """Async: List versions of a specific agent.""" + local_vars = locals() + try: + loop = asyncio.get_event_loop() + kwargs["alist_agent_versions"] = True + func = partial( + list_versions, + name=name, + custom_llm_provider=custom_llm_provider or "gemini", + extra_headers=extra_headers, + timeout=timeout, + **kwargs, + ) + ctx = contextvars.copy_context() + init_response = await loop.run_in_executor(None, partial(ctx.run, func)) + if asyncio.iscoroutine(init_response): + return await init_response + return init_response + except Exception as e: + raise litellm.exception_type( + model=name, + custom_llm_provider=custom_llm_provider or "gemini", + original_exception=e, + completion_kwargs=local_vars, + extra_kwargs=kwargs, + ) + + +@client +def list_versions( + name: str, + custom_llm_provider: Optional[str] = None, + extra_headers: Optional[Dict[str, Any]] = None, + timeout: Optional[Union[float, httpx.Timeout]] = None, + **kwargs, +) -> Union[AgentVersionsResponse, Coroutine[Any, Any, AgentVersionsResponse]]: + """Sync: List versions of a specific agent.""" + local_vars = locals() + custom_llm_provider = ( + custom_llm_provider or kwargs.get("custom_llm_provider") or "gemini" + ) + try: + _is_async = kwargs.pop("alist_agent_versions", False) is True + kwargs.setdefault("custom_llm_provider", custom_llm_provider) + litellm_params = GenericLiteLLMParams(**kwargs) + logging_obj = _make_logging_obj( + kwargs, name, custom_llm_provider, "list_agent_versions", {"name": name} + ) + config = _get_agents_api_config(custom_llm_provider) + return agents_http_handler.list_agent_versions( + agents_api_config=config, + name=name, + litellm_params=litellm_params, + logging_obj=logging_obj, + extra_headers=extra_headers, + timeout=timeout, + _is_async=_is_async, + ) + except Exception as e: + raise litellm.exception_type( + model=name, + custom_llm_provider=custom_llm_provider, + original_exception=e, + completion_kwargs=local_vars, + extra_kwargs=kwargs, + ) diff --git a/litellm/interactions/agents/utils.py b/litellm/interactions/agents/utils.py new file mode 100644 index 00000000000..d16a9597f53 --- /dev/null +++ b/litellm/interactions/agents/utils.py @@ -0,0 +1,23 @@ +""" +Utility functions for the Agents API SDK. +""" + +from typing import Optional + +from litellm.llms.base_llm.agents.transformation import BaseAgentsAPIConfig + + +def get_provider_agents_api_config( + custom_llm_provider: Optional[str], +) -> Optional[BaseAgentsAPIConfig]: + """ + Return a provider-specific BaseAgentsAPIConfig if the provider has a + native agent-creation API, or None otherwise. + """ + from litellm.types.utils import LlmProviders + + if custom_llm_provider == LlmProviders.GEMINI.value: + from litellm.llms.gemini.agents.transformation import GeminiAgentsConfig + + return GeminiAgentsConfig() + return None diff --git a/litellm/interactions/http_handler.py b/litellm/interactions/http_handler.py index 7fead07043f..695da2be89a 100644 --- a/litellm/interactions/http_handler.py +++ b/litellm/interactions/http_handler.py @@ -41,27 +41,55 @@ from litellm.types.interactions import ( from litellm.types.router import GenericLiteLLMParams -class InteractionsHTTPHandler: +class _BaseHTTPHandler: + """ + Shared HTTP infrastructure for LiteLLM handler classes. + + Provides common client resolution and error-mapping helpers so that + handler subclasses (InteractionsHTTPHandler, AgentsHTTPHandler, …) do + not duplicate this boilerplate. + """ + + def _handle_error(self, e: Exception, provider_config: Any) -> Exception: + if isinstance(e, httpx.HTTPStatusError): + return provider_config.get_error_class( + error_message=e.response.text, + status_code=e.response.status_code, + headers=dict(e.response.headers), + ) + return e + + def _sync_client( + self, + litellm_params: GenericLiteLLMParams, + client: Optional[HTTPHandler], + ) -> HTTPHandler: + return client or _get_httpx_client( + params={"ssl_verify": litellm_params.get("ssl_verify", None)} + ) + + def _async_client( + self, + litellm_params: GenericLiteLLMParams, + client: Optional[AsyncHTTPHandler], + ) -> AsyncHTTPHandler: + # GenericLiteLLMParams.get uses getattr; an unset field is None, not the default. + custom_llm_provider = litellm_params.get("custom_llm_provider") or "gemini" + return client or get_async_httpx_client( + llm_provider=litellm.LlmProviders(custom_llm_provider), + params={"ssl_verify": litellm_params.get("ssl_verify", None)}, + ) + + +class InteractionsHTTPHandler(_BaseHTTPHandler): """ HTTP handler for Interactions API requests. """ - def _handle_error( - self, - e: Exception, - provider_config: BaseInteractionsAPIConfig, - ) -> Exception: - """Handle errors from HTTP requests.""" - if isinstance(e, httpx.HTTPStatusError): - error_message = e.response.text - status_code = e.response.status_code - headers = dict(e.response.headers) - return provider_config.get_error_class( - error_message=error_message, - status_code=status_code, - headers=headers, - ) - return e + # _handle_error is inherited from _BaseHTTPHandler (accepts Any provider_config). + # AgentsHTTPHandler also extends this class and passes BaseAgentsAPIConfig, which + # is structurally compatible but a different type — keeping the override here with + # BaseInteractionsAPIConfig would cause type errors in the subclass. # ========================================================= # CREATE INTERACTION diff --git a/litellm/interactions/litellm_responses_transformation/streaming_iterator.py b/litellm/interactions/litellm_responses_transformation/streaming_iterator.py index 72a3afbc3c5..4a3eb63084e 100644 --- a/litellm/interactions/litellm_responses_transformation/streaming_iterator.py +++ b/litellm/interactions/litellm_responses_transformation/streaming_iterator.py @@ -2,7 +2,17 @@ Streaming iterator for transforming Responses API stream to Interactions API stream. """ -from typing import Any, AsyncIterator, Dict, Iterator, Optional, cast +from collections import deque +from typing import ( + Any, + AsyncIterator, + Deque, + Dict, + Iterator, + List, + Optional, + cast, +) from litellm.responses.streaming_iterator import ( BaseResponsesAPIStreamingIterator, @@ -29,7 +39,13 @@ class LiteLLMResponsesInteractionsStreamingIterator: This class handles both sync and async iteration, transforming Responses API streaming events (output.text.delta, response.completed, etc.) to Interactions - API streaming events (content.delta, interaction.complete, etc.). + API streaming events. + + Schema selection: + - New schema (default, use_legacy_interactions_schema=False): + interaction.created -> step.start -> step.delta ... -> step.stop -> interaction.completed + - Legacy schema (use_legacy_interactions_schema=True, remove after June 8 2026): + interaction.start -> content.start -> content.delta ... -> content.stop -> interaction.complete """ def __init__( @@ -41,6 +57,8 @@ class LiteLLMResponsesInteractionsStreamingIterator: custom_llm_provider: Optional[str] = None, litellm_metadata: Optional[Dict[str, Any]] = None, ): + import litellm + self.model = model self.responses_stream_iterator = litellm_custom_stream_wrapper self.request_input = request_input @@ -51,66 +69,156 @@ class LiteLLMResponsesInteractionsStreamingIterator: self.collected_text = "" self.sent_interaction_start = False self.sent_content_start = False + # Capture the schema flag once at construction time so all events + # emitted by this stream use a consistent schema, even if the global + # flag is mutated mid-stream (e.g. by a config reload). + self._use_legacy: bool = litellm.use_legacy_interactions_schema + # Buffer of events that have been derived from upstream chunks but not + # yet returned to the caller. A single Responses API chunk may expand + # into multiple Interactions API events (e.g. the first text delta + # produces interaction.created + step.start + step.delta), and the + # terminal sequence on stream end may also span multiple events + # (step.stop + interaction.completed). + self._pending_events: Deque[InteractionsAPIStreamingResponse] = deque() + # Tracks whether we've already emitted a terminal completion event so + # the StopIteration fallback path doesn't double-emit. + self._sent_completion_event = False + # ID resolved from the first upstream chunk (item_id on a text delta or + # response.id on response.created). Persisted so the EOF terminal + # events stay correlated with the start events delivered earlier. + self._interaction_id: Optional[str] = None - def _transform_responses_chunk_to_interactions_chunk( - self, - responses_chunk: ResponsesAPIStreamingResponse, - ) -> Optional[InteractionsAPIStreamingResponse]: + # ------------------------------------------------------------------ + # Event builders + # ------------------------------------------------------------------ + + def _build_interaction_start_event( + self, interaction_id: str + ) -> InteractionsAPIStreamingResponse: + event_type = "interaction.start" if self._use_legacy else "interaction.created" + return InteractionsAPIStreamingResponse( + event_type=event_type, + id=interaction_id, + object="interaction", + status="in_progress", + model=self.model, + ) + + def _build_content_start_event( + self, interaction_id: str + ) -> InteractionsAPIStreamingResponse: + if self._use_legacy: + return InteractionsAPIStreamingResponse( + event_type="content.start", + id=interaction_id, + object="content", + delta={"type": "text", "text": ""}, + ) + return InteractionsAPIStreamingResponse( + event_type="step.start", + index=0, + step={"type": "model_output", "content": []}, + ) + + def _build_text_delta_event( + self, interaction_id: str, delta_text: str + ) -> InteractionsAPIStreamingResponse: + if self._use_legacy: + return InteractionsAPIStreamingResponse( + event_type="content.delta", + id=interaction_id, + object="content", + delta={"type": "text", "text": delta_text}, + ) + return InteractionsAPIStreamingResponse( + event_type="step.delta", + index=0, + delta={"type": "text", "text": delta_text}, + ) + + def _build_content_stop_event( + self, interaction_id: Optional[str] + ) -> InteractionsAPIStreamingResponse: + if self._use_legacy: + return InteractionsAPIStreamingResponse( + event_type="content.stop", + id=interaction_id, + object="content", + delta={"type": "text", "text": self.collected_text}, + ) + return InteractionsAPIStreamingResponse( + event_type="step.stop", + index=0, + ) + + def _build_completion_event( + self, response_id: str + ) -> InteractionsAPIStreamingResponse: + if self._use_legacy: + return InteractionsAPIStreamingResponse( + event_type="interaction.complete", + id=response_id, + object="interaction", + status="completed", + model=self.model, + outputs=[{"type": "text", "text": self.collected_text}], + ) + return InteractionsAPIStreamingResponse( + event_type="interaction.completed", + id=response_id, + object="interaction", + status="completed", + model=self.model, + steps=[ + { + "type": "model_output", + "content": [{"type": "text", "text": self.collected_text}], + } + ], + ) + + # ------------------------------------------------------------------ + # Per-chunk transform (returns a list of events to enqueue) + # ------------------------------------------------------------------ + + def _events_for_chunk( + self, responses_chunk: ResponsesAPIStreamingResponse + ) -> List[InteractionsAPIStreamingResponse]: """ - Transform a Responses API streaming chunk to an Interactions API streaming chunk. + Translate a single upstream Responses API chunk into the list of + Interactions API events it should produce. - Responses API events: - - output.text.delta -> content.delta - - response.completed -> interaction.complete - - Interactions API events: - - interaction.start - - content.start - - content.delta - - content.stop - - interaction.complete + Returning a list (rather than a single event) lets a chunk emit any + synthetic start events that haven't been sent yet *together with* the + actual delta event, so we never silently drop the chunk's payload. """ if not responses_chunk: - return None + return [] - # Handle OutputTextDeltaEvent -> content.delta + # Text delta: emit any missing start events, then the delta itself. if isinstance(responses_chunk, OutputTextDeltaEvent): delta_text = ( responses_chunk.delta if isinstance(responses_chunk.delta, str) else "" ) self.collected_text += delta_text + interaction_id = ( + getattr(responses_chunk, "item_id", None) or f"interaction_{id(self)}" + ) + if self._interaction_id is None: + self._interaction_id = interaction_id - # Send interaction.start if not sent + events: List[InteractionsAPIStreamingResponse] = [] if not self.sent_interaction_start: self.sent_interaction_start = True - return InteractionsAPIStreamingResponse( - event_type="interaction.start", - id=getattr(responses_chunk, "item_id", None) - or f"interaction_{id(self)}", - object="interaction", - status="in_progress", - model=self.model, - ) - - # Send content.start if not sent + events.append(self._build_interaction_start_event(interaction_id)) if not self.sent_content_start: self.sent_content_start = True - return InteractionsAPIStreamingResponse( - event_type="content.start", - id=getattr(responses_chunk, "item_id", None), - object="content", - delta={"type": "text", "text": ""}, - ) + events.append(self._build_content_start_event(interaction_id)) + events.append(self._build_text_delta_event(interaction_id, delta_text)) + return events - # Send content.delta - return InteractionsAPIStreamingResponse( - event_type="content.delta", - id=getattr(responses_chunk, "item_id", None), - object="content", - delta={"text": delta_text}, - ) - - # Handle ResponseCreatedEvent or ResponseInProgressEvent -> interaction.start + # Response created / in-progress: synthesize interaction start if we + # haven't already sent one. if isinstance(responses_chunk, (ResponseCreatedEvent, ResponseInProgressEvent)): if not self.sent_interaction_start: self.sent_interaction_start = True @@ -118,169 +226,136 @@ class LiteLLMResponsesInteractionsStreamingIterator: getattr(responses_chunk.response, "id", None) if hasattr(responses_chunk, "response") else None - ) - return InteractionsAPIStreamingResponse( - event_type="interaction.start", - id=response_id or f"interaction_{id(self)}", - object="interaction", - status="in_progress", - model=self.model, - ) + ) or f"interaction_{id(self)}" + if self._interaction_id is None: + self._interaction_id = response_id + return [self._build_interaction_start_event(response_id)] + return [] - # Handle ResponseCompletedEvent -> interaction.complete + # Response completed: emit step.stop (if content was started) followed + # by the terminal completion event. Prefer the interaction id already + # established by earlier events so consumers can correlate the start + # and completion events by id (response.id may differ from the item_id + # used to derive the initial id when the stream starts directly with a + # text delta). if isinstance(responses_chunk, ResponseCompletedEvent): self.finished = True response = responses_chunk.response - - # Send content.stop first if content was started - if self.sent_content_start: - # Note: We'll send this in the iterator, not here - pass - - # Send interaction.complete - return InteractionsAPIStreamingResponse( - event_type="interaction.complete", - id=getattr(response, "id", None) or f"interaction_{id(self)}", - object="interaction", - status="completed", - model=self.model, - outputs=[ - { - "type": "text", - "text": self.collected_text, - } - ], + response_id = ( + self._interaction_id + or getattr(response, "id", None) + or f"interaction_{id(self)}" ) - # For other event types, return None (skip) - return None + terminal: List[InteractionsAPIStreamingResponse] = [] + if self.sent_content_start: + terminal.append(self._build_content_stop_event(response_id)) + terminal.append(self._build_completion_event(response_id)) + self._sent_completion_event = True + return terminal + + return [] + + def _build_terminal_events_on_eof( + self, + ) -> List[InteractionsAPIStreamingResponse]: + """ + Build the events to flush when the upstream stream ends without a + ResponseCompletedEvent. Ensures consumers always observe a terminal + interaction.completed/interaction.complete carrying the full text. + """ + if self._sent_completion_event: + return [] + + fallback_id = self._interaction_id or f"interaction_{id(self)}" + terminal: List[InteractionsAPIStreamingResponse] = [] + if self.sent_content_start: + terminal.append(self._build_content_stop_event(fallback_id)) + if self.sent_interaction_start or self.collected_text: + terminal.append(self._build_completion_event(fallback_id)) + self._sent_completion_event = True + return terminal + + # ------------------------------------------------------------------ + # Iteration + # ------------------------------------------------------------------ def __iter__(self) -> Iterator[InteractionsAPIStreamingResponse]: - """Sync iterator implementation.""" return self def __next__(self) -> InteractionsAPIStreamingResponse: - """Get next chunk in sync mode.""" + if self._pending_events: + return self._pending_events.popleft() + if self.finished: raise StopIteration - # Check if we have a pending interaction.complete to send - if hasattr(self, "_pending_interaction_complete"): - pending: InteractionsAPIStreamingResponse = getattr( - self, "_pending_interaction_complete" - ) - delattr(self, "_pending_interaction_complete") - return pending - - # Use a loop instead of recursion to avoid stack overflow sync_iterator = cast( SyncResponsesAPIStreamingIterator, self.responses_stream_iterator ) while True: try: - # Get next chunk from responses API stream chunk = next(sync_iterator) - - # Transform chunk (chunk is already a ResponsesAPIStreamingResponse) - transformed = self._transform_responses_chunk_to_interactions_chunk( - chunk - ) - - if transformed: - # If we finished and content was started, send content.stop before interaction.complete - if ( - self.finished - and self.sent_content_start - and transformed.event_type == "interaction.complete" - ): - # Send content.stop first - content_stop = InteractionsAPIStreamingResponse( - event_type="content.stop", - id=transformed.id, - object="content", - delta={"type": "text", "text": self.collected_text}, - ) - # Store the interaction.complete to send next - self._pending_interaction_complete = transformed - return content_stop - return transformed - - # If no transformation, continue to next chunk (loop continues) - except StopIteration: self.finished = True + self._pending_events.extend(self._build_terminal_events_on_eof()) + if self._pending_events: + return self._pending_events.popleft() + raise - # Send final events if needed - if self.sent_content_start: - return InteractionsAPIStreamingResponse( - event_type="content.stop", - object="content", - delta={"type": "text", "text": self.collected_text}, - ) - - raise StopIteration + events = self._events_for_chunk(chunk) + if events: + self._pending_events.extend(events) + return self._pending_events.popleft() def __aiter__(self) -> AsyncIterator[InteractionsAPIStreamingResponse]: - """Async iterator implementation.""" return self async def __anext__(self) -> InteractionsAPIStreamingResponse: - """Get next chunk in async mode.""" + if self._pending_events: + return self._pending_events.popleft() + if self.finished: raise StopAsyncIteration - # Check if we have a pending interaction.complete to send - if hasattr(self, "_pending_interaction_complete"): - pending: InteractionsAPIStreamingResponse = getattr( - self, "_pending_interaction_complete" - ) - delattr(self, "_pending_interaction_complete") - return pending - - # Use a loop instead of recursion to avoid stack overflow async_iterator = cast( ResponsesAPIStreamingIterator, self.responses_stream_iterator ) while True: try: - # Get next chunk from responses API stream chunk = await async_iterator.__anext__() - - # Transform chunk (chunk is already a ResponsesAPIStreamingResponse) - transformed = self._transform_responses_chunk_to_interactions_chunk( - chunk - ) - - if transformed: - # If we finished and content was started, send content.stop before interaction.complete - if ( - self.finished - and self.sent_content_start - and transformed.event_type == "interaction.complete" - ): - # Send content.stop first - content_stop = InteractionsAPIStreamingResponse( - event_type="content.stop", - id=transformed.id, - object="content", - delta={"type": "text", "text": self.collected_text}, - ) - # Store the interaction.complete to send next - self._pending_interaction_complete = transformed - return content_stop - return transformed - - # If no transformation, continue to next chunk (loop continues) - except StopAsyncIteration: self.finished = True + self._pending_events.extend(self._build_terminal_events_on_eof()) + if self._pending_events: + return self._pending_events.popleft() + raise - # Send final events if needed - if self.sent_content_start: - return InteractionsAPIStreamingResponse( - event_type="content.stop", - object="content", - delta={"type": "text", "text": self.collected_text}, - ) + events = self._events_for_chunk(chunk) + if events: + self._pending_events.extend(events) + return self._pending_events.popleft() - raise StopAsyncIteration + # ------------------------------------------------------------------ + # Backwards-compatible single-chunk transform (used by tests and any + # external callers that drove the iterator chunk-by-chunk pre-fix). + # ------------------------------------------------------------------ + + def _transform_responses_chunk_to_interactions_chunk( + self, + responses_chunk: ResponsesAPIStreamingResponse, + ) -> Optional[InteractionsAPIStreamingResponse]: + """ + Compatibility shim: returns the *first* event produced for this chunk + and queues any remaining events on ``self._pending_events`` so they + are surfaced on subsequent calls/iterations. + + Prefer ``_events_for_chunk`` in new code. + """ + events = self._events_for_chunk(responses_chunk) + if not events: + return None + first = events[0] + if len(events) > 1: + self._pending_events.extend(events[1:]) + return first diff --git a/litellm/interactions/litellm_responses_transformation/transformation.py b/litellm/interactions/litellm_responses_transformation/transformation.py index 100300af7b5..173d4ca8764 100644 --- a/litellm/interactions/litellm_responses_transformation/transformation.py +++ b/litellm/interactions/litellm_responses_transformation/transformation.py @@ -226,29 +226,37 @@ class LiteLLMResponsesInteractionsConfig: - Map status - Extract usage """ - # Extract text from outputs - outputs = [] + # Extract text from outputs and build both `outputs` (legacy) and `steps` (new schema). + outputs: List[Dict[str, Any]] = [] + steps: List[Dict[str, Any]] = [] if hasattr(responses_response, "output") and responses_response.output: for output_item in responses_response.output: # Use getattr with None default to safely access content content = getattr(output_item, "content", None) if content is not None: content_items = content if isinstance(content, list) else [content] + model_output_contents: List[Dict[str, Any]] = [] for content_item in content_items: # Check if content_item has text attribute text = getattr(content_item, "text", None) if text is not None: - outputs.append( - { - "type": "text", - "text": text, - } - ) + # Use independent dict instances so mutations to one + # of `outputs` / `steps` don't leak into the other. + outputs.append({"type": "text", "text": text}) + model_output_contents.append({"type": "text", "text": text}) elif ( isinstance(content_item, dict) and content_item.get("type") == "text" ): - outputs.append(content_item) + outputs.append({**content_item}) + model_output_contents.append({**content_item}) + if model_output_contents: + steps.append( + { + "type": "model_output", + "content": model_output_contents, + } + ) # Convert created_at to ISO string created_at = getattr(responses_response, "created_at", None) @@ -270,12 +278,14 @@ class LiteLLMResponsesInteractionsConfig: else: interactions_status = status - # Build interactions response + # Build interactions response — populate both `outputs` (legacy schema) and + # `steps` (new schema) so callers work regardless of which schema they expect. interactions_response_dict: Dict[str, Any] = { "id": getattr(responses_response, "id", ""), "object": "interaction", "status": interactions_status, "outputs": outputs, + "steps": steps, "model": model or getattr(responses_response, "model", ""), "created": created, } diff --git a/litellm/interactions/main.py b/litellm/interactions/main.py index ab429ef6db5..d99cc3d11c7 100644 --- a/litellm/interactions/main.py +++ b/litellm/interactions/main.py @@ -8,25 +8,25 @@ Per OpenAPI spec (https://ai.google.dev/static/api/interactions.openapi.json): Usage: import litellm - + # Create an interaction with a model response = litellm.interactions.create( model="gemini-2.5-flash", input="Hello, how are you?" ) - + # Create an interaction with an agent response = litellm.interactions.create( agent="deep-research-pro-preview-12-2025", input="Research the current state of cancer research" ) - + # Async version response = await litellm.interactions.acreate(...) - + # Get an interaction response = litellm.interactions.get(interaction_id="...") - + # Delete an interaction result = litellm.interactions.delete(interaction_id="...") """ @@ -48,6 +48,7 @@ from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging from litellm.types.interactions import ( CancelInteractionResult, DeleteInteractionResult, + InteractionEnvironment, InteractionInput, InteractionsAPIResponse, InteractionsAPIStreamingResponse, @@ -80,6 +81,8 @@ async def acreate( store: Optional[bool] = None, # Background execution background: Optional[bool] = None, + # Agent execution environment ("remote", env id, or remote config object) + environment: Optional[InteractionEnvironment] = None, # Response format response_modalities: Optional[List[str]] = None, response_format: Optional[Dict[str, Any]] = None, @@ -109,6 +112,10 @@ async def acreate( stream: Whether to stream the response store: Whether to store the response for later retrieval background: Whether to run in background + environment: Agent execution environment — ``"remote"``, an existing env id + string, or a config object such as + ``{"type": "remote", "sources": [...]}`` / + ``{"type": "remote", "network": {...}}`` response_modalities: Requested response modalities (TEXT, IMAGE, AUDIO) response_format: JSON schema for response format response_mime_type: MIME type of the response @@ -144,6 +151,7 @@ async def acreate( stream=stream, store=store, background=background, + environment=environment, response_modalities=response_modalities, response_format=response_format, response_mime_type=response_mime_type, @@ -194,6 +202,8 @@ def create( store: Optional[bool] = None, # Background execution background: Optional[bool] = None, + # Agent execution environment ("remote", env id, or remote config object) + environment: Optional[InteractionEnvironment] = None, # Response format response_modalities: Optional[List[str]] = None, response_format: Optional[Dict[str, Any]] = None, @@ -231,6 +241,10 @@ def create( stream: Whether to stream the response store: Whether to store the response for later retrieval background: Whether to run in background + environment: Agent execution environment — ``"remote"``, an existing env id + string, or a config object such as + ``{"type": "remote", "sources": [...]}`` / + ``{"type": "remote", "network": {...}}`` response_modalities: Requested response modalities (TEXT, IMAGE, AUDIO) response_format: JSON schema for response format response_mime_type: MIME type of the response @@ -252,7 +266,14 @@ def create( litellm_params = GenericLiteLLMParams(**kwargs) - if model: + # Routing logic: + # - agent provided (no model, or model accidentally set to agent name) → gemini + # - model provided → resolve provider via get_llm_provider (normal routing) + if agent and model == agent: + model = None + if agent and not model: + custom_llm_provider = custom_llm_provider or "gemini" + elif model: model, custom_llm_provider, _, _ = litellm.get_llm_provider( model=model, custom_llm_provider=custom_llm_provider, diff --git a/litellm/interactions/streaming_iterator.py b/litellm/interactions/streaming_iterator.py index a5a7f9e06e5..561686a3e1b 100644 --- a/litellm/interactions/streaming_iterator.py +++ b/litellm/interactions/streaming_iterator.py @@ -101,10 +101,14 @@ class BaseInteractionsAPIStreamingIterator: ) ) - # Store the completed response (check for status=completed) - if ( - streaming_response - and getattr(streaming_response, "status", None) == "completed" + # Store the completed response. + # Legacy schema signals completion via status="completed". + # New schema (Api-Revision: 2026-05-20) uses event_type="interaction.completed". + # Remove the legacy check after June 8, 2026. + if streaming_response and ( + getattr(streaming_response, "status", None) == "completed" + or getattr(streaming_response, "event_type", None) + == "interaction.completed" ): self.completed_response = streaming_response self._handle_logging_completed_response() diff --git a/litellm/interactions/utils.py b/litellm/interactions/utils.py index 3a18ddf52fe..84437f4d3d8 100644 --- a/litellm/interactions/utils.py +++ b/litellm/interactions/utils.py @@ -15,6 +15,7 @@ INTERACTIONS_API_OPTIONAL_PARAMS = { "stream", "store", "background", + "environment", "response_modalities", "response_format", "response_mime_type", diff --git a/litellm/litellm_core_utils/get_litellm_params.py b/litellm/litellm_core_utils/get_litellm_params.py index ad9538ac171..b32803b5dfc 100644 --- a/litellm/litellm_core_utils/get_litellm_params.py +++ b/litellm/litellm_core_utils/get_litellm_params.py @@ -1,5 +1,7 @@ from typing import Optional +from litellm.llms.openai.data_residency import infer_openai_data_residency + # Pre-define optional kwargs keys as frozenset for O(1) lookups # These are extracted from kwargs only if present, avoiding unnecessary .get() calls _OPTIONAL_KWARGS_KEYS = frozenset( @@ -103,6 +105,10 @@ def get_litellm_params( if litellm_trace_id is None: litellm_trace_id = _meta.get("trace_id") or _meta.get("session_id") + data_residency: Optional[str] = infer_openai_data_residency( + custom_llm_provider, api_base + ) + # Build base dict with explicit parameters (always included) litellm_params = { "acompletion": acompletion, @@ -112,6 +118,7 @@ def get_litellm_params( "verbose": verbose, "custom_llm_provider": custom_llm_provider, "api_base": api_base, + "data_residency": data_residency, "litellm_call_id": litellm_call_id, "model_alias_map": model_alias_map, "completion_call_id": completion_call_id, diff --git a/litellm/litellm_core_utils/get_supported_openai_params.py b/litellm/litellm_core_utils/get_supported_openai_params.py index 9d8bd7523db..b8cdc8210fc 100644 --- a/litellm/litellm_core_utils/get_supported_openai_params.py +++ b/litellm/litellm_core_utils/get_supported_openai_params.py @@ -11,6 +11,7 @@ def get_supported_openai_params( # noqa: PLR0915 request_type: Literal[ "chat_completion", "embeddings", "transcription" ] = "chat_completion", + base_model: Optional[str] = None, ) -> Optional[list]: """ Returns the supported openai params for a given model + provider @@ -20,6 +21,11 @@ def get_supported_openai_params( # noqa: PLR0915 get_supported_openai_params(model="anthropic.claude-3", custom_llm_provider="bedrock") ``` + Args: + base_model: For Azure, the true underlying model (e.g. ``"azure/gpt-5.2"``) + when the deployment name differs. Used for model-type detection so that + non-standard deployment names route to the correct config. + Returns: - List if custom_llm_provider is mapped - None if unmapped @@ -32,17 +38,21 @@ def get_supported_openai_params( # noqa: PLR0915 if custom_llm_provider in LlmProvidersSet: provider_config = litellm.ProviderConfigManager.get_provider_chat_config( - model=model, provider=LlmProviders(custom_llm_provider) + model=model, + provider=LlmProviders(custom_llm_provider), + base_model=base_model, ) elif custom_llm_provider.split("/")[0] in LlmProvidersSet: provider_config = litellm.ProviderConfigManager.get_provider_chat_config( - model=model, provider=LlmProviders(custom_llm_provider.split("/")[0]) + model=model, + provider=LlmProviders(custom_llm_provider.split("/")[0]), + base_model=base_model, ) else: provider_config = None if provider_config and request_type == "chat_completion": - return provider_config.get_supported_openai_params(model=model) + return provider_config.get_supported_openai_params(model=base_model or model) if custom_llm_provider == "bedrock": return litellm.AmazonConverseConfig().get_supported_openai_params(model=model) @@ -130,16 +140,23 @@ def get_supported_openai_params( # noqa: PLR0915 model=model ) elif custom_llm_provider == "azure": - if litellm.AzureOpenAIO1Config().is_o_series_model(model=model): + _azure_detection_model = base_model or model + if litellm.AzureOpenAIO1Config().is_o_series_model( + model=_azure_detection_model + ): return litellm.AzureOpenAIO1Config().get_supported_openai_params( - model=model + model=_azure_detection_model ) - elif litellm.AzureOpenAIGPT5Config.is_model_gpt_5_model(model=model): + elif litellm.AzureOpenAIGPT5Config.is_model_gpt_5_model( + model=_azure_detection_model + ): return litellm.AzureOpenAIGPT5Config().get_supported_openai_params( - model=model + model=_azure_detection_model ) else: - return litellm.AzureOpenAIConfig().get_supported_openai_params(model=model) + return litellm.AzureOpenAIConfig().get_supported_openai_params( + model=_azure_detection_model + ) elif custom_llm_provider == "openrouter": return litellm.OpenrouterConfig().get_supported_openai_params(model=model) elif custom_llm_provider == "vercel_ai_gateway": diff --git a/litellm/litellm_core_utils/litellm_logging.py b/litellm/litellm_core_utils/litellm_logging.py index c73d914e6cc..c127b3873a7 100644 --- a/litellm/litellm_core_utils/litellm_logging.py +++ b/litellm/litellm_core_utils/litellm_logging.py @@ -994,10 +994,8 @@ class Logging(LiteLLMLoggingBaseClass): try: # [Non-blocking Extra Debug Information in metadata] if turn_off_message_logging is True: - _metadata["raw_request"] = ( - "redacted by litellm. \ + _metadata["raw_request"] = "redacted by litellm. \ 'litellm.turn_off_message_logging=True'" - ) else: curl_command = self._get_request_curl_command( api_base=additional_args.get("api_base", ""), @@ -1031,12 +1029,8 @@ class Logging(LiteLLMLoggingBaseClass): error=str(e), ) ) - _metadata["raw_request"] = ( - "Unable to Log \ - raw request: {}".format( - str(e) - ) - ) + _metadata["raw_request"] = "Unable to Log \ + raw request: {}".format(str(e)) if getattr(self, "logger_fn", None) and callable(self.logger_fn): try: self.logger_fn( @@ -1050,6 +1044,16 @@ class Logging(LiteLLMLoggingBaseClass): ) self.model_call_details["api_call_start_time"] = datetime.datetime.now() + # Set-once first provider-handoff instant. api_call_start_time + # is overwritten on every retry, so it can't measure one-time + # preprocessing; pinning the first attempt excludes retry loops + # + backoff. Logging object only — must NOT go into + # litellm_params["metadata"] (caller request metadata, typed + # Dict[str, str], echoed downstream; a datetime breaks it). + if self.model_call_details.get("first_api_call_start_time") is None: + self.model_call_details["first_api_call_start_time"] = ( + self.model_call_details["api_call_start_time"] + ) # Input Integration Logging -> If you want to log the fact that an attempt to call the model was made callbacks = litellm.input_callback + (self.dynamic_input_callbacks or []) for callback in callbacks: @@ -1542,6 +1546,11 @@ class Logging(LiteLLMLoggingBaseClass): if self.optional_params else None ), + "data_residency": ( + self.litellm_params.get("data_residency") + if hasattr(self, "litellm_params") and self.litellm_params + else None + ), } except Exception as e: # error creating kwargs for cost calculation debug_info = StandardLoggingModelCostFailureDebugInformation( @@ -1603,6 +1612,90 @@ class Logging(LiteLLMLoggingBaseClass): ) -> Optional[float]: return self._response_cost_calculator(result=result, cache_hit=cache_hit) + @staticmethod + def _is_sync_litellm_request(litellm_params: dict) -> bool: + """True for sync SDK entrypoints (``completion``), false for async (``acompletion``, etc.).""" + return ( + litellm_params.get(CallTypes.acompletion.value, False) is not True + and litellm_params.get(CallTypes.aresponses.value, False) is not True + and litellm_params.get(CallTypes.aembedding.value, False) is not True + and litellm_params.get(CallTypes.aimage_generation.value, False) is not True + and litellm_params.get(CallTypes.atranscription.value, False) is not True + ) + + def _is_assembled_stream_success(self, result=None) -> bool: + """Final assembled stream export (not a per-chunk success call). + + Per-chunk callers pass a ``ModelResponseStream`` (or ``None``); the + final assembled response is any other non-``None`` value (typically a + ``ModelResponse``). Treating a chunk as the assembled response would + prematurely set the ``has_dispatched_final_stream_success`` dedup + guard and silently suppress the real final stream log. + """ + if self.stream is not True: + return False + if result is not None and not isinstance(result, ModelResponseStream): + return True + return ( + "async_complete_streaming_response" in self.model_call_details + or self.model_call_details.get("complete_streaming_response") is not None + ) + + async def dispatch_success_handlers( + self, + result=None, + start_time=None, + end_time=None, + cache_hit=None, + prefer_async_handlers: bool = False, + **kwargs, + ) -> None: + """Route success logging to async and/or sync handlers for this request. + + ``prefer_async_handlers`` only bypasses the sync-SDK-only shortcut (e.g. + ``async for`` on a stream from ``completion()``). Legacy string callbacks + still run via ``executor.submit(success_handler)`` when configured. + """ + from litellm.litellm_core_utils.thread_pool_executor import executor + + if self._is_assembled_stream_success(result): + if self.model_call_details.get("has_dispatched_final_stream_success"): + return + self.model_call_details["has_dispatched_final_stream_success"] = True + + litellm_params = self.model_call_details.get("litellm_params", {}) or {} + sync_sdk = self._is_sync_litellm_request(litellm_params) + passthrough = self.call_type == CallTypes.pass_through.value + if sync_sdk and not prefer_async_handlers and not passthrough: + self.success_handler( + result, + start_time=start_time, + end_time=end_time, + cache_hit=cache_hit, + **kwargs, + ) + return + + await self.async_success_handler( + result, + start_time=start_time, + end_time=end_time, + cache_hit=cache_hit, + **kwargs, + ) + + if not self._should_run_sync_callbacks_for_async_calls(): + return + + executor.submit( + self.success_handler, + result, + start_time=start_time, + end_time=end_time, + cache_hit=cache_hit, + **kwargs, + ) + def should_run_logging( self, event_type: Literal[ @@ -1759,9 +1852,12 @@ class Logging(LiteLLMLoggingBaseClass): self.model_call_details["response_cost"] = 0.0 elif "response_cost" in hidden_params: self.model_call_details["response_cost"] = hidden_params["response_cost"] - elif self.model_call_details.get("response_cost") is not None: + elif ( + existing_cost := self.model_call_details.get("response_cost") + ) is not None and existing_cost != 0: # Preserve response_cost if already calculated (e.g., by pass-through - # handlers like Gemini/Vertex which call completion_cost directly) + # handlers like Gemini/Vertex which call completion_cost directly). + # Do not preserve 0 from failure_handler on intermediate router retries. pass else: self.model_call_details["response_cost"] = self._response_cost_calculator( @@ -2022,13 +2118,7 @@ class Logging(LiteLLMLoggingBaseClass): standard_logging_object=kwargs.get("standard_logging_object", None), ) litellm_params = self.model_call_details.get("litellm_params", {}) - is_sync_request = ( - litellm_params.get(CallTypes.acompletion.value, False) is not True - and litellm_params.get(CallTypes.aresponses.value, False) is not True - and litellm_params.get(CallTypes.aembedding.value, False) is not True - and litellm_params.get(CallTypes.aimage_generation.value, False) is not True - and litellm_params.get(CallTypes.atranscription.value, False) is not True - ) + is_sync_request = self._is_sync_litellm_request(litellm_params) try: ## BUILD COMPLETE STREAMED RESPONSE complete_streaming_response: Optional[ @@ -2484,9 +2574,11 @@ class Logging(LiteLLMLoggingBaseClass): print_verbose( "Logging Details LiteLLM-Async Success Call, cache_hit={}".format(cache_hit) ) - if not self.should_run_logging( + if not self._is_assembled_stream_success( + result + ) and not self.should_run_logging( event_type="async_success" - ): # prevent double logging + ): # prevent double logging (non-streaming) return ## CALCULATE COST FOR BATCH JOBS @@ -2936,13 +3028,7 @@ class Logging(LiteLLMLoggingBaseClass): ): # prevent double logging return litellm_params = self.model_call_details.get("litellm_params", {}) - is_sync_request = ( - litellm_params.get(CallTypes.acompletion.value, False) is not True - and litellm_params.get(CallTypes.aresponses.value, False) is not True - and litellm_params.get(CallTypes.aembedding.value, False) is not True - and litellm_params.get(CallTypes.aimage_generation.value, False) is not True - and litellm_params.get(CallTypes.atranscription.value, False) is not True - ) + is_sync_request = self._is_sync_litellm_request(litellm_params) try: start_time, end_time = self._failure_handler_helper_fn( @@ -3706,6 +3792,9 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915 try: custom_logger_init_args = custom_logger_init_args or {} if logging_integration == "agentops": # Add AgentOps initialization + _v2 = _maybe_construct_otel_v2("agentops", _in_memory_loggers) + if _v2 is not None: + return _v2 # type: ignore for callback in _in_memory_loggers: if isinstance(callback, AgentOps): return callback # type: ignore @@ -3858,6 +3947,9 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915 _in_memory_loggers.append(_opik_logger) return _opik_logger # type: ignore elif logging_integration == "arize": + _v2 = _maybe_construct_otel_v2("arize", _in_memory_loggers) + if _v2 is not None: + return _v2 # type: ignore from litellm.integrations.opentelemetry import ( OpenTelemetry, OpenTelemetryConfig, @@ -3887,6 +3979,9 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915 _in_memory_loggers.append(_arize_otel_logger) return _arize_otel_logger # type: ignore elif logging_integration == "arize_phoenix": + _v2 = _maybe_construct_otel_v2("arize_phoenix", _in_memory_loggers) + if _v2 is not None: + return _v2 # type: ignore from litellm.integrations.opentelemetry import ( OpenTelemetry, OpenTelemetryConfig, @@ -3898,31 +3993,6 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915 endpoint=arize_phoenix_config.endpoint, headers=arize_phoenix_config.otlp_auth_headers, ) - if arize_phoenix_config.project_name: - existing_attrs = os.environ.get("OTEL_RESOURCE_ATTRIBUTES", "") - # Add openinference.project.name attribute - if existing_attrs: - os.environ["OTEL_RESOURCE_ATTRIBUTES"] = ( - f"{existing_attrs},openinference.project.name={arize_phoenix_config.project_name}" - ) - else: - os.environ["OTEL_RESOURCE_ATTRIBUTES"] = ( - f"openinference.project.name={arize_phoenix_config.project_name}" - ) - - # Set Phoenix project name from environment variable - phoenix_project_name = os.environ.get("PHOENIX_PROJECT_NAME", None) - if phoenix_project_name: - existing_attrs = os.environ.get("OTEL_RESOURCE_ATTRIBUTES", "") - # Add openinference.project.name attribute - if existing_attrs: - os.environ["OTEL_RESOURCE_ATTRIBUTES"] = ( - f"{existing_attrs},openinference.project.name={phoenix_project_name}" - ) - else: - os.environ["OTEL_RESOURCE_ATTRIBUTES"] = ( - f"openinference.project.name={phoenix_project_name}" - ) # auth can be disabled on local deployments of arize phoenix if arize_phoenix_config.otlp_auth_headers is not None: @@ -3942,6 +4012,9 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915 _in_memory_loggers.append(_arize_phoenix_otel_logger) return _arize_phoenix_otel_logger # type: ignore elif logging_integration == "levo": + _v2 = _maybe_construct_otel_v2("levo", _in_memory_loggers) + if _v2 is not None: + return _v2 # type: ignore from litellm.integrations.levo.levo import LevoLogger from litellm.integrations.opentelemetry import ( OpenTelemetry, @@ -3967,6 +4040,28 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915 _in_memory_loggers.append(_levo_otel_logger) return _levo_otel_logger # type: ignore elif logging_integration == "otel": + # Gate the new typed V2 adapter behind LITELLM_OTEL_V2. When off, + # the legacy 3,227-line god-class is used unchanged. The two are + # never registered simultaneously — the dedup loop below treats + # any module under ``litellm.integrations.otel`` or + # ``litellm.integrations.opentelemetry`` as "the OTel callback". + from litellm.integrations.otel.model.config import is_otel_v2_enabled + + if is_otel_v2_enabled(): + from litellm.integrations.otel.logger import OpenTelemetryV2 + + for callback in _in_memory_loggers: + if type(callback) is OpenTelemetryV2: + return callback # type: ignore + otel_logger_v2 = OpenTelemetryV2( + **_get_custom_logger_settings_from_proxy_server( + callback_name=logging_integration + ) + ) + _in_memory_loggers.append(otel_logger_v2) + _maybe_auto_initialize_arize_phoenix(_in_memory_loggers) + return otel_logger_v2 # type: ignore + from litellm.integrations.opentelemetry import OpenTelemetry for callback in _in_memory_loggers: @@ -4105,6 +4200,9 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915 elif logging_integration == "langtrace": if "LANGTRACE_API_KEY" not in os.environ: raise ValueError("LANGTRACE_API_KEY not found in environment variables") + _v2 = _maybe_construct_otel_v2("langtrace", _in_memory_loggers) + if _v2 is not None: + return _v2 # type: ignore from litellm.integrations.opentelemetry import ( OpenTelemetry, @@ -4145,6 +4243,9 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915 _in_memory_loggers.append(langfuse_logger) return langfuse_logger # type: ignore elif logging_integration == "langfuse_otel": + _v2 = _maybe_construct_otel_v2("langfuse_otel", _in_memory_loggers) + if _v2 is not None: + return _v2 # type: ignore from litellm.integrations.langfuse.langfuse_otel import LangfuseOtelLogger for callback in _in_memory_loggers: @@ -4161,6 +4262,9 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915 _in_memory_loggers.append(_otel_logger) return _otel_logger # type: ignore elif logging_integration == "weave_otel": + _v2 = _maybe_construct_otel_v2("weave_otel", _in_memory_loggers) + if _v2 is not None: + return _v2 # type: ignore from litellm.integrations.opentelemetry import OpenTelemetryConfig from litellm.integrations.weave.weave_otel import ( WeaveOtelLogger, @@ -4309,6 +4413,42 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915 return None +def _maybe_construct_otel_v2( + callback_name: str, _in_memory_loggers: list +) -> Optional[Any]: + """If ``LITELLM_OTEL_V2`` is on, build (or reuse) a single ``OpenTelemetryV2`` + instance configured via the preset for ``callback_name``. + + Returns ``None`` when V2 is off OR when there's no preset registered for + ``callback_name`` — callers should then fall through to the legacy path. + """ + from litellm.integrations.otel.model.config import is_otel_v2_enabled + + if not is_otel_v2_enabled(): + return None + from litellm.integrations.otel.logger import OpenTelemetryV2 + from litellm.integrations.otel.presets import PRESET_BY_CALLBACK + + preset_fn = PRESET_BY_CALLBACK.get(callback_name) + if preset_fn is None: + return None + for callback in _in_memory_loggers: + if ( + isinstance(callback, OpenTelemetryV2) + and getattr(callback, "callback_name", None) == callback_name + ): + return callback + try: + config = preset_fn() + except Exception: + # If env vars are missing or the preset raises, defer to the legacy path + # so customers get the same error story they had before V2 landed. + return None + v2_logger = OpenTelemetryV2(config=config, callback_name=callback_name) + _in_memory_loggers.append(v2_logger) + return v2_logger + + def _maybe_auto_initialize_arize_phoenix(_in_memory_loggers: list) -> None: """ Auto-initialize ArizePhoenixLogger when Phoenix env vars are detected. @@ -5133,13 +5273,17 @@ class StandardLoggingPayloadSetup: ) -> StandardLoggingPayloadErrorInformation: from litellm.constants import MAXIMUM_TRACEBACK_LINES_TO_LOG - # Check for 'code' first (used by ProxyException), then fall back to 'status_code' (used by LiteLLM exceptions) - # Ensure error_code is always a string for Prisma Python JSON field compatibility + # ProxyException uses .code, LiteLLM exceptions use .status_code, + # httpx.HTTPStatusError exposes status only as .response.status_code. + # Stringified for Prisma JSON compatibility. error_code_attr = getattr(original_exception, "code", None) if error_code_attr is not None and str(error_code_attr) not in ("", "None"): error_status: str = str(error_code_attr) else: status_code_attr = getattr(original_exception, "status_code", None) + if status_code_attr is None: + response_attr = getattr(original_exception, "response", None) + status_code_attr = getattr(response_attr, "status_code", None) error_status = str(status_code_attr) if status_code_attr is not None else "" error_class: str = ( str(original_exception.__class__.__name__) if original_exception else "" diff --git a/litellm/litellm_core_utils/llm_cost_calc/utils.py b/litellm/litellm_core_utils/llm_cost_calc/utils.py index 59d0465e6d4..882561ed2e8 100644 --- a/litellm/litellm_core_utils/llm_cost_calc/utils.py +++ b/litellm/litellm_core_utils/llm_cost_calc/utils.py @@ -9,6 +9,7 @@ from litellm.types.utils import ( CacheCreationTokenDetails, CallTypes, CompletionTokensDetailsWrapper, + DataResidency, ImageResponse, ModelInfo, PassthroughCallTypes, @@ -29,6 +30,9 @@ _IMAGE_RESPONSE_CALL_TYPES = frozenset( } ) +# Pre-resolved DataResidency enum values for fast membership checks +_VALID_DATA_RESIDENCIES = frozenset(r.value for r in DataResidency) + def _is_above_128k(tokens: float) -> bool: if tokens > 128000: @@ -617,11 +621,46 @@ def _calculate_input_cost( return prompt_cost +def _get_regional_uplift_multiplier( + model_info: ModelInfo, data_residency: Optional[str] +) -> float: + """ + Resolve the per-model regional-processing uplift multiplier for a given + data-residency region. + + OpenAI applies a flat percentage uplift (e.g. +10%) on all token costs for + requests served from a regionalized hostname (eu./us.api.openai.com). The + multiplier is stored on the model entry as + ``regional_processing_uplift_multiplier_`` (e.g. 1.10). + + Returns 1.0 (no uplift) when ``data_residency`` is ``None`` or when the + model has no multiplier configured for the given region. + """ + if data_residency is None: + return 1.0 + residency = data_residency.lower() + if residency not in _VALID_DATA_RESIDENCIES: + return 1.0 + multiplier = model_info.get(f"regional_processing_uplift_multiplier_{residency}") + if multiplier is None: + return 1.0 + try: + return float(cast(float, multiplier)) + except (TypeError, ValueError): + verbose_logger.exception( + "Invalid regional_processing_uplift_multiplier_%s for model; " + "defaulting to 1.0", + residency, + ) + return 1.0 + + def generic_cost_per_token( # noqa: PLR0915 model: str, usage: Usage, custom_llm_provider: str, service_tier: Optional[str] = None, + data_residency: Optional[str] = None, ) -> Tuple[float, float]: """ Calculates the cost per token for a given model, prompt tokens, and completion tokens. @@ -631,6 +670,8 @@ def generic_cost_per_token( # noqa: PLR0915 Input: - model: str, the model name without provider prefix - usage: LiteLLM Usage block, containing anthropic caching information + - data_residency: optional OpenAI data-residency region (e.g. "eu", "us"), + used to apply the per-model regional-processing uplift multiplier. Returns: Tuple[float, float] - prompt_cost_in_usd, completion_cost_in_usd @@ -781,6 +822,14 @@ def generic_cost_per_token( # noqa: PLR0915 ) completion_cost += float(image_tokens) * _output_cost_per_image_token + ## REGIONAL DATA-RESIDENCY UPLIFT + # Applied as a flat multiplier across all token costs for the request + # when the upstream is a regionalized OpenAI host (eu./us.api.openai.com). + uplift = _get_regional_uplift_multiplier(model_info, data_residency) + if uplift != 1.0: + prompt_cost *= uplift + completion_cost *= uplift + return prompt_cost, completion_cost diff --git a/litellm/litellm_core_utils/prompt_templates/common_utils.py b/litellm/litellm_core_utils/prompt_templates/common_utils.py index f417b4a5f61..b44d21368f8 100644 --- a/litellm/litellm_core_utils/prompt_templates/common_utils.py +++ b/litellm/litellm_core_utils/prompt_templates/common_utils.py @@ -3,6 +3,7 @@ Common utility functions used for translating messages across providers """ import io +import json import mimetypes import re from os import PathLike @@ -20,6 +21,7 @@ from typing import ( cast, ) +import litellm from litellm import verbose_logger from litellm.router_utils.batch_utils import InMemoryFile from litellm.types.llms.openai import ( @@ -131,6 +133,39 @@ def strip_none_values_from_message(message: AllMessageValues) -> AllMessageValue return cast(AllMessageValues, {k: v for k, v in message.items() if v is not None}) +def extract_search_results_text(search_results: object) -> str: + """ + Extract model-visible text from OpenAI tool-message ``search_results``. + + Used by token estimators and TPM limiters so large search result payloads + cannot bypass preflight checks via a small ``content`` field. + + Counts every string field forwarded on Bedrock ``SearchResultBlock``: + ``source``, ``title``, ``content[].text``, and ``citations``. + """ + if not isinstance(search_results, list): + return "" + texts = "" + for result in search_results: + if not isinstance(result, dict): + continue + for key in ("source", "title"): + value = result.get(key) + if isinstance(value, str): + texts += value + content = result.get("content") + if isinstance(content, list): + for block in content: + if isinstance(block, dict): + text = block.get("text") + if isinstance(text, str): + texts += text + citations = result.get("citations") + if citations is not None: + texts += json.dumps(citations, separators=(",", ":")) + return texts + + def convert_content_list_to_str( message: Union[AllMessageValues, ChatCompletionResponseMessage], ) -> str: @@ -151,6 +186,7 @@ def convert_content_list_to_str( elif message_content is not None and isinstance(message_content, str): texts = message_content + texts += extract_search_results_text(message.get("search_results")) return texts @@ -1170,9 +1206,16 @@ def migrate_file_to_image_url( ChatCompletionImageUrlObject, ) - file_id = message["file"].get("file_id") - file_data = message["file"].get("file_data") - format = message["file"].get("format") + file_sub = message.get("file") + if file_sub is None: + raise litellm.BadRequestError( + message="Content block has type='file' but is missing the required 'file' field", + model=None, + llm_provider=None, + ) + file_id = file_sub.get("file_id") + file_data = file_sub.get("file_data") + format = file_sub.get("format") if not file_id and not file_data: raise ValueError("file_id and file_data are both None") image_url_object = ChatCompletionImageObject( @@ -1196,12 +1239,8 @@ def get_last_user_message(messages: List[AllMessageValues]) -> Optional[str]: {"role": "assistant", "content": "I'm good, thank you!"}, {"role": "user", "content": "What is the weather in Tokyo?"}, ] - get_user_prompt(messages) -> "What is the weather in Tokyo?" + get_last_user_message(messages) -> "What is the weather in Tokyo?" """ - from litellm.litellm_core_utils.prompt_templates.common_utils import ( - convert_content_list_to_str, - ) - if not messages: return None diff --git a/litellm/litellm_core_utils/prompt_templates/factory.py b/litellm/litellm_core_utils/prompt_templates/factory.py index abe9e016e26..46e9b43a429 100644 --- a/litellm/litellm_core_utils/prompt_templates/factory.py +++ b/litellm/litellm_core_utils/prompt_templates/factory.py @@ -1233,6 +1233,7 @@ def infer_protocol_value( def _gemini_tool_call_invoke_helper( function_call_params: ChatCompletionToolCallFunctionChunk, + tool_call_id: Optional[str] = None, ) -> Optional[VertexFunctionCall]: name = function_call_params.get("name", "") or "" arguments = function_call_params.get("arguments", "") @@ -1248,6 +1249,10 @@ def _gemini_tool_call_invoke_helper( name=name, args=arguments_dict, ) + if tool_call_id: + clean_id = tool_call_id.split(THOUGHT_SIGNATURE_SEPARATOR, 1)[0] + if clean_id: + function_call["id"] = clean_id return function_call @@ -1339,6 +1344,7 @@ def _get_dummy_thought_signature() -> str: def convert_to_gemini_tool_call_invoke( message: ChatCompletionAssistantMessage, model: Optional[str] = None, + custom_llm_provider: Optional[str] = None, ) -> List[VertexPartType]: """ OpenAI tool invokes: @@ -1384,12 +1390,26 @@ def convert_to_gemini_tool_call_invoke( tool_calls = message.get("tool_calls", None) function_call = message.get("function_call", None) + from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import ( + VertexGeminiConfig, + ) + + forward_tool_call_id = bool( + model + and VertexGeminiConfig._forward_gemini_function_call_id( + model, custom_llm_provider + ) + ) + if tool_calls is not None: for idx, tool in enumerate(tool_calls): if "function" in tool: gemini_function_call: Optional[VertexFunctionCall] = ( _gemini_tool_call_invoke_helper( - function_call_params=tool["function"] + function_call_params=tool["function"], + tool_call_id=( + tool.get("id") if forward_tool_call_id else None + ), ) ) if gemini_function_call is not None: @@ -1429,10 +1449,6 @@ def convert_to_gemini_tool_call_invoke( thought_signature = provider_fields.get("thought_signature") # If no signature found and model is gemini-3, use dummy signature - from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import ( - VertexGeminiConfig, - ) - if ( not thought_signature and model @@ -1462,6 +1478,8 @@ def convert_to_gemini_tool_call_invoke( def convert_to_gemini_tool_call_result( # noqa: PLR0915 message: Union[ChatCompletionToolMessage, ChatCompletionFunctionMessage], last_message_with_tool_calls: Optional[dict], + model: Optional[str] = None, + custom_llm_provider: Optional[str] = None, ) -> Union[VertexPartType, List[VertexPartType]]: """ OpenAI message with a tool result looks like: @@ -1602,6 +1620,23 @@ def convert_to_gemini_tool_call_result( # noqa: PLR0915 ): name = tool.get("function", {}).get("name", "") + # Echo the OpenAI tool_call_id on functionResponse (strip thought-signature suffix). + # Only Google AI Studio Gemini 3+ accepts `id` on function_response parts. + # Vertex AI and older Gemini models reject the field with HTTP 400. + from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import ( + VertexGeminiConfig, + ) + + gemini_call_id: Optional[str] = None + if model and VertexGeminiConfig._forward_gemini_function_call_id( + model, custom_llm_provider + ): + raw_tool_call_id = message.get("tool_call_id") + if raw_tool_call_id and isinstance(raw_tool_call_id, str): + stripped_id = raw_tool_call_id.split(THOUGHT_SIGNATURE_SEPARATOR, 1)[0] + if stripped_id: + gemini_call_id = stripped_id + if not name: raise Exception( "Missing corresponding tool call for tool response message. Received - message={}, last_message_with_tool_calls={}".format( @@ -1632,6 +1667,8 @@ def convert_to_gemini_tool_call_result( # noqa: PLR0915 name=name, response=response_data, # type: ignore ) + if gemini_call_id: + _function_response["id"] = gemini_call_id # Create part with function_response, and optionally inline_data for images (Computer Use) _part: VertexPartType = {"function_response": _function_response} @@ -2057,9 +2094,16 @@ def anthropic_process_openai_file_message( AnthropicMessagesContainerUploadParam, ]: file_message = cast(ChatCompletionFileObject, message) - file_data = file_message["file"].get("file_data") - file_id = file_message["file"].get("file_id") - format = file_message["file"].get("format") + file_sub = file_message.get("file") + if file_sub is None: + raise litellm.BadRequestError( + message="Content block has type='file' but is missing the required 'file' field", + model=None, + llm_provider="anthropic", + ) + file_data = file_sub.get("file_data") + file_id = file_sub.get("file_id") + format = file_sub.get("format") if file_data: image_chunk = convert_to_anthropic_image_obj( openai_image_url=file_data, @@ -3614,6 +3658,7 @@ from litellm.types.llms.bedrock import ( ToolInputSchemaBlock as BedrockToolInputSchemaBlock, ) from litellm.types.llms.bedrock import ToolJsonSchemaBlock as BedrockToolJsonSchemaBlock +from litellm.types.llms.bedrock import SearchResultBlock from litellm.types.llms.bedrock import ToolResultBlock as BedrockToolResultBlock from litellm.types.llms.bedrock import ( ToolResultContentBlock as BedrockToolResultContentBlock, @@ -3953,7 +3998,7 @@ def _convert_to_bedrock_tool_call_invoke( for tool in tool_calls: if "function" in tool: tool_id = tool["id"] - name = tool["function"].get("name", "") + name = make_valid_bedrock_tool_name(tool["function"].get("name", "")) arguments = tool["function"].get("arguments", "") if not arguments or not arguments.strip(): @@ -4019,6 +4064,122 @@ def _convert_to_bedrock_tool_call_invoke( ) +def _append_bedrock_tool_result_media_block( + tool_result_content_blocks: List[BedrockToolResultContentBlock], + processed_block: BedrockContentBlock, + content: dict, + content_type: str, +) -> None: + if "image" in processed_block: + tool_result_content_blocks.append( + BedrockToolResultContentBlock(image=processed_block["image"]) + ) + elif "document" in processed_block: + tool_result_content_blocks.append( + BedrockToolResultContentBlock(document=processed_block["document"]) + ) + else: + verbose_logger.warning( + "Bedrock Converse: unrecognized BedrockContentBlock keys " + "%s for %s tool-result block %s; dropping.", + list(processed_block.keys()), + content_type, + content, + ) + + +def _append_bedrock_tool_result_image_url_block( + tool_result_content_blocks: List[BedrockToolResultContentBlock], + content: dict, +) -> None: + format: Optional[str] = None + if isinstance(content["image_url"], dict): + image_url = content["image_url"]["url"] + format = content["image_url"].get("format") + else: + image_url = content["image_url"] + processed_block = BedrockImageProcessor.process_image_sync( + image_url=image_url, + format=format, + ) + _append_bedrock_tool_result_media_block( + tool_result_content_blocks, processed_block, content, "image_url" + ) + + +def _append_bedrock_tool_result_file_block( + tool_result_content_blocks: List[BedrockToolResultContentBlock], + content: dict, +) -> None: + # Match the user-message path (_process_file_message): accept either + # file_data (base64 data URI) or file_id (server-side reference / URL). + file_obj = content.get("file") or {} + file_data = file_obj.get("file_data") + file_id = file_obj.get("file_id") + if file_data is None and file_id is None: + raise litellm.BadRequestError( + message="file_data and file_id cannot both be None. Got={}".format(content), + model="", + llm_provider="bedrock", + ) + processed_block = BedrockImageProcessor.process_image_sync( + image_url=cast(str, file_id or file_data), + format=file_obj.get("format"), + ) + _append_bedrock_tool_result_media_block( + tool_result_content_blocks, processed_block, content, "file" + ) + + +def _parse_bedrock_tool_result_content_list( + content_list: List, +) -> List[BedrockToolResultContentBlock]: + tool_result_content_blocks: List[BedrockToolResultContentBlock] = [] + for content in content_list: + if content["type"] == "text": + tool_result_content_blocks.append( + BedrockToolResultContentBlock(text=content["text"]) + ) + elif content["type"] == "image_url": + _append_bedrock_tool_result_image_url_block( + tool_result_content_blocks, content + ) + elif content["type"] == "file": + _append_bedrock_tool_result_file_block(tool_result_content_blocks, content) + return tool_result_content_blocks + + +def _build_bedrock_tool_result_content_blocks( + message: Union[ChatCompletionToolMessage, ChatCompletionFunctionMessage], +) -> tuple[List[BedrockToolResultContentBlock], bool]: + # Optional OpenAI tool-message extension: + # allow structured Bedrock search results on tool messages and map them + # directly to toolResult.content[].searchResult for Converse API. + # + # If `search_results` is present, we intentionally prefer it over `content` + # to avoid generating mixed text + searchResult blocks. + search_results = message.get("search_results") + if isinstance(search_results, list): + tool_result_content_blocks: List[BedrockToolResultContentBlock] = [] + for result in search_results: + if not isinstance(result, dict): + continue + tool_result_content_blocks.append( + BedrockToolResultContentBlock( + searchResult=cast(SearchResultBlock, result) + ) + ) + if tool_result_content_blocks: + return tool_result_content_blocks, True + + message_content = message["content"] + if isinstance(message_content, str): + return [BedrockToolResultContentBlock(text=message_content)], False + if isinstance(message_content, List): + return _parse_bedrock_tool_result_content_list(message_content), False + return [], False + + def _convert_to_bedrock_tool_call_result( message: Union[ChatCompletionToolMessage, ChatCompletionFunctionMessage], ) -> BedrockContentBlock: @@ -4062,90 +4223,18 @@ def _convert_to_bedrock_tool_call_result( """ - """ - tool_result_content_blocks: List[BedrockToolResultContentBlock] = [] - if isinstance(message["content"], str): - tool_result_content_blocks.append( - BedrockToolResultContentBlock(text=message["content"]) - ) - elif isinstance(message["content"], List): - content_list = message["content"] - for content in content_list: - if content["type"] == "text": - tool_result_content_blocks.append( - BedrockToolResultContentBlock(text=content["text"]) - ) - elif content["type"] == "image_url": - format: Optional[str] = None - if isinstance(content["image_url"], dict): - image_url = content["image_url"]["url"] - format = content["image_url"].get("format") - else: - image_url = content["image_url"] - _block: BedrockContentBlock = BedrockImageProcessor.process_image_sync( - image_url=image_url, - format=format, - ) - if "image" in _block: - tool_result_content_blocks.append( - BedrockToolResultContentBlock(image=_block["image"]) - ) - elif "document" in _block: - tool_result_content_blocks.append( - BedrockToolResultContentBlock(document=_block["document"]) - ) - else: - verbose_logger.warning( - "Bedrock Converse: unrecognized BedrockContentBlock keys " - "%s for image_url tool-result block %s; dropping.", - list(_block.keys()), - content, - ) - elif content["type"] == "file": - # Match the user-message path (_process_file_message): accept - # either file_data (base64 data URI) or file_id (server-side - # reference / URL) and hand off to BedrockImageProcessor. Raise - # BadRequestError on both-None rather than silently dropping. - file_obj = content.get("file") or {} - file_data = file_obj.get("file_data") - file_id = file_obj.get("file_id") - if file_data is None and file_id is None: - raise litellm.BadRequestError( - message="file_data and file_id cannot both be None. Got={}".format( - content - ), - model="", - llm_provider="bedrock", - ) - file_format = file_obj.get("format") - _file_block: BedrockContentBlock = ( - BedrockImageProcessor.process_image_sync( - image_url=cast(str, file_id or file_data), - format=file_format, - ) - ) - if "document" in _file_block: - tool_result_content_blocks.append( - BedrockToolResultContentBlock(document=_file_block["document"]) - ) - elif "image" in _file_block: - tool_result_content_blocks.append( - BedrockToolResultContentBlock(image=_file_block["image"]) - ) - else: - verbose_logger.warning( - "Bedrock Converse: unrecognized BedrockContentBlock keys " - "%s for file tool-result block %s; dropping.", - list(_file_block.keys()), - content, - ) + tool_result_content_blocks, used_search_results = ( + _build_bedrock_tool_result_content_blocks(message) + ) message.get("name", "") id = str(message.get("tool_call_id", str(uuid.uuid4()))) tool_result = BedrockToolResultBlock( - content=tool_result_content_blocks, - toolUseId=id, + content=tool_result_content_blocks, toolUseId=id ) + if used_search_results: + tool_result["status"] = cast(Literal["success"], "success") content_block = BedrockContentBlock(toolResult=tool_result) @@ -4879,7 +4968,13 @@ class BedrockConverseMessagesProcessor: @staticmethod def _process_file_message(message: ChatCompletionFileObject) -> BedrockContentBlock: - file_message = message["file"] + file_message = message.get("file") + if file_message is None: + raise litellm.BadRequestError( + message="Content block has type='file' but is missing the required 'file' field", + model=None, + llm_provider="bedrock", + ) file_data = file_message.get("file_data") file_id = file_message.get("file_id") @@ -4900,7 +4995,13 @@ class BedrockConverseMessagesProcessor: async def _async_process_file_message( message: ChatCompletionFileObject, ) -> BedrockContentBlock: - file_message = message["file"] + file_message = message.get("file") + if file_message is None: + raise litellm.BadRequestError( + message="Content block has type='file' but is missing the required 'file' field", + model=None, + llm_provider="bedrock", + ) file_data = file_message.get("file_data") file_id = file_message.get("file_id") format = file_message.get("format") @@ -4977,8 +5078,9 @@ class BedrockConverseMessagesProcessor: ) if reasoning_text and not reasoning_text.get("signature"): reasoning_text_text = reasoning_text["text"] - assistants_part = BedrockContentBlock(text=reasoning_text_text) - assistant_parts.append(assistants_part) + if reasoning_text_text.strip(): + assistants_part = BedrockContentBlock(text=reasoning_text_text) + assistant_parts.append(assistants_part) else: filtered_thinking_blocks.append(block) if len(filtered_thinking_blocks) > 0: @@ -5266,16 +5368,10 @@ def _bedrock_converse_messages_pt( # noqa: PLR0915 def make_valid_bedrock_tool_name(input_tool_name: str) -> str: - """ - Replaces any invalid characters in the input tool name with underscores - and ensures the resulting string is a valid identifier for Bedrock tools - """ + """Normalize tool names to Bedrock pattern [a-zA-Z][a-zA-Z0-9_-]*.""" def replace_invalid(char): - """ - Bedrock tool names only supports alpha-numeric characters and underscores - """ - if char.isalnum() or char == "_": + if char.isalnum() or char in ("_", "-"): return char return "_" @@ -5435,7 +5531,7 @@ def _bedrock_tools_pt( raw_name = f"litellm_unnamed_tool_{tool_idx}" # related issue: https://github.com/BerriAI/litellm/issues/5007 - # Bedrock tool names must satisfy regular expression pattern: [a-zA-Z][a-zA-Z0-9_]* ensure this is true + # Bedrock tool names must satisfy pattern: [a-zA-Z][a-zA-Z0-9_-]* name = make_valid_bedrock_tool_name(input_tool_name=raw_name) if _tool_description: # bedrock doesn't accept empty "" or None descriptions description = _tool_description @@ -5533,9 +5629,7 @@ def default_response_schema_prompt(response_schema: dict) -> str: prompt_str = """Use this JSON schema: ```json {} - ```""".format( - response_schema - ) + ```""".format(response_schema) return prompt_str diff --git a/litellm/litellm_core_utils/realtime_streaming.py b/litellm/litellm_core_utils/realtime_streaming.py index c4528ff74e3..33bb6d7d2ea 100644 --- a/litellm/litellm_core_utils/realtime_streaming.py +++ b/litellm/litellm_core_utils/realtime_streaming.py @@ -86,6 +86,12 @@ class RealTimeStreaming: # When a text message is blocked, hold the guardrail reason so the next # response.create can be rewritten to include the failure context. self._pending_guardrail_message: Optional[str] = None + # Track whether session.created has already been sent to the client + # (e.g. synthetic event in deferred setup mode). + self._session_created_sent_to_client: bool = False + # Track whether we have already sent the guardrail turn-detection update + # that disables provider auto-response for transcription guardrails. + self._guardrail_turn_detection_update_sent: bool = False _SESSION_EVENT_TYPES = frozenset(["session.created", "session.updated"]) _AUDIO_FORMAT_MAP: Dict[str, Dict[str, Any]] = { @@ -248,40 +254,82 @@ class RealTimeStreaming: ## SYNC LOGGING executor.submit(self.logging_obj.success_handler(self.messages)) - async def _send_to_backend(self, message: str) -> None: + async def _send_to_backend(self, message: str) -> bool: """Send a message to the backend WebSocket. If a provider_config is set the message is first passed through transform_realtime_request so that provider-specific translation (e.g. dropping session.update for Vertex AI) is applied even for guardrail-injected messages. + + Returns True if at least one message was actually delivered to the + backend, False if the provider transformation produced no output and + the message was effectively dropped. """ if self.provider_config: transformed = self.provider_config.transform_realtime_request( message, self.model, self.session_configuration_request ) + sent = False for msg in transformed: + # Send first; only cache the setup payload once the backend + # has actually accepted it. Caching before send would leave + # ``session_configuration_request`` populated after a failed + # send, causing subsequent client session.update messages to + # be treated as "subsequent" and dropped even though the + # backend never received the original setup. await self.backend_ws.send(msg) # type: ignore[union-attr, attr-defined] - else: - await self.backend_ws.send(message) # type: ignore[union-attr, attr-defined] + self._cache_session_configuration_request(msg) + sent = True + return sent + await self.backend_ws.send(message) # type: ignore[union-attr, attr-defined] + return True + + def _cache_session_configuration_request(self, transformed_message: str) -> None: + """Store setup payload once sent to backend. + + Updates the cached setup on every successful setup send so follow-up + ``session.update`` messages (which produce a merged setup with new + ``generationConfig`` / ``systemInstruction`` / etc.) are reflected in + the cache used by downstream readers (``transform_session_created_event``, + ``return_new_content_delta_events`` modality lookup, ...). + """ + try: + message_obj = json.loads(transformed_message) + if "setup" in message_obj: + self.session_configuration_request = transformed_message + except (json.JSONDecodeError, TypeError): + return def _make_disable_auto_response_message(self) -> str: """Return a session.update that disables VAD auto-response.""" + turn_detection: Dict[str, Any] = { + "type": "server_vad", + "create_response": False, + } if self._backend_uses_beta_protocol: - session: Dict[str, Any] = { - "turn_detection": {"create_response": False}, - } + session: Dict[str, Any] = {"turn_detection": turn_detection} else: session = { "type": "realtime", - "audio": { - "input": { - "turn_detection": {"create_response": False}, - } - }, + "audio": {"input": {"turn_detection": turn_detection}}, } return json.dumps({"type": "session.update", "session": session}) + async def _maybe_send_guardrail_turn_detection_update(self) -> None: + """Disable provider auto-response once when transcription guardrails are enabled.""" + if self._guardrail_turn_detection_update_sent: + return + if not self._has_audio_transcription_guardrails(): + return + sent = await self._send_to_backend(self._make_disable_auto_response_message()) + # Only mark as sent when the provider transformation actually delivered + # the update to the backend. Otherwise (e.g. Gemini drops session.update + # after the initial setup), leave the flag unset so future opportunities + # — such as a duplicate session.created — can retry. + if sent: + self._guardrail_turn_detection_update_sent = True + def _has_realtime_guardrails(self) -> bool: """Return True if any callback is registered for realtime guardrail event types.""" from litellm.integrations.custom_guardrail import CustomGuardrail @@ -320,12 +368,20 @@ class RealTimeStreaming: self, transcript: str, item_id: Optional[str] = None, + pre_block_backend_message: Optional[str] = None, ) -> bool: """ Run registered guardrails on a completed speech transcription. Returns True if blocked (synthetic warning already sent to client). Returns False if clean (caller should send response.create to the backend). + + ``pre_block_backend_message`` (if provided) is sent to the backend + BEFORE any of the guardrail's own backend messages when a block is + triggered. This is needed for protocol contracts that require a + specific message to be sent first — e.g. Gemini Live requires a + matching ``toolResponse`` immediately after a ``toolCall`` before any + other client messages can be accepted. """ from litellm.integrations.custom_guardrail import CustomGuardrail from litellm.types.guardrails import GuardrailEventHooks @@ -385,6 +441,13 @@ class RealTimeStreaming: getattr(callback, "realtime_violation_message", None) or safe_msg ) + # Deliver any caller-supplied backend message FIRST so that + # protocol contracts requiring a specific ordering (e.g. + # Gemini Live's mandatory ``toolResponse`` after a + # ``toolCall``) are honored before the guardrail's own + # clientContent / cancel messages are sent. + if pre_block_backend_message is not None: + await self._send_to_backend(pre_block_backend_message) # Cancel any in-progress LLM response (e.g. VAD auto-response). await self._send_to_backend(json.dumps({"type": "response.cancel"})) # Send the policy violation hint (shows as small gray status text in UI). @@ -480,16 +543,34 @@ class RealTimeStreaming: else [transformed_response] ) for event in events: + is_session_created_event = ( + isinstance(event, dict) and event.get("type") == "session.created" + ) + if is_session_created_event: + if self._session_created_sent_to_client: + # A synthetic session.created (with placeholder defaults) was + # already forwarded to the client when we connected. The + # provider's real session.created (e.g. emitted from Gemini + # `setupComplete`) carries the authoritative modalities/model + # from the client's session.update. Re-emit it as + # `session.updated` so the client learns the corrected + # configuration without seeing two `session.created` events. + event = {**event, "type": "session.updated"} + else: + self._session_created_sent_to_client = True event_str = json.dumps(event) - ## For audio/VAD guardrail path: forward session.created first, then inject. - if ( - isinstance(event, dict) - and event.get("type") == "session.created" - and self._has_audio_transcription_guardrails() - ): + ## For audio/VAD guardrail path: forward the (possibly retyped) + ## session.created first, then invoke the one-time guardrail + ## turn-detection update. ``_maybe_send_guardrail_turn_detection_update`` + ## is idempotent (gated by ``_guardrail_turn_detection_update_sent``), + ## so duplicate session.created events — including those emitted + ## after a synthetic session.created from ``llm_http_handler`` in + ## deferred-setup mode — still get a single chance to inject the + ## update if a prior attempt was dropped by the provider transform. + if is_session_created_event and self._has_audio_transcription_guardrails(): self.store_message(event_str) await self.websocket.send_text(event_str) - await self._send_to_backend(self._make_disable_auto_response_message()) + await self._maybe_send_guardrail_turn_detection_update() continue ## GUARDRAIL: run on transcription events in provider_config path too if ( @@ -564,10 +645,19 @@ class RealTimeStreaming: try: raw_response = await self.backend_ws.recv( # type: ignore[union-attr] decode=False - ) # improves performance + ) except TypeError: raw_response = await self.backend_ws.recv() # type: ignore[union-attr, assignment] + if isinstance(raw_response, bytes): + try: + raw_response = raw_response.decode("utf-8") + except UnicodeDecodeError: + verbose_logger.warning( + "Received non-UTF-8 binary frame from backend, skipping." + ) + continue + if self.provider_config: try: await self._handle_provider_config_message(raw_response) @@ -783,12 +873,13 @@ class RealTimeStreaming: item["content"] = new_content return item - async def client_ack_messages(self): + async def client_ack_messages(self): # noqa: PLR0915 try: while True: message = await self.websocket.receive_text() ## GUARDRAIL: intercept conversation.item.create for text-based injection. + guardrail_turn_detection_injected = False try: msg_obj = json.loads(message) msg_type = msg_obj.get("type") @@ -796,7 +887,68 @@ class RealTimeStreaming: if msg_type == "conversation.item.create": # Check user text messages for prompt injection item = msg_obj.get("item", {}) - if item.get("role") == "user": + # Check function_call_output first so a client cannot + # bypass the tool-result guardrail by also setting + # role="user" on a function_call_output item. + if item.get("type") == "function_call_output": + # Tool results are client-controlled and fed to the + # model; check them with the same guardrail used for + # user text so an attacker cannot smuggle blocked + # content into a function_call_output. + output = item.get("output", "") + output_text = ( + output + if isinstance(output, str) + else json.dumps(output) + ) + if output_text: + # Build the sanitized function_call_output up + # front so we can hand it to the guardrail + # runner as the pre-block message. Providers + # that pair every toolCall with a toolResponse + # (e.g. Gemini/Vertex Live) require the + # toolResponse to arrive BEFORE any other + # client message — otherwise the guardrail's + # own clientContent would violate the + # pending-tool-call protocol contract and the + # backend could close the connection before + # the sanitized response ever lands. Dropping + # the blocked item outright would similarly + # leave such providers waiting indefinitely. + # The sanitized payload carries no blocked + # content — only a generic policy marker. + sanitized_msg = json.dumps( + { + **msg_obj, + "item": { + **item, + "output": json.dumps( + { + "error": "Tool output blocked by content policy", + } + ), + }, + } + ) + blocked = await self.run_realtime_guardrails( + output_text, + pre_block_backend_message=sanitized_msg, + ) + if blocked: + # ``_pending_guardrail_message`` is + # intentionally NOT set here. That flag + # exists to swallow the reflexive + # ``response.create`` an OpenAI client + # sends immediately after a user text + # message. In a tool-calling flow the + # client may not send a ``response.create`` + # at all (e.g. Gemini SDKs auto-respond), + # so leaving the flag set would + # incorrectly drop an unrelated + # ``response.create`` from a later + # interaction turn. + continue + elif item.get("role") == "user": content_list = item.get("content", []) texts = [ c.get("text", "") @@ -824,6 +976,89 @@ class RealTimeStreaming: self._pending_guardrail_message = None continue + ## GUARDRAIL: Inject turn_detection into first session.update + # if needed. Done BEFORE the GA remap so the injected + # ``create_response`` rides along with any client-provided + # turn_detection fields (e.g. silence_duration_ms) into the + # nested ``audio.input.turn_detection`` path produced by the + # remap. Doing this after the remap would create a separate + # minimal root-level ``turn_detection`` and silently drop + # the client's nested settings. + if ( + msg_type == "session.update" + and self.session_configuration_request is None + and not self._guardrail_turn_detection_update_sent + and self._has_audio_transcription_guardrails() + ): + session = msg_obj.setdefault("session", {}) + if isinstance(session, dict): + existing_td = session.get("turn_detection") + if not isinstance(existing_td, dict): + existing_td = {} + existing_td["create_response"] = False + session["turn_detection"] = existing_td + message = json.dumps(msg_obj) + guardrail_turn_detection_injected = True + verbose_logger.debug( + "Injected turn_detection into first session.update for audio transcription guardrails" + ) + + ## GUARDRAIL: Force ``create_response`` to False in any + # client-provided ``turn_detection`` so a later + # ``session.update`` cannot re-enable VAD auto-response + # and bypass the transcription guardrail after the + # initial disable. Covers both the flat beta key and the + # nested GA ``audio.input.turn_detection`` shape, since + # the GA remap below also accepts either form. Skipped + # when the injection block above already ran for this + # message, to avoid redundant double-serialization. + if ( + msg_type == "session.update" + and not guardrail_turn_detection_injected + and self._has_audio_transcription_guardrails() + ): + session = msg_obj.get("session") + if isinstance(session, dict): + td_overridden = False + flat_td = session.get("turn_detection") + flat_td_present = flat_td is not None + if flat_td_present: + if not isinstance(flat_td, dict): + flat_td = {} + if flat_td.get("create_response") is not False: + flat_td["create_response"] = False + session["turn_detection"] = flat_td + td_overridden = True + nested_td_present = False + audio = session.get("audio") + if isinstance(audio, dict): + audio_input = audio.get("input") + if isinstance(audio_input, dict): + nested_td = audio_input.get("turn_detection") + if nested_td is not None: + nested_td_present = True + if not isinstance(nested_td, dict): + nested_td = {} + if ( + nested_td.get("create_response") + is not False + ): + nested_td["create_response"] = False + audio_input["turn_detection"] = nested_td + td_overridden = True + # Symmetric with the first-update injection block: + # if the client omitted turn_detection entirely on + # a subsequent session.update, still inject the + # ``create_response: False`` override so the + # transcription guardrail cannot be re-enabled by + # any downstream merge that drops the original + # disable. + if not flat_td_present and not nested_td_present: + session["turn_detection"] = {"create_response": False} + td_overridden = True + if td_overridden: + message = json.dumps(msg_obj) + # GA compatibility: remap beta-style session fields only when # the upstream is in GA mode. Beta upstreams expect the flat # session shape unchanged. @@ -841,17 +1076,20 @@ class RealTimeStreaming: pass ## LOGGING + # Log after any in-place modifications (GA remap, guardrail + # turn_detection injection) so audit logs reflect what we + # actually forward to the backend. self.store_input(message=message) - ## FORWARD TO BACKEND - if self.provider_config: - message = self.provider_config.transform_realtime_request( - message, self.model - ) - for msg in message: - await self.backend_ws.send(msg) # type: ignore[union-attr] - else: - await self.backend_ws.send(message) # type: ignore[union-attr] + ## FORWARD TO BACKEND + # Only mark the guardrail turn_detection update as sent after the + # backend actually accepted the message. Setting the flag earlier + # would permanently disable the injection if ``_send_to_backend`` + # raised — neither this loop nor + # ``_maybe_send_guardrail_turn_detection_update`` would retry. + sent = await self._send_to_backend(message) + if guardrail_turn_detection_injected and sent: + self._guardrail_turn_detection_update_sent = True except Exception as e: verbose_logger.debug(f"Error in client ack messages: {e}") diff --git a/litellm/litellm_core_utils/sensitive_data_masker.py b/litellm/litellm_core_utils/sensitive_data_masker.py index d7803455b4a..4928dd08386 100644 --- a/litellm/litellm_core_utils/sensitive_data_masker.py +++ b/litellm/litellm_core_utils/sensitive_data_masker.py @@ -146,6 +146,37 @@ class SensitiveDataMasker: return masked_data +_default_masker = SensitiveDataMasker() + + +def mask_sensitive_keys( + data: Dict[str, Any], sensitive_fields: Set[str] +) -> Dict[str, Any]: + """Return a new dict with values masked for keys listed in ``sensitive_fields``. + + Unlike :meth:`SensitiveDataMasker.mask_dict`, this does exact key-name + matching (not segment matching), so callers explicitly enumerate which + fields to mask. Non-string and None values are passed through unchanged. + + Values shorter than ``visible_prefix + visible_suffix`` (8 by default) + fall outside :meth:`SensitiveDataMasker._mask_value`'s partial-reveal + range and are replaced with a fixed-length all-mask string, so a short + credential is never returned verbatim. + """ + masked: Dict[str, Any] = {} + mask_char = _default_masker.mask_char + min_visible = _default_masker.visible_prefix + _default_masker.visible_suffix + for key, value in data.items(): + if value is not None and key in sensitive_fields and isinstance(value, str): + if len(value) < min_visible: + masked[key] = mask_char * len(value) if value else value + else: + masked[key] = _default_masker._mask_value(value) + else: + masked[key] = value + return masked + + # Usage example: """ masker = SensitiveDataMasker() diff --git a/litellm/litellm_core_utils/specialty_caches/dynamic_logging_cache.py b/litellm/litellm_core_utils/specialty_caches/dynamic_logging_cache.py index 13341f27a61..0a6a4e82c72 100644 --- a/litellm/litellm_core_utils/specialty_caches/dynamic_logging_cache.py +++ b/litellm/litellm_core_utils/specialty_caches/dynamic_logging_cache.py @@ -1,9 +1,9 @@ """ This is a cache for LangfuseLoggers. -Langfuse Python SDK initializes a thread for each client. +Langfuse Python SDK initializes a thread for each client. -This ensures we do +This ensures we do 1. Proper cleanup of Langfuse initialized clients. 2. Re-use created langfuse clients. """ diff --git a/litellm/litellm_core_utils/streaming_handler.py b/litellm/litellm_core_utils/streaming_handler.py index fa7faf3035d..55042a733ed 100644 --- a/litellm/litellm_core_utils/streaming_handler.py +++ b/litellm/litellm_core_utils/streaming_handler.py @@ -59,6 +59,8 @@ FUNCTION_CALL_ATTRIBUTE = "function_call" _SYNC_ITER_EXHAUSTED = object() +_GCHUNK_FIELDS: frozenset = frozenset(GChunk.__annotations__) + def _next_sync_or_exhausted(it: Any) -> Any: """ @@ -181,6 +183,30 @@ class CustomStreamWrapper: self.created: Optional[int] = None self._last_returned_hidden_params: Optional[dict] = None + _cached_logging_provider = self.logging_obj.model_call_details.get( + "custom_llm_provider", None + ) + self._cached_logging_llm_provider: Optional[str] = _cached_logging_provider + _effective_model = model or "" + if ( + custom_llm_provider == "openai" + and custom_llm_provider != _cached_logging_provider + ): + _effective_model = "{}/{}".format( + _cached_logging_provider, _effective_model + ) + self._cached_model_name: str = _effective_model + + # Snapshot assumes self._hidden_params is populated from litellm_params + # at init and never mutated during the stream. If that ever changes, + # this cache must be removed. + self._base_hidden_params: Dict[str, Any] = { + **self._hidden_params, + "response_cost": None, + } + + self._post_streaming_hooks: Optional[List] = None + def _check_max_streaming_duration(self) -> None: """Raise litellm.Timeout if the stream has exceeded LITELLM_MAX_STREAMING_DURATION_SECONDS.""" from litellm.constants import LITELLM_MAX_STREAMING_DURATION_SECONDS @@ -681,29 +707,16 @@ class CustomStreamWrapper: def model_response_creator( self, chunk: Optional[dict] = None, hidden_params: Optional[dict] = None ): - _model = self.model - _received_llm_provider = self.custom_llm_provider - _logging_obj_llm_provider = self.logging_obj.model_call_details.get("custom_llm_provider", None) # type: ignore - if ( - _received_llm_provider == "openai" - and _received_llm_provider != _logging_obj_llm_provider - ): - _model = "{}/{}".format(_logging_obj_llm_provider, _model) + _model = self._cached_model_name + _logging_obj_llm_provider = self._cached_logging_llm_provider + if chunk is None: - chunk = {} + args: Dict[str, Any] = {"model": _model} else: - # pop model keyword chunk.pop("model", None) - - chunk_dict = {} - for key, value in chunk.items(): - if key != "stream": - chunk_dict[key] = value - - args = { - "model": _model, - **chunk_dict, - } + args = {"model": _model} + if chunk: + args.update({k: v for k, v in chunk.items() if k != "stream"}) model_response = ModelResponseStream(**args) if self.response_id is not None: @@ -717,15 +730,23 @@ class CustomStreamWrapper: model_response.created = self.created else: self.created = model_response.created + + # Spread order is load-bearing: _base_hidden_params (model_id, api_base, ...) + # must win over both caller-supplied hidden_params and the computed + # custom_llm_provider/created_at values, so it comes last. if hidden_params is not None: - model_response._hidden_params = hidden_params - model_response._hidden_params["custom_llm_provider"] = _logging_obj_llm_provider - model_response._hidden_params["created_at"] = time.time() - model_response._hidden_params = { - **model_response._hidden_params, - **self._hidden_params, - "response_cost": None, - } + model_response._hidden_params = { + **hidden_params, + "custom_llm_provider": _logging_obj_llm_provider, + "created_at": time.time(), + **self._base_hidden_params, + } + else: + model_response._hidden_params = { + "custom_llm_provider": _logging_obj_llm_provider, + "created_at": time.time(), + **self._base_hidden_params, + } if ( len(model_response.choices) > 0 @@ -1627,7 +1648,17 @@ class CustomStreamWrapper: from litellm.integrations.custom_logger import CustomLogger from litellm.types.utils import CallTypes - # Get request kwargs from logging object + if self._post_streaming_hooks is None: + self._post_streaming_hooks = [ + cb + for cb in litellm.callbacks + if isinstance(cb, CustomLogger) + and hasattr(cb, "async_post_call_streaming_deployment_hook") + ] + + if not self._post_streaming_hooks: + return chunk + request_data = self.logging_obj.model_call_details call_type_str = self.logging_obj.call_type @@ -1636,18 +1667,14 @@ class CustomStreamWrapper: except ValueError: typed_call_type = None - # Call hooks for all callbacks - for callback in litellm.callbacks: - if isinstance(callback, CustomLogger) and hasattr( - callback, "async_post_call_streaming_deployment_hook" - ): - result = await callback.async_post_call_streaming_deployment_hook( - request_data=request_data, - response_chunk=chunk, - call_type=typed_call_type, - ) - if result is not None: - chunk = result + for callback in self._post_streaming_hooks: + result = await callback.async_post_call_streaming_deployment_hook( + request_data=request_data, + response_chunk=chunk, + call_type=typed_call_type, + ) + if result is not None: + chunk = result return chunk except Exception as e: @@ -1808,8 +1835,10 @@ class CustomStreamWrapper: processed_chunk, None, None, cache_hit ) ) - ## SYNC LOGGING - self.logging_obj.success_handler(processed_chunk, None, None, cache_hit) + ## SYNC LOGGING — only for sync SDK entrypoints; async proxy paths export via async_success_handler + litellm_params = self.logging_obj.model_call_details.get("litellm_params", {}) + if self.logging_obj._is_sync_litellm_request(litellm_params): + self.logging_obj.success_handler(processed_chunk, None, None, cache_hit) def finish_reason_handler(self): model_response = self.model_response_creator() @@ -1888,17 +1917,15 @@ class CustomStreamWrapper: response = self._add_mcp_list_tools_to_first_chunk(response) self.sent_first_chunk = True - if hasattr( - response, "usage" - ): # remove usage from chunk, only send on final chunk - # Convert the object to a dictionary + # ModelResponseStream declares `usage` as a field, so + # hasattr(response, "usage") is always True — must check + # `is not None` to avoid running this path on every chunk. + if getattr(response, "usage", None) is not None: obj_dict = response.model_dump() - # Remove an attribute (e.g., 'attr2') if "usage" in obj_dict: del obj_dict["usage"] - # Create a new object without the removed attribute response = self.model_response_creator( chunk=obj_dict, hidden_params=response._hidden_params ) @@ -2206,23 +2233,19 @@ class CustomStreamWrapper: cache_hit, ) else: + # prefer_async_handlers routes CustomLogger to async_success_handler + # when consumers use ``async for`` on sync-SDK streams. Legacy string + # callbacks still run via executor.submit inside dispatch_success_handlers. asyncio.create_task( - self.logging_obj.async_success_handler( + self.logging_obj.dispatch_success_handlers( complete_streaming_response, cache_hit=cache_hit, start_time=None, end_time=None, + prefer_async_handlers=True, ) ) - executor.submit( - self.logging_obj.success_handler, - complete_streaming_response, - cache_hit=cache_hit, - start_time=None, - end_time=None, - ) - raise StopAsyncIteration # Re-raise StopIteration else: self.sent_last_chunk = True @@ -2398,10 +2421,7 @@ def generic_chunk_has_all_required_fields(chunk: dict) -> bool: :param chunk: The dictionary to check. :return: True if all required fields are present, False otherwise. """ - _all_fields = GChunk.__annotations__ - - decision = all(key in _all_fields for key in chunk) - return decision + return all(key in _GCHUNK_FIELDS for key in chunk) def convert_generic_chunk_to_model_response_stream( diff --git a/litellm/litellm_core_utils/token_counter.py b/litellm/litellm_core_utils/token_counter.py index e6a68de07e9..74b41062174 100644 --- a/litellm/litellm_core_utils/token_counter.py +++ b/litellm/litellm_core_utils/token_counter.py @@ -486,6 +486,14 @@ def _count_messages( use_default_image_token_count, default_token_count, ) + elif key == "search_results" and isinstance(value, list): + from litellm.litellm_core_utils.prompt_templates.common_utils import ( + extract_search_results_text, + ) + + search_results_text = extract_search_results_text(value) + if search_results_text: + num_tokens += params.count_function(search_results_text) else: # Skip unsupported keys instead of raising an error continue @@ -764,11 +772,29 @@ def _format_function_definitions(tools): lines.append("namespace functions {") lines.append("") for tool in tools: + if not isinstance(tool, dict): + continue function = tool.get("function") + if not isinstance(function, dict): + # Anthropic tool shape → OpenAI function dict for token counting. + params = tool.get("input_schema") or tool.get("parameters") or {} + if not isinstance(params, dict): + params = {} + function = { + "name": tool.get("name"), + "description": tool.get("description"), + "parameters": params, + } + function_name = function.get("name") + if not function_name: + # Skip malformed tools missing a name to avoid emitting + # ``type None = ...`` which would produce inaccurate token counts. + continue if function_description := function.get("description"): lines.append(f"// {function_description}") - function_name = function.get("name") - parameters = function.get("parameters", {}) + parameters = function.get("parameters") or {} + if not isinstance(parameters, dict): + parameters = {} properties = parameters.get("properties") if properties and properties.keys(): lines.append(f"type {function_name} = (_: {{") diff --git a/litellm/llms/anthropic/chat/transformation.py b/litellm/llms/anthropic/chat/transformation.py index 1ce80207552..57609cfcd26 100644 --- a/litellm/llms/anthropic/chat/transformation.py +++ b/litellm/llms/anthropic/chat/transformation.py @@ -337,13 +337,12 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig): ) @staticmethod - def _supports_effort_level(model: str, level: str) -> bool: - """Check ``supports_{level}_reasoning_effort`` in the model map. + def _supports_model_capability(model: str, key: str) -> bool: + """Check a boolean capability ``key`` in the model map. Strips bedrock/vertex prefixes so a provider-routed Claude still resolves to the Anthropic model-map entry. """ - key = f"supports_{level}_reasoning_effort" try: if _supports_factory( model=model, @@ -372,8 +371,6 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig): except Exception: pass try: - import litellm - for cand in candidates: if cand in litellm.model_cost and ( litellm.model_cost[cand].get(key) is True @@ -383,6 +380,13 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig): pass return False + @staticmethod + def _supports_effort_level(model: str, level: str) -> bool: + """Check ``supports_{level}_reasoning_effort`` in the model map.""" + return AnthropicConfig._supports_model_capability( + model, f"supports_{level}_reasoning_effort" + ) + @staticmethod def _validate_effort_for_model(model: str, effort: Optional[str]) -> Optional[str]: """Return ``None`` if ``effort`` is allowed on ``model``, else an error message.""" @@ -400,7 +404,15 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig): @staticmethod def _model_supports_effort_param(model: str) -> bool: - """Whether the model accepts ``output_config.effort`` at all.""" + """Whether the model accepts ``output_config.effort`` at all. + + A model qualifies if its map entry advertises ``supports_output_config`` + or any ``supports_*_reasoning_effort`` flag. The two are independent + signals: e.g. Claude Opus 4.5 supports ``output_config`` without + advertising a non-default (max/xhigh) effort level. + """ + if AnthropicConfig._supports_model_capability(model, "supports_output_config"): + return True return any( AnthropicConfig._supports_effort_level(model, level) for level in ("low", "minimal", "medium", "high", "xhigh", "max") @@ -1506,9 +1518,21 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig): optional_params["metadata"] = {"user_id": value} elif param == "thinking": optional_params["thinking"] = value - elif param == "reasoning_effort" and isinstance(value, str): + elif param == "reasoning_effort": + # Accept both string ("low") and dict ({"effort": "low", + # "summary": "concise"}). The Responses->Chat parser keeps the + # full dict when `summary` is set (see #25359), so a dict here + # is the standard shape Otto/OpenAI-Responses-Bridge callers + # send. Coerce to the effort string before mapping — same + # shape-tolerance the GPT-5 path already implements in + # `_normalize_reasoning_effort_for_chat_completion`. + effort_value = value + if isinstance(effort_value, dict): + effort_value = effort_value.get("effort") + if not isinstance(effort_value, str): + continue mapped_thinking = AnthropicConfig._map_reasoning_effort( - reasoning_effort=value, + reasoning_effort=effort_value, model=model, llm_provider=self.custom_llm_provider or "anthropic", ) @@ -1519,12 +1543,12 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig): optional_params["thinking"] = mapped_thinking if AnthropicConfig._is_adaptive_thinking_model(model): mapped_effort = REASONING_EFFORT_TO_OUTPUT_CONFIG_EFFORT.get( - value + effort_value ) if mapped_effort is None: AnthropicConfig._raise_invalid_reasoning_effort( model=model, - value=value, + value=effort_value, llm_provider=self.custom_llm_provider or "anthropic", ) optional_params["output_config"] = {"effort": mapped_effort} @@ -1781,7 +1805,10 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig): self._ensure_context_management_beta_header( headers, optional_params["context_management"] ) - if optional_params.get("output_format") is not None: + output_config = optional_params.get("output_config") + if optional_params.get("output_format") is not None or ( + isinstance(output_config, dict) and output_config.get("format") is not None + ): self._ensure_beta_header( headers, ANTHROPIC_BETA_HEADER_VALUES.STRUCTURED_OUTPUT_2025_09_25.value ) diff --git a/litellm/llms/anthropic/common_utils.py b/litellm/llms/anthropic/common_utils.py index 869a7c5fbc4..31131d722ab 100644 --- a/litellm/llms/anthropic/common_utils.py +++ b/litellm/llms/anthropic/common_utils.py @@ -832,6 +832,49 @@ def strip_thinking_blocks_from_anthropic_messages_request_dict( data.pop("thinking", None) +def strip_empty_text_blocks_from_anthropic_messages( + messages: List[Any], +) -> List[Any]: + """ + Return a new message list with empty or whitespace-only ``{"type": "text"}`` + content blocks removed. + + Anthropic's API rejects requests containing such blocks with + ``"messages: text content blocks must be non-empty"``, but assistant + messages from Anthropic routinely arrive with ``{"type": "text", "text": ""}`` + alongside ``tool_use`` blocks (see anthropics/anthropic-sdk-python#461). + Multi-turn tool-use clients (e.g. Claude Code) loop these prior responses + back as conversation history, which then causes the next request to 400 + on the unified ``/v1/messages`` path. ``/v1/chat/completions`` already + handles this in ``anthropic_messages_pt``; this helper provides the + equivalent guarantee for the native Anthropic Messages path. + + Messages whose content is a list and becomes empty after stripping are + omitted, matching :func:`strip_thinking_blocks_from_anthropic_messages`. + The caller's list and its content blocks are never mutated; modified + messages are returned as shallow copies with a fresh content list. + """ + out: List[Any] = [] + for m in messages: + if not isinstance(m, dict) or not isinstance(m.get("content"), list): + out.append(m) + continue + content = m["content"] + filtered = [b for b in content if not _is_empty_text_block(b)] + if len(filtered) == len(content): + out.append(m) + elif filtered: + out.append({**m, "content": filtered}) + return out + + +def _is_empty_text_block(block: Any) -> bool: + if not isinstance(block, dict) or block.get("type") != "text": + return False + text = block.get("text") + return not isinstance(text, str) or not text.strip() + + def process_anthropic_headers(headers: Union[httpx.Headers, dict]) -> dict: openai_headers = {} if "anthropic-ratelimit-requests-limit" in headers: diff --git a/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py b/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py index 8ed6126d2eb..efb913f709a 100644 --- a/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py +++ b/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py @@ -4,6 +4,7 @@ from typing import ( AsyncIterator, Coroutine, Dict, + Iterator, List, Optional, Tuple, @@ -12,9 +13,16 @@ from typing import ( ) import litellm +from litellm._logging import verbose_logger +from litellm.litellm_core_utils.asyncify import run_async_function from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import ( AnthropicAdapter, ) +from litellm.llms.anthropic.experimental_pass_through.context_management import ( + AnthropicContextManagementError, + PolyfillResult, + apply_context_management, +) from litellm.llms.anthropic.experimental_pass_through.utils import ( is_reasoning_auto_summary_enabled, ) @@ -28,15 +36,266 @@ if TYPE_CHECKING: pass -# Anthropic-only fields that the translator above already maps into the -# OpenAI-format completion_kwargs (output_config → reasoning_effort / -# response_format, etc.). They must be filtered out of the raw -# extra_kwargs re-merge below or non-Anthropic backends reject the call -# with 400 "Extra inputs are not permitted". Add new entries here when -# extending AnthropicMessagesRequestOptionalParams with another Anthropic- -# specific key. +# Anthropic-only keys already mapped by the translator; strip on extra_kwargs re-merge. ANTHROPIC_ONLY_REQUEST_KEYS: frozenset[str] = frozenset({"output_config"}) + +def _messages_have_compaction_block(messages: List[Dict]) -> bool: + """Return True when any message carries a ``compaction`` content block.""" + for msg in messages: + content = msg.get("content") + if not isinstance(content, list): + continue + for block in content: + if isinstance(block, dict) and block.get("type") == "compaction": + return True + return False + + +def _extract_proxy_litellm_metadata(kwargs: Dict[str, Any]) -> Optional[Dict[str, Any]]: + """Return ``kwargs["litellm_metadata"]`` when it's a dict; ``None`` otherwise. + + The proxy attaches its auth/spend-attribution fields (``user_api_key``, + ``user_api_key_team_id``, ``litellm_call_id``, the full ``UserAPIKeyAuth`` + object under ``user_api_key_auth``, ...) to ``data["litellm_metadata"]`` + for ``/v1/messages`` (see + ``LiteLLMProxyRequestSetup.add_user_api_key_auth_to_request_metadata`` and + ``LITELLM_METADATA_ROUTES``). The Anthropic-shape ``metadata`` arg only + carries ``user_id`` and must not be conflated. Returns ``None`` for SDK + callers that bypass the proxy entirely. + """ + litellm_metadata = kwargs.get("litellm_metadata") + if not isinstance(litellm_metadata, dict): + return None + return litellm_metadata + + +async def _prepare_context_managed_request( + *, + model: str, + messages: List[Dict], + tools: Optional[List[Dict]], + system: Optional[Any], + context_management_spec: Any, + litellm_metadata: Optional[Dict], + drop_params: Optional[bool], + llm_router: Any, + user_api_key_auth: Any = None, +) -> Optional[PolyfillResult]: + """Apply client compaction history, then optional context_management polyfill.""" + from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import ( + apply_client_compaction_block_history, + ) + + # Skip the client-history pre-processing when a ``compact_20260112`` + # polyfill spec will run: that editor already slices around any client-sent + # compaction block in its Phase A (and uses the full post-compaction tail + # for its token-threshold check). Pre-collapsing to just the latest user + # question here would starve the polyfill of conversation context and + # silently drop intermediate turns. + polyfill_will_run = _polyfill_will_run( + context_management_spec=context_management_spec, + drop_params=drop_params, + ) + + if polyfill_will_run: + history_result: Optional[PolyfillResult] = None + working_messages: List[Dict] = messages + working_system: Optional[Any] = system + else: + history_result = apply_client_compaction_block_history( + messages=cast(List[Dict[str, Any]], messages), + system=system, + ) + working_messages = ( + history_result.messages if history_result is not None else messages + ) + working_system = history_result.system if history_result is not None else system + + polyfill_result = await _run_polyfill_if_enabled( + model=model, + messages=working_messages, + tools=tools, + system=working_system, + context_management_spec=context_management_spec, + litellm_metadata=litellm_metadata, + drop_params=drop_params, + llm_router=llm_router, + user_api_key_auth=user_api_key_auth, + ) + + if polyfill_result is not None: + return polyfill_result + + # Safety net: if we skipped client-history pre-processing because a + # ``compact_20260112`` polyfill was expected to handle the compaction + # block itself but the polyfill ultimately did not produce a result + # (e.g. it crashed and was best-effort swallowed in + # ``_run_polyfill_if_enabled``), apply the slice-only fallback now so + # Anthropic-specific ``compaction`` content blocks don't leak through + # to non-Anthropic backends that would reject them. + if polyfill_will_run and history_result is None: + history_result = apply_client_compaction_block_history( + messages=cast(List[Dict[str, Any]], messages), + system=system, + ) + return history_result + + +def _polyfill_will_run( + *, + context_management_spec: Any, + drop_params: Optional[bool], +) -> bool: + """Return True when ``compact_20260112`` will run via the polyfill dispatcher. + + Mirrors the gating in ``_run_polyfill_if_enabled``: an empty spec or + effective ``drop_params`` short-circuits the polyfill. The pre-processing + skip only applies when the dispatcher will actually invoke + ``apply_compact_20260112`` (which has its own compaction-block slicing). + """ + edits = _normalize_spec_edits( + context_management_spec=context_management_spec, + drop_params=drop_params, + ) + if edits is None: + return False + + from litellm.llms.anthropic.experimental_pass_through.context_management.constants import ( + COMPACT_EDIT_TYPE, + ) + + return any( + isinstance(edit, dict) and edit.get("type") == COMPACT_EDIT_TYPE + for edit in edits + ) + + +def _spec_has_non_compact_edits( + *, + context_management_spec: Any, + drop_params: Optional[bool], +) -> bool: + """Return True when the spec includes edits other than ``compact_20260112``. + + Used to decide whether a polyfill failure can be silently swallowed + (compact-only specs have a safe compaction-block slicing fallback) or + must be surfaced (other editors like ``clear_tool_uses_20250919`` have + no slice-only fallback and would otherwise be dropped without notice). + """ + edits = _normalize_spec_edits( + context_management_spec=context_management_spec, + drop_params=drop_params, + ) + if edits is None: + return False + + from litellm.llms.anthropic.experimental_pass_through.context_management.constants import ( + COMPACT_EDIT_TYPE, + ) + + return any( + isinstance(edit, dict) + and isinstance(edit.get("type"), str) + and edit.get("type") != COMPACT_EDIT_TYPE + for edit in edits + ) + + +def _normalize_spec_edits( + *, + context_management_spec: Any, + drop_params: Optional[bool], +) -> Optional[List[Dict[str, Any]]]: + """Return the normalized ``edits`` list, or ``None`` if the polyfill won't run. + + Delegates spec-shape normalization to the dispatcher's ``_normalize_spec`` + so the prediction here can't drift from what the dispatcher actually does. + """ + if not context_management_spec: + return None + + effective_drop_params = ( + drop_params if drop_params is not None else litellm.drop_params + ) + if effective_drop_params: + return None + + from litellm.llms.anthropic.experimental_pass_through.context_management.dispatcher import ( + _normalize_spec, + ) + + try: + return _normalize_spec(context_management_spec) + except Exception: + return None + + +async def _run_polyfill_if_enabled( + *, + model: str, + messages: List[Dict], + tools: Optional[List[Dict]], + system: Optional[Any], + context_management_spec: Any, + litellm_metadata: Optional[Dict], + drop_params: Optional[bool], + llm_router: Any, + user_api_key_auth: Any = None, +) -> Optional[PolyfillResult]: + """Run the async context_management polyfill if a spec is present. + + Returns ``None`` when the spec is empty or drop_params is on. Raises + ``AnthropicContextManagementError`` so the /v1/messages endpoint can + emit an Anthropic-format 400. All other exceptions are best-effort + swallowed (matches v0 behavior). + """ + if not context_management_spec: + return None + + effective_drop_params = ( + drop_params if drop_params is not None else litellm.drop_params + ) + if effective_drop_params: + return None + + try: + return await apply_context_management( + model=model, + messages=messages, + tools=tools, + system=system, + context_management_spec=context_management_spec, + litellm_metadata=litellm_metadata, + llm_router=llm_router, + user_api_key_auth=user_api_key_auth, + ) + except AnthropicContextManagementError: + # Surface validation errors so the endpoint can emit an Anthropic-format + # 400. Other exception types fall into the best-effort branch below. + raise + except Exception as e: + verbose_logger.exception( + "context_management polyfill: skipping edits due to error: %s", e + ) + # Best-effort swallow is only safe for compact-only specs, where the + # caller's compaction-block-slicing safety net produces a correct + # (if degraded) result. When the spec also requested non-compact + # edits (e.g. ``clear_tool_uses_20250919``), the safety net does + # NOT re-run those editors, so silently returning ``None`` would + # drop them with no error surface. Raise instead so the endpoint + # emits an Anthropic-format error. + if _spec_has_non_compact_edits( + context_management_spec=context_management_spec, + drop_params=drop_params, + ): + raise AnthropicContextManagementError( + status_code=500, + message=f"context_management polyfill failed: {e}", + ) from e + return None + + ######################################################## # init adapter ANTHROPIC_ADAPTER = AnthropicAdapter() @@ -163,7 +422,7 @@ class LiteLLMMessagesToCompletionTransformationHandler: metadata: Optional[Dict] = None, stop_sequences: Optional[List[str]] = None, stream: Optional[bool] = False, - system: Optional[str] = None, + system: Optional[Union[str, List[Dict[str, Any]]]] = None, temperature: Optional[float] = None, thinking: Optional[Dict] = None, tool_choice: Optional[Dict] = None, @@ -307,19 +566,56 @@ class LiteLLMMessagesToCompletionTransformationHandler: top_p: Optional[float] = None, output_format: Optional[Dict] = None, **kwargs, - ) -> Union[AnthropicMessagesResponse, AsyncIterator]: + ) -> Union[AnthropicMessagesResponse, AsyncIterator[Any], Iterator[bytes]]: """Handle non-Anthropic models asynchronously using the adapter""" + context_management = kwargs.pop("context_management", None) + drop_params: Optional[bool] = kwargs.get("drop_params", None) + litellm_router = kwargs.pop("litellm_router", None) + if litellm_router is None: + try: + from litellm.proxy.proxy_server import llm_router as _proxy_router + + litellm_router = _proxy_router + except Exception: + pass + + proxy_litellm_metadata = _extract_proxy_litellm_metadata(kwargs) + user_api_key_auth = ( + proxy_litellm_metadata.get("user_api_key_auth") + if proxy_litellm_metadata is not None + else None + ) + + polyfill_result = await _prepare_context_managed_request( + model=model, + messages=messages, + tools=tools, + system=system, + context_management_spec=context_management, + litellm_metadata=proxy_litellm_metadata, + drop_params=drop_params, + llm_router=litellm_router, + user_api_key_auth=user_api_key_auth, + ) + + effective_messages = ( + polyfill_result.messages if polyfill_result is not None else messages + ) + effective_system = ( + polyfill_result.system if polyfill_result is not None else system + ) + ( completion_kwargs, tool_name_mapping, ) = LiteLLMMessagesToCompletionTransformationHandler._prepare_completion_kwargs( max_tokens=max_tokens, - messages=messages, + messages=effective_messages, model=model, metadata=metadata, stop_sequences=stop_sequences, stream=stream, - system=system, + system=effective_system, temperature=temperature, thinking=thinking, tool_choice=tool_choice, @@ -338,6 +634,8 @@ class LiteLLMMessagesToCompletionTransformationHandler: completion_response, model=model, tool_name_mapping=tool_name_mapping, + polyfill_result=polyfill_result, + is_async=True, ) ) if transformed_stream is not None: @@ -347,6 +645,7 @@ class LiteLLMMessagesToCompletionTransformationHandler: anthropic_response = ANTHROPIC_ADAPTER.translate_completion_output_params( cast(ModelResponse, completion_response), tool_name_mapping=tool_name_mapping, + polyfill_result=polyfill_result, ) if anthropic_response is not None: return anthropic_response @@ -372,8 +671,13 @@ class LiteLLMMessagesToCompletionTransformationHandler: **kwargs, ) -> Union[ AnthropicMessagesResponse, + Iterator[bytes], AsyncIterator[Any], - Coroutine[Any, Any, Union[AnthropicMessagesResponse, AsyncIterator[Any]]], + Coroutine[ + Any, + Any, + Union[AnthropicMessagesResponse, AsyncIterator[Any], Iterator[bytes]], + ], ]: """Handle non-Anthropic models using the adapter.""" if _is_async is True: @@ -395,17 +699,72 @@ class LiteLLMMessagesToCompletionTransformationHandler: **kwargs, ) + # Run the context_management polyfill on the sync path too so that + # ``litellm.messages.create()`` callers don't silently lose edits like + # ``clear_tool_uses_20250919``. The dispatcher is async (so the + # ``compact_20260112`` editor can ``await`` the summarization model); + # bridge to it via ``run_async_function``. + context_management = kwargs.pop("context_management", None) + drop_params: Optional[bool] = kwargs.get("drop_params", None) + # Deliberately do NOT auto-attach the proxy ``llm_router`` here: + # ``run_async_function`` spawns a new event loop in a worker thread + # to bridge to the async dispatcher, but the proxy router's httpx + # ``AsyncClient`` instances are bound to the proxy's main event loop. + # Reusing them from the new thread's loop violates httpx's single-loop + # invariant and can raise ``RuntimeError: Event loop is closed`` or + # produce stalled connections. The summary editor falls back to + # ``litellm.acompletion`` (which creates a fresh client per call) when + # ``llm_router`` is ``None``, which is safe to call from the bridged + # loop. The async ``async_anthropic_messages_handler`` path is + # unaffected because it ``await``s within the original event loop. + litellm_router = kwargs.pop("litellm_router", None) + + # Skip the async bridge entirely when there is nothing for either the + # polyfill or the client-history slice-only fallback to do. The vast + # majority of sync ``litellm.messages.create()`` requests carry no + # ``context_management`` spec and no client-sent ``compaction`` block, + # and bridging through a worker-thread event loop just to discover + # there is no work is pure overhead. + if context_management is None and not _messages_have_compaction_block(messages): + polyfill_result: Optional[PolyfillResult] = None + else: + proxy_litellm_metadata = _extract_proxy_litellm_metadata(kwargs) + user_api_key_auth = ( + proxy_litellm_metadata.get("user_api_key_auth") + if proxy_litellm_metadata is not None + else None + ) + polyfill_result = run_async_function( + _prepare_context_managed_request, + model=model, + messages=messages, + tools=tools, + system=system, + context_management_spec=context_management, + litellm_metadata=proxy_litellm_metadata, + drop_params=drop_params, + llm_router=litellm_router, + user_api_key_auth=user_api_key_auth, + ) + + effective_messages = ( + polyfill_result.messages if polyfill_result is not None else messages + ) + effective_system = ( + polyfill_result.system if polyfill_result is not None else system + ) + ( completion_kwargs, tool_name_mapping, ) = LiteLLMMessagesToCompletionTransformationHandler._prepare_completion_kwargs( max_tokens=max_tokens, - messages=messages, + messages=effective_messages, model=model, metadata=metadata, stop_sequences=stop_sequences, stream=stream, - system=system, + system=effective_system, temperature=temperature, thinking=thinking, tool_choice=tool_choice, @@ -424,6 +783,8 @@ class LiteLLMMessagesToCompletionTransformationHandler: completion_response, model=model, tool_name_mapping=tool_name_mapping, + polyfill_result=polyfill_result, + is_async=False, ) ) if transformed_stream is not None: @@ -433,6 +794,7 @@ class LiteLLMMessagesToCompletionTransformationHandler: anthropic_response = ANTHROPIC_ADAPTER.translate_completion_output_params( cast(ModelResponse, completion_response), tool_name_mapping=tool_name_mapping, + polyfill_result=polyfill_result, ) if anthropic_response is not None: return anthropic_response diff --git a/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py b/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py index c65dfb22730..bacb9f8ddf6 100644 --- a/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py +++ b/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py @@ -3,11 +3,26 @@ import json import traceback from collections import deque -from typing import TYPE_CHECKING, Any, AsyncIterator, Dict, Iterator, Literal, Optional +from typing import ( + TYPE_CHECKING, + Any, + AsyncIterator, + Dict, + Iterator, + List, + Literal, + Optional, +) -from litellm import verbose_logger +from litellm._logging import verbose_logger from litellm._uuid import uuid -from litellm.types.llms.anthropic import UsageDelta +from litellm.types.llms.anthropic import ( + AppliedEdit, + CompactionBlock, + ContextManagementResponse, + UsageDelta, + UsageIteration, +) from litellm.types.utils import AdapterCompletionStreamWrapper if TYPE_CHECKING: @@ -37,22 +52,208 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): holding_stop_reason_chunk: Optional[Any] = None queued_usage_chunk: bool = False current_content_block_index: int = 0 - current_content_block_start: ContentBlockContentBlockDict = TextBlock( - type="text", - text="", - ) - chunk_queue: deque = deque() # Queue for buffering multiple chunks def __init__( self, completion_stream: Any, model: str, tool_name_mapping: Optional[Dict[str, str]] = None, + applied_edits: Optional[List[AppliedEdit]] = None, + compaction_block: Optional[CompactionBlock] = None, + iterations_usage: Optional[List[UsageIteration]] = None, ): super().__init__(completion_stream) self.model = model # Mapping of truncated tool names to original names (for OpenAI's 64-char limit) self.tool_name_mapping = tool_name_mapping or {} + # Polyfill applied_edits on final message_delta. + self.applied_edits: List[AppliedEdit] = list(applied_edits or []) + # Synthesized compaction block from compact_20260112 polyfill (streaming). + self.compaction_block = compaction_block + self.iterations_usage = iterations_usage + self.sent_compaction_block: bool = False + # Per-phase flags so the compaction block's start/delta/stop events + # are emitted (and the public state machine is advanced) in + # lock-step with the caller actually consuming each event. Pre- + # queuing all three would set ``sent_content_block_finish=True`` + # before the client received ``content_block_stop``, leaving the + # observable state inconsistent during the drain window. + self.sent_compaction_block_start: bool = False + self.sent_compaction_block_delta: bool = False + # Per-instance queue for buffering multiple chunks. Must be initialized + # here (not at class level) so concurrent streams don't share the same + # deque and corrupt each other's SSE event order. + self.chunk_queue: deque = deque() + # Per-instance default content block. Must be initialized here (not at + # class level) so concurrent streams don't share the same mutable dict + # — `_should_start_new_content_block` mutates `tool_block["name"]` in + # place, which would otherwise leak across streams. + self.current_content_block_start: ( + "AnthropicStreamWrapper.ContentBlockContentBlockDict" + ) = self.TextBlock( + type="text", + text="", + ) + + def _merge_usage_into_held_stop_reason_chunk(self, chunk: Any) -> Dict[str, Any]: + """Merge usage data from ``chunk`` into the held ``message_delta`` chunk. + + Shared by both the sync ``__next__`` and async ``__anext__`` paths so + the subtle hold-and-merge logic (cache tokens, ``context_management`` + attachment, ``UsageDelta`` shape) lives in exactly one place. + + Caller is responsible for managing ``self.holding_stop_reason_chunk`` + and ``self.queued_usage_chunk`` state and for queuing the returned + merged chunk. + """ + assert self.holding_stop_reason_chunk is not None + merged_chunk = self.holding_stop_reason_chunk.copy() + if "delta" not in merged_chunk: + merged_chunk["delta"] = {} + + uncached_input_tokens = chunk.usage.prompt_tokens or 0 + if ( + hasattr(chunk.usage, "prompt_tokens_details") + and chunk.usage.prompt_tokens_details + ): + cached_tokens = ( + getattr(chunk.usage.prompt_tokens_details, "cached_tokens", 0) or 0 + ) + uncached_input_tokens -= cached_tokens + + usage_dict: UsageDelta = { + "input_tokens": uncached_input_tokens, + "output_tokens": chunk.usage.completion_tokens or 0, + } + if ( + hasattr(chunk.usage, "_cache_creation_input_tokens") + and chunk.usage._cache_creation_input_tokens > 0 + ): + usage_dict["cache_creation_input_tokens"] = ( + chunk.usage._cache_creation_input_tokens + ) + if ( + hasattr(chunk.usage, "_cache_read_input_tokens") + and chunk.usage._cache_read_input_tokens > 0 + ): + usage_dict["cache_read_input_tokens"] = chunk.usage._cache_read_input_tokens + merged_chunk["usage"] = usage_dict + if self.applied_edits and "context_management" not in merged_chunk: + merged_chunk["context_management"] = ContextManagementResponse( + applied_edits=list(self.applied_edits) + ) + return self._augment_message_delta_usage(merged_chunk) + + def _ensure_context_management_attached( + self, message_delta_chunk: Dict[str, Any] + ) -> Dict[str, Any]: + """Attach ``context_management`` to a ``message_delta`` chunk if + ``self.applied_edits`` is non-empty and the chunk does not already + carry it. Returns the (possibly new) chunk dict. + + Centralizing this guard ensures every ``message_delta`` emission + path (merge-with-usage and direct-flush-of-held) consistently + surfaces ``applied_edits`` to the client. + """ + if not self.applied_edits or "context_management" in message_delta_chunk: + return message_delta_chunk + augmented = message_delta_chunk.copy() + augmented["context_management"] = ContextManagementResponse( + applied_edits=list(self.applied_edits) + ) + return augmented + + def _augment_message_delta_usage( + self, message_delta_chunk: Dict[str, Any] + ) -> Dict[str, Any]: + """Attach polyfill compaction iteration usage to the final message_delta. + + Also defensively re-attaches ``context_management`` so the direct + held-chunk flush path stays in sync with the merge path's guarantee + when ``self.applied_edits`` is non-empty. + """ + message_delta_chunk = self._ensure_context_management_attached( + message_delta_chunk + ) + if self.iterations_usage is None: + return message_delta_chunk + usage = message_delta_chunk.get("usage") + if not isinstance(usage, dict) or "iterations" in usage: + return message_delta_chunk + + input_tokens = usage.get("input_tokens", 0) or 0 + output_tokens = usage.get("output_tokens", 0) or 0 + augmented = message_delta_chunk.copy() + augmented_usage = dict(usage) + iterations: List[UsageIteration] = list(self.iterations_usage) + # Only emit a ``message`` iteration when we have real token data. + # Without a separate usage chunk (e.g. provider sent finish_reason + # alone), the held ``message_delta`` carries placeholder zeros from + # the translate step; reporting a zero-token iteration would be + # misleading and inconsistent with the non-streaming path. + if input_tokens > 0 or output_tokens > 0: + message_iteration: UsageIteration = { + "type": "message", + "input_tokens": input_tokens, + "output_tokens": output_tokens, + } + iterations.append(message_iteration) + augmented_usage["iterations"] = iterations # type: ignore[typeddict-unknown-key] + augmented["usage"] = augmented_usage + return augmented + + def _next_compaction_event(self) -> Optional[Dict[str, Any]]: + """Return the next compaction content-block SSE event, or ``None``. + + Anthropic delivers compaction as a single delta (no token-by-token + streaming), but we still surface it as a proper + start → delta → stop trio. Each call returns exactly one event so + the state machine (``sent_content_block_finish``, + ``current_content_block_index``) is advanced *only* when the + terminal stop event is actually handed back to the caller. This + prevents an observable window where the flags claim the block is + finished while the stop event is still buffered. + """ + if self.compaction_block is None or self.sent_compaction_block: + return None + + compaction_index = self.current_content_block_index + + if not self.sent_compaction_block_start: + self.sent_compaction_block_start = True + return { + "type": "content_block_start", + "index": compaction_index, + # Mirror the text-block shape ({"type": "text", "text": ""}): + # send an empty ``content`` field so clients that introspect + # ``content_block_start`` see the full block schema. The + # actual summary text arrives via the ``content_block_delta`` + # below. + "content_block": {"type": "compaction", "content": ""}, + } + + if not self.sent_compaction_block_delta: + self.sent_compaction_block_delta = True + summary_content = self.compaction_block.get("content") or "" + return { + "type": "content_block_delta", + "index": compaction_index, + "delta": {"type": "compaction_delta", "content": summary_content}, + } + + stop_event = { + "type": "content_block_stop", + "index": compaction_index, + } + # Don't touch ``sent_content_block_finish`` here: that flag is the + # state machine for the regular text/tool_use/thinking block and is + # independent of the synthetic compaction block lifecycle. Conflating + # them would let outside observers (subclass overrides, introspection + # hooks, exception paths) see ``sent_content_block_finish=True`` + # without any regular content block ever having started. + self._increment_content_block_index() + self.sent_compaction_block = True + return stop_event def _create_initial_usage_delta(self) -> UsageDelta: """ @@ -75,7 +276,7 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): cache_read_input_tokens=0, ) - def __next__(self): + def __next__(self): # noqa: PLR0915 from .transformation import LiteLLMAnthropicMessagesAdapter try: @@ -103,8 +304,17 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): ) return self.chunk_queue.popleft() + if ( + self.sent_compaction_block is False + and self.compaction_block is not None + ): + compaction_event = self._next_compaction_event() + if compaction_event is not None: + return compaction_event + if self.sent_content_block_start is False: self.sent_content_block_start = True + self.sent_content_block_finish = False self.chunk_queue.append( { "type": "content_block_start", @@ -122,11 +332,45 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): if should_start_new_block: self._increment_content_block_index() + # applied_edits only needs to flow to the final message_delta + # (when finish_reason is set); skip threading it through every + # intermediate chunk. For the hold-and-merge path below, + # context_management is attached directly to the merged chunk, + # so the translated ``processed_chunk`` would be discarded — + # skip the applied_edits attachment in that case to avoid + # allocating a throwaway ``MessageBlockDelta``. + will_merge_into_held = ( + self.holding_stop_reason_chunk is not None + and getattr(chunk, "usage", None) is not None + ) + is_final_chunk = chunk.choices[0].finish_reason is not None processed_chunk = LiteLLMAnthropicMessagesAdapter().translate_streaming_openai_response_to_anthropic( response=chunk, current_content_block_index=self.current_content_block_index, + applied_edits=( + self.applied_edits + if is_final_chunk and not will_merge_into_held + else None + ), ) + # Check if this is a usage chunk and we have a held stop_reason chunk + if will_merge_into_held: + merged_chunk = self._merge_usage_into_held_stop_reason_chunk(chunk) + self.chunk_queue.append(merged_chunk) + self.queued_usage_chunk = True + self.holding_stop_reason_chunk = None + return self.chunk_queue.popleft() + + if self.queued_usage_chunk: + # Usage has already been merged + emitted. Any trailing + # provider events would violate Anthropic SSE ordering + # (no chunks may follow the final ``message_delta``), so + # silently drop them — matches the async ``__anext__`` + # behavior where the block-handling logic is gated on + # ``not self.queued_usage_chunk``. + continue + if should_start_new_block and not self.sent_content_block_finish: # Queue the sequence: content_block_stop -> content_block_start # For text blocks the trigger chunk is not emitted as a separate @@ -178,20 +422,64 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): } ) self.sent_content_block_finish = True - self.chunk_queue.append(processed_chunk) + if processed_chunk.get("delta", {}).get("stop_reason") is not None: + self.holding_stop_reason_chunk = processed_chunk + else: + processed_chunk = self._augment_message_delta_usage( + processed_chunk + ) + self.chunk_queue.append(processed_chunk) return self.chunk_queue.popleft() elif self.holding_chunk is not None: self.chunk_queue.append(self.holding_chunk) + if processed_chunk.get("type") == "message_delta": + processed_chunk = self._augment_message_delta_usage( + processed_chunk + ) self.chunk_queue.append(processed_chunk) self.holding_chunk = None return self.chunk_queue.popleft() else: + if processed_chunk.get("type") == "message_delta": + processed_chunk = self._augment_message_delta_usage( + processed_chunk + ) self.chunk_queue.append(processed_chunk) return self.chunk_queue.popleft() - # Handle any remaining held chunks after stream ends - if self.holding_chunk is not None: - self.chunk_queue.append(self.holding_chunk) + # Handle any remaining held chunks after stream ends. The + # buffered ``holding_chunk`` (a ``content_block_delta``) must + # precede the final ``message_delta`` so Anthropic SSE event + # ordering is preserved. When ``queued_usage_chunk`` is True, + # the final ``message_delta`` has already been emitted; any + # buffered content delta is dropped rather than emitted after + # ``message_delta`` (which would violate SSE ordering and may + # confuse strict Anthropic SDK clients). + if not self.queued_usage_chunk: + if self.holding_chunk is not None: + self.chunk_queue.append(self.holding_chunk) + self.holding_chunk = None + if self.holding_stop_reason_chunk is not None: + # A final ``message_delta`` must be preceded by + # ``content_block_stop`` so the emitted SSE stays in + # valid Anthropic order (... -> content_block_stop -> + # message_delta). Emit ``content_block_stop`` here if + # the active content block was not already closed. + if not self.sent_content_block_finish: + self.chunk_queue.append( + { + "type": "content_block_stop", + "index": self.current_content_block_index, + } + ) + self.sent_content_block_finish = True + self.chunk_queue.append( + self._augment_message_delta_usage( + self.holding_stop_reason_chunk + ) + ) + self.holding_stop_reason_chunk = None + else: self.holding_chunk = None if not self.sent_last_message: @@ -205,6 +493,26 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): except StopIteration: if self.chunk_queue: return self.chunk_queue.popleft() + # Handle any held stop_reason chunk. Emit ``content_block_stop`` + # first if the active content block was not already closed, so + # Anthropic SSE ordering is preserved (content_block_stop -> + # message_delta). + if self.holding_stop_reason_chunk is not None: + if not self.sent_content_block_finish: + self.sent_content_block_finish = True + self.chunk_queue.append( + self._augment_message_delta_usage( + self.holding_stop_reason_chunk + ) + ) + self.holding_stop_reason_chunk = None + return { + "type": "content_block_stop", + "index": self.current_content_block_index, + } + held = self._augment_message_delta_usage(self.holding_stop_reason_chunk) + self.holding_stop_reason_chunk = None + return held if self.sent_last_message is False: self.sent_last_message = True return {"type": "message_stop"} @@ -213,7 +521,7 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): verbose_logger.error( "Anthropic Adapter - {}\n{}".format(e, traceback.format_exc()) ) - raise StopAsyncIteration + raise StopIteration async def __anext__(self): # noqa: PLR0915 from .transformation import LiteLLMAnthropicMessagesAdapter @@ -243,8 +551,17 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): ) return self.chunk_queue.popleft() + if ( + self.sent_compaction_block is False + and self.compaction_block is not None + ): + compaction_event = self._next_compaction_event() + if compaction_event is not None: + return compaction_event + if self.sent_content_block_start is False: self.sent_content_block_start = True + self.sent_content_block_finish = False self.chunk_queue.append( { "type": "content_block_start", @@ -263,57 +580,31 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): if should_start_new_block: self._increment_content_block_index() + # applied_edits only needs to flow to the final message_delta + # (when finish_reason is set); skip threading it through every + # intermediate chunk. For the hold-and-merge path below, + # context_management is attached directly to the merged chunk, + # so the translated ``processed_chunk`` would be discarded — + # skip the applied_edits attachment in that case to avoid + # allocating a throwaway ``MessageBlockDelta``. + will_merge_into_held = ( + self.holding_stop_reason_chunk is not None + and getattr(chunk, "usage", None) is not None + ) + is_final_chunk = chunk.choices[0].finish_reason is not None processed_chunk = LiteLLMAnthropicMessagesAdapter().translate_streaming_openai_response_to_anthropic( response=chunk, current_content_block_index=self.current_content_block_index, + applied_edits=( + self.applied_edits + if is_final_chunk and not will_merge_into_held + else None + ), ) # Check if this is a usage chunk and we have a held stop_reason chunk - if ( - self.holding_stop_reason_chunk is not None - and getattr(chunk, "usage", None) is not None - ): - # Merge usage into the held stop_reason chunk - merged_chunk = self.holding_stop_reason_chunk.copy() - if "delta" not in merged_chunk: - merged_chunk["delta"] = {} - - # Add usage to the held chunk - uncached_input_tokens = chunk.usage.prompt_tokens or 0 - if ( - hasattr(chunk.usage, "prompt_tokens_details") - and chunk.usage.prompt_tokens_details - ): - cached_tokens = ( - getattr( - chunk.usage.prompt_tokens_details, "cached_tokens", 0 - ) - or 0 - ) - uncached_input_tokens -= cached_tokens - - usage_dict: UsageDelta = { - "input_tokens": uncached_input_tokens, - "output_tokens": chunk.usage.completion_tokens or 0, - } - # Add cache tokens if available (for prompt caching support) - if ( - hasattr(chunk.usage, "_cache_creation_input_tokens") - and chunk.usage._cache_creation_input_tokens > 0 - ): - usage_dict["cache_creation_input_tokens"] = ( - chunk.usage._cache_creation_input_tokens - ) - if ( - hasattr(chunk.usage, "_cache_read_input_tokens") - and chunk.usage._cache_read_input_tokens > 0 - ): - usage_dict["cache_read_input_tokens"] = ( - chunk.usage._cache_read_input_tokens - ) - merged_chunk["usage"] = usage_dict - - # Queue the merged chunk and reset + if will_merge_into_held: + merged_chunk = self._merge_usage_into_held_stop_reason_chunk(chunk) self.chunk_queue.append(merged_chunk) self.queued_usage_chunk = True self.holding_stop_reason_chunk = None @@ -379,28 +670,63 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): ): self.holding_stop_reason_chunk = processed_chunk else: + processed_chunk = self._augment_message_delta_usage( + processed_chunk + ) self.chunk_queue.append(processed_chunk) return self.chunk_queue.popleft() elif self.holding_chunk is not None: # Queue both chunks self.chunk_queue.append(self.holding_chunk) + if processed_chunk.get("type") == "message_delta": + processed_chunk = self._augment_message_delta_usage( + processed_chunk + ) self.chunk_queue.append(processed_chunk) self.holding_chunk = None return self.chunk_queue.popleft() else: - # Queue the current chunk + if processed_chunk.get("type") == "message_delta": + processed_chunk = self._augment_message_delta_usage( + processed_chunk + ) self.chunk_queue.append(processed_chunk) return self.chunk_queue.popleft() - # Handle any remaining held chunks after stream ends + # Handle any remaining held chunks after stream ends. The + # buffered ``holding_chunk`` (a ``content_block_delta``) must + # precede the final ``message_delta`` so Anthropic SSE event + # ordering is preserved. When ``queued_usage_chunk`` is True, + # the final ``message_delta`` has already been emitted; any + # buffered content delta is dropped rather than emitted after + # ``message_delta`` (which would violate SSE ordering and may + # confuse strict Anthropic SDK clients). if not self.queued_usage_chunk: - if self.holding_stop_reason_chunk is not None: - self.chunk_queue.append(self.holding_stop_reason_chunk) - self.holding_stop_reason_chunk = None - if self.holding_chunk is not None: self.chunk_queue.append(self.holding_chunk) self.holding_chunk = None + if self.holding_stop_reason_chunk is not None: + # A final ``message_delta`` must be preceded by + # ``content_block_stop`` so the emitted SSE stays in + # valid Anthropic order (... -> content_block_stop -> + # message_delta). Emit ``content_block_stop`` here if + # the active content block was not already closed. + if not self.sent_content_block_finish: + self.chunk_queue.append( + { + "type": "content_block_stop", + "index": self.current_content_block_index, + } + ) + self.sent_content_block_finish = True + self.chunk_queue.append( + self._augment_message_delta_usage( + self.holding_stop_reason_chunk + ) + ) + self.holding_stop_reason_chunk = None + else: + self.holding_chunk = None if not self.sent_last_message: self.sent_last_message = True @@ -416,9 +742,28 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): # Handle any remaining queued chunks before stopping if self.chunk_queue: return self.chunk_queue.popleft() - # Handle any held stop_reason chunk + # Handle any held stop_reason chunk — clear after capturing so a + # subsequent ``__anext__`` call doesn't re-emit the same chunk + # (matches the sync ``__next__`` path). Emit ``content_block_stop`` + # first if the active content block was not already closed, so + # Anthropic SSE ordering is preserved (content_block_stop -> + # message_delta). if self.holding_stop_reason_chunk is not None: - return self.holding_stop_reason_chunk + if not self.sent_content_block_finish: + self.sent_content_block_finish = True + self.chunk_queue.append( + self._augment_message_delta_usage( + self.holding_stop_reason_chunk + ) + ) + self.holding_stop_reason_chunk = None + return { + "type": "content_block_stop", + "index": self.current_content_block_index, + } + held = self._augment_message_delta_usage(self.holding_stop_reason_chunk) + self.holding_stop_reason_chunk = None + return held if not self.sent_last_message: self.sent_last_message = True return {"type": "message_stop"} diff --git a/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py b/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py index fe8e694efe5..02e0c562654 100644 --- a/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py +++ b/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py @@ -6,6 +6,7 @@ from typing import ( Any, AsyncIterator, Dict, + Iterator, List, Literal, Optional, @@ -75,6 +76,9 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import ( from litellm.litellm_core_utils.prompt_templates.factory import ( THOUGHT_SIGNATURE_SEPARATOR, ) +from litellm.llms.anthropic.experimental_pass_through.context_management import ( + PolyfillResult, +) from litellm.types.llms.anthropic import ( ANTHROPIC_HOSTED_TOOLS, AllAnthropicToolsValues, @@ -87,14 +91,17 @@ from litellm.types.llms.anthropic import ( AnthropicResponseContentBlockText, AnthropicResponseContentBlockThinking, AnthropicResponseContentBlockToolUse, + AppliedEdit, ContentBlockDelta, ContentJsonBlockDelta, ContentTextBlockDelta, ContentThinkingBlockDelta, ContentThinkingSignatureBlockDelta, + ContextManagementResponse, MessageBlockDelta, MessageDelta, UsageDelta, + UsageIteration, ) from litellm.types.llms.anthropic_messages.anthropic_response import ( AnthropicMessagesResponse, @@ -195,6 +202,7 @@ class AnthropicAdapter: self, response: ModelResponse, tool_name_mapping: Optional[Dict[str, str]] = None, + polyfill_result: Optional[PolyfillResult] = None, ) -> Optional[AnthropicMessagesResponse]: """ Translate OpenAI response to Anthropic format. @@ -204,10 +212,12 @@ class AnthropicAdapter: tool_name_mapping: Optional mapping of truncated tool names to original names. Used to restore original names for tools that exceeded OpenAI's 64-char limit. + polyfill_result: PolyfillResult from context_management polyfill. """ return LiteLLMAnthropicMessagesAdapter().translate_openai_response_to_anthropic( response=response, tool_name_mapping=tool_name_mapping, + polyfill_result=polyfill_result, ) def translate_completion_output_params_streaming( @@ -215,7 +225,9 @@ class AnthropicAdapter: completion_stream: Any, model: str, tool_name_mapping: Optional[Dict[str, str]] = None, - ) -> Union[AsyncIterator[bytes], None]: + polyfill_result: Optional[PolyfillResult] = None, + is_async: bool = True, + ) -> Union[AsyncIterator[bytes], Iterator[bytes], None]: """ Translate OpenAI streaming response to Anthropic format. @@ -223,14 +235,35 @@ class AnthropicAdapter: completion_stream: The OpenAI streaming response model: The model name tool_name_mapping: Optional mapping of truncated tool names to original names. + polyfill_result: PolyfillResult from context_management polyfill. + is_async: When ``True`` (default, for back-compat with existing + async callers) returns an ``AsyncIterator[bytes]``. When + ``False`` returns a sync ``Iterator[bytes]`` so sync callers + (e.g. ``litellm.anthropic.messages.create(stream=True)`` via + the sync handler) don't get back an async iterator they + can't iterate without an event loop. """ + applied_edits = ( + polyfill_result.applied_edits_for_response() if polyfill_result else None + ) + compaction_block = ( + polyfill_result.compaction_block if polyfill_result is not None else None + ) + iterations_usage = ( + polyfill_result.iterations_usage if polyfill_result is not None else None + ) anthropic_wrapper = AnthropicStreamWrapper( completion_stream=completion_stream, model=model, tool_name_mapping=tool_name_mapping, + applied_edits=applied_edits, + compaction_block=compaction_block, + iterations_usage=iterations_usage, ) - # Return the SSE-wrapped version for proper event formatting - return anthropic_wrapper.async_anthropic_sse_wrapper() + # Return the SSE-wrapped version for proper event formatting. + if is_async: + return anthropic_wrapper.async_anthropic_sse_wrapper() + return anthropic_wrapper.anthropic_sse_wrapper() class LiteLLMAnthropicMessagesAdapter: @@ -1299,9 +1332,18 @@ class LiteLLMAnthropicMessagesAdapter: else truncated_name ) + # Strip Gemini thought-signature suffix from id (mirrors streaming + # path below); base64 chars (+ / =) violate Anthropic's + # `^[a-zA-Z0-9_-]+$` tool_use.id pattern when replayed. + raw_id = tool_call.id or "" + base_id = ( + raw_id.split(THOUGHT_SIGNATURE_SEPARATOR, 1)[0] + if THOUGHT_SIGNATURE_SEPARATOR in raw_id + else raw_id + ) tool_use_block = AnthropicResponseContentBlockToolUse( type="tool_use", - id=tool_call.id, + id=base_id, name=original_name, input=parse_tool_call_arguments( tool_call.function.arguments, @@ -1333,6 +1375,7 @@ class LiteLLMAnthropicMessagesAdapter: self, response: ModelResponse, tool_name_mapping: Optional[Dict[str, str]] = None, + polyfill_result: Optional[PolyfillResult] = None, ) -> AnthropicMessagesResponse: """ Translate OpenAI response to Anthropic format. @@ -1342,12 +1385,17 @@ class LiteLLMAnthropicMessagesAdapter: tool_name_mapping: Optional mapping of truncated tool names to original names. Used to restore original names for tools that exceeded OpenAI's 64-char limit. + polyfill_result: PolyfillResult from context_management polyfill. """ ## translate content block anthropic_content = self._translate_openai_content_to_anthropic( choices=response.choices, # type: ignore tool_name_mapping=tool_name_mapping, ) + + if polyfill_result is not None and polyfill_result.compaction_block is not None: + anthropic_content.insert(0, polyfill_result.compaction_block) # type: ignore[arg-type] + ## extract finish reason anthropic_finish_reason = self._translate_openai_finish_reason_to_anthropic( openai_finish_reason=response.choices[0].finish_reason # type: ignore @@ -1376,6 +1424,14 @@ class LiteLLMAnthropicMessagesAdapter: if cached_tokens > 0: anthropic_usage["cache_read_input_tokens"] = cached_tokens + if polyfill_result is not None and polyfill_result.iterations_usage is not None: + message_iteration: UsageIteration = { + "type": "message", + "input_tokens": uncached_input_tokens, + "output_tokens": usage.completion_tokens or 0, + } + anthropic_usage["iterations"] = list(polyfill_result.iterations_usage) + [message_iteration] # type: ignore[typeddict-unknown-key] + translated_obj = AnthropicMessagesResponse( id=response.id, type="message", @@ -1387,6 +1443,14 @@ class LiteLLMAnthropicMessagesAdapter: stop_reason=anthropic_finish_reason, ) + applied_edits = ( + polyfill_result.applied_edits_for_response() if polyfill_result else None + ) + if applied_edits: + translated_obj["context_management"] = ContextManagementResponse( + applied_edits=list(applied_edits) + ) + return translated_obj def _translate_streaming_openai_chunk_to_anthropic_content_block( @@ -1467,7 +1531,7 @@ class LiteLLMAnthropicMessagesAdapter: for choice in choices: if choice.delta.content is not None and len(choice.delta.content) > 0: text += choice.delta.content - if choice.delta.tool_calls is not None: + if choice.delta.tool_calls: partial_json = "" for tool in choice.delta.tool_calls: if ( @@ -1519,7 +1583,10 @@ class LiteLLMAnthropicMessagesAdapter: return "text_delta", ContentTextBlockDelta(type="text_delta", text=text) def translate_streaming_openai_response_to_anthropic( - self, response: ModelResponse, current_content_block_index: int + self, + response: ModelResponse, + current_content_block_index: int, + applied_edits: Optional[List[AppliedEdit]] = None, ) -> Union[ContentBlockDelta, MessageBlockDelta]: ## base case - final chunk w/ finish reason if response.choices[0].finish_reason is not None: @@ -1569,9 +1636,14 @@ class LiteLLMAnthropicMessagesAdapter: usage_delta["cache_read_input_tokens"] = cached_tokens else: usage_delta = UsageDelta(input_tokens=0, output_tokens=0) - return MessageBlockDelta( + message_block = MessageBlockDelta( type="message_delta", delta=delta, usage=usage_delta # type: ignore ) + if applied_edits: + message_block["context_management"] = ContextManagementResponse( + applied_edits=list(applied_edits) + ) + return message_block ( type_of_content, content_block_delta, diff --git a/litellm/llms/anthropic/experimental_pass_through/context_management/__init__.py b/litellm/llms/anthropic/experimental_pass_through/context_management/__init__.py new file mode 100644 index 00000000000..729b2864524 --- /dev/null +++ b/litellm/llms/anthropic/experimental_pass_through/context_management/__init__.py @@ -0,0 +1,11 @@ +from .constants import CLEARED_TOOL_RESULT_PLACEHOLDER +from .dispatcher import apply_context_management +from .errors import AnthropicContextManagementError +from .result import PolyfillResult + +__all__ = [ + "apply_context_management", + "AnthropicContextManagementError", + "CLEARED_TOOL_RESULT_PLACEHOLDER", + "PolyfillResult", +] diff --git a/litellm/llms/anthropic/experimental_pass_through/context_management/constants.py b/litellm/llms/anthropic/experimental_pass_through/context_management/constants.py new file mode 100644 index 00000000000..ebbc182c427 --- /dev/null +++ b/litellm/llms/anthropic/experimental_pass_through/context_management/constants.py @@ -0,0 +1,45 @@ +"""Constants for the in-gateway context-management polyfill.""" + +CLEAR_TOOL_USES_EDIT_TYPE = "clear_tool_uses_20250919" + +DEFAULT_INPUT_TOKENS_TRIGGER = 100_000 +DEFAULT_KEEP_TOOL_USES = 3 + +CLEARED_TOOL_RESULT_PLACEHOLDER = "[Cleared by context management]" + +# compact_20260112 +COMPACT_EDIT_TYPE = "compact_20260112" +COMPACT_DEFAULT_TRIGGER_TOKENS = 150_000 +COMPACT_MIN_TRIGGER_TOKENS = 50_000 +# Default ``max_tokens`` for the summary call. Required by providers like +# Anthropic that reject requests without it; safely accepted by providers that +# don't strictly require it. Chosen to comfortably fit a long structured +# summary. Operators can override via +# ``general_settings.context_management_summary_max_tokens``. +COMPACT_SUMMARY_MAX_TOKENS = 4096 +COMPACT_SUMMARY_MAX_TOKENS_SETTING_KEY = "context_management_summary_max_tokens" +# Wall-clock bound for the summary sub-call. Without this a slow or +# unresponsive summary model would hang the parent ``/v1/messages`` request +# with no escape hatch; on timeout the editor falls into the standard +# ``summary_call_failed`` path and forwards the request without compaction. +COMPACT_SUMMARY_TIMEOUT_SECONDS = 60.0 +COMPACT_SUMMARY_MODEL_SETTING_KEY = "context_management_summary_model" +COMPACT_SUMMARY_SYSTEM_PREFIX = "Previous conversation summary: " + +# Default summarization prompt from the Anthropic spec. +COMPACT_DEFAULT_INSTRUCTIONS = ( + "You have written a partial transcript for the initial task above. Please " + "write a summary of the transcript. The purpose of this summary is to " + "provide continuity so you can continue to make progress towards solving " + "the task in a future context, where the raw history above may not be " + "accessible and will be replaced with this summary. Write down anything " + "that would be helpful, including the state, next steps, learnings etc. " + "You must wrap your summary in a

block." +) + +# Appended to the default prompt when ``tools`` are present and the caller +# did not supply custom ``instructions``. Matches the guidance in the +# Anthropic docs under "Compaction might fail when tools are defined". +COMPACT_NO_TOOL_CALLS_SUFFIX = ( + " Do not call any tools while writing this summary; respond with text only." +) diff --git a/litellm/llms/anthropic/experimental_pass_through/context_management/dispatcher.py b/litellm/llms/anthropic/experimental_pass_through/context_management/dispatcher.py new file mode 100644 index 00000000000..f7af09ee62a --- /dev/null +++ b/litellm/llms/anthropic/experimental_pass_through/context_management/dispatcher.py @@ -0,0 +1,127 @@ +"""Dispatch ``context_management`` edits to registered polyfill editors.""" + +import inspect +from typing import Any, Awaitable, Callable, Dict, List, Optional, Tuple, Union, cast + +from litellm._logging import verbose_logger +from litellm.types.llms.anthropic import AppliedEdit + +from .constants import CLEAR_TOOL_USES_EDIT_TYPE, COMPACT_EDIT_TYPE +from .editors import apply_clear_tool_uses_20250919, apply_compact_20260112 +from .result import PolyfillResult + +EditorFn = Callable[..., Any] + +_EDITOR_REGISTRY: Dict[str, EditorFn] = { + CLEAR_TOOL_USES_EDIT_TYPE: apply_clear_tool_uses_20250919, + COMPACT_EDIT_TYPE: apply_compact_20260112, +} + + +def _normalize_spec( + spec: Union[Dict[str, Any], List[Dict[str, Any]], None], +) -> Optional[List[Dict[str, Any]]]: + """Accept Anthropic-native dict form or OpenAI list form; return edits list.""" + if isinstance(spec, list): + # Local import to avoid an import cycle at module load. + from litellm.llms.anthropic.chat.transformation import AnthropicConfig + + spec = AnthropicConfig.map_openai_context_management_to_anthropic(spec) + + edits = spec.get("edits") if isinstance(spec, dict) else None + if not edits or not isinstance(edits, list): + return None + return [edit for edit in edits if isinstance(edit, dict)] + + +def _wrap_editor_return(raw: Any, *, fallback_system: Any) -> PolyfillResult: + """Coerce an editor's native return shape into a ``PolyfillResult``. + + v0 sync editors (e.g. ``clear_tool_uses_20250919``) return a 2-tuple + ``(messages, Optional[AppliedEdit])``. The new async ``compact_20260112`` + editor returns a ``PolyfillResult`` directly. + """ + if isinstance(raw, PolyfillResult): + return raw + # Legacy 2-tuple return — sync editors don't mutate ``system``, so + # carry the caller's value forward. + messages, applied = cast(Tuple[List[Dict[str, Any]], Any], raw) + return PolyfillResult( + messages=messages, + system=fallback_system, + applied_edits=[applied] if applied is not None else [], + ) + + +async def apply_context_management( + *, + model: str, + messages: List[Dict[str, Any]], + tools: Optional[List[Dict[str, Any]]], + system: Any, + context_management_spec: Union[Dict[str, Any], List[Dict[str, Any]], None], + litellm_metadata: Optional[Dict[str, Any]] = None, + llm_router: Any = None, + user_api_key_auth: Any = None, +) -> PolyfillResult: + """Run edits in order; return a single ``PolyfillResult``. + + The dispatcher is async so async editors (``compact_20260112``) can + ``await`` the configured summarization model. Sync editors are called + inline — ``inspect.iscoroutinefunction`` decides how each editor is + invoked. + """ + edits = _normalize_spec(context_management_spec) + if not edits: + return PolyfillResult(messages=messages, system=system, applied_edits=[]) + + current_messages = messages + current_system = system + aggregated_applied: List[AppliedEdit] = [] + aggregated_compaction_block = None + aggregated_iterations_usage = None + + for edit_spec in edits: + edit_type = edit_spec.get("type") + editor = _EDITOR_REGISTRY.get(edit_type) if isinstance(edit_type, str) else None + if editor is None: + verbose_logger.debug( + "context_management polyfill: unknown edit type '%s' — skipping", + edit_type, + ) + continue + + kwargs: Dict[str, Any] = { + "model": model, + "messages": current_messages, + "tools": tools, + "system": current_system, + "edit_spec": edit_spec, + } + # Only async editors accept these — passing them to sync v0 editors + # would break their signature. + if inspect.iscoroutinefunction(editor): + kwargs["litellm_metadata"] = litellm_metadata + kwargs["llm_router"] = llm_router + kwargs["user_api_key_auth"] = user_api_key_auth + raw_result = await cast(Callable[..., Awaitable[Any]], editor)(**kwargs) + else: + raw_result = editor(**kwargs) + + result = _wrap_editor_return(raw_result, fallback_system=current_system) + + current_messages = result.messages + current_system = result.system + aggregated_applied.extend(result.applied_edits) + if result.compaction_block is not None: + aggregated_compaction_block = result.compaction_block + if result.iterations_usage is not None: + aggregated_iterations_usage = result.iterations_usage + + return PolyfillResult( + messages=current_messages, + system=current_system, + applied_edits=aggregated_applied, + compaction_block=aggregated_compaction_block, + iterations_usage=aggregated_iterations_usage, + ) diff --git a/litellm/llms/anthropic/experimental_pass_through/context_management/editors/__init__.py b/litellm/llms/anthropic/experimental_pass_through/context_management/editors/__init__.py new file mode 100644 index 00000000000..3e933a9880a --- /dev/null +++ b/litellm/llms/anthropic/experimental_pass_through/context_management/editors/__init__.py @@ -0,0 +1,4 @@ +from .clear_tool_uses import apply_clear_tool_uses_20250919 +from .compact import apply_compact_20260112 + +__all__ = ["apply_clear_tool_uses_20250919", "apply_compact_20260112"] diff --git a/litellm/llms/anthropic/experimental_pass_through/context_management/editors/clear_tool_uses.py b/litellm/llms/anthropic/experimental_pass_through/context_management/editors/clear_tool_uses.py new file mode 100644 index 00000000000..7b1c20ff522 --- /dev/null +++ b/litellm/llms/anthropic/experimental_pass_through/context_management/editors/clear_tool_uses.py @@ -0,0 +1,210 @@ +"""``clear_tool_uses_20250919`` polyfill (v0: ``trigger`` and ``keep`` only).""" + +from typing import Any, Dict, List, Optional, Tuple, cast + +import litellm +from litellm._logging import verbose_logger +from litellm.types.llms.anthropic import AppliedEdit + +from ..constants import ( + CLEAR_TOOL_USES_EDIT_TYPE, + DEFAULT_INPUT_TOKENS_TRIGGER, + DEFAULT_KEEP_TOOL_USES, +) +from ..placeholders import build_cleared_tool_result_content + + +def _count_tool_uses(messages: List[Dict[str, Any]]) -> int: + """Return the number of tool_use content blocks across all messages. + + Only counts blocks with a string ``id`` to stay consistent with + :func:`_collect_tool_use_ids_in_order`, which is the source of truth for + which blocks are clearable. + """ + count = 0 + for msg in messages: + content = msg.get("content") + if isinstance(content, list): + for block in content: + if isinstance(block, dict) and block.get("type") == "tool_use": + if isinstance(block.get("id"), str): + count += 1 + return count + + +def _collect_tool_use_ids_in_order(messages: List[Dict[str, Any]]) -> List[str]: + """Return tool_use ids in the chronological order they appear in messages.""" + ids: List[str] = [] + for msg in messages: + content = msg.get("content") + if isinstance(content, list): + for block in content: + if isinstance(block, dict) and block.get("type") == "tool_use": + block_id = block.get("id") + if isinstance(block_id, str): + ids.append(block_id) + return ids + + +def _trigger_met( + trigger: Dict[str, Any], + model: str, + messages: List[Dict[str, Any]], + tools: Optional[List[Dict[str, Any]]], +) -> Tuple[bool, Optional[int]]: + """Return (trigger_met, input_tokens if counted for reuse).""" + trigger_type = trigger.get("type", "input_tokens") + threshold = trigger.get("value") + + if trigger_type == "tool_uses": + if not isinstance(threshold, int): + return False, None + return _count_tool_uses(messages) > threshold, None + + if not isinstance(threshold, int): + threshold = DEFAULT_INPUT_TOKENS_TRIGGER + current_tokens = litellm.token_counter( + model=model, + messages=messages, + tools=cast(Any, tools), + ) + verbose_logger.debug( + f"context_management polyfill: current_tokens: {current_tokens}" + ) + verbose_logger.debug(f"context_management polyfill: threshold: {threshold}") + return current_tokens > threshold, current_tokens + + +def _resolve_keep_count(keep: Dict[str, Any]) -> int: + keep_type = keep.get("type", "tool_uses") + if keep_type != "tool_uses": + return DEFAULT_KEEP_TOOL_USES + value = keep.get("value") + if not isinstance(value, int) or value < 0: + return DEFAULT_KEEP_TOOL_USES + return value + + +def _last_completed_tool_use_id( + messages: List[Dict[str, Any]], +) -> Optional[str]: + """Latest completed tool_result id; never cleared.""" + last_id: Optional[str] = None + for msg in messages: + content = msg.get("content") + if isinstance(content, list): + for block in content: + if isinstance(block, dict) and block.get("type") == "tool_result": + block_id = block.get("tool_use_id") + if isinstance(block_id, str): + last_id = block_id + return last_id + + +def _clear_tool_results( + messages: List[Dict[str, Any]], ids_to_clear: set +) -> Tuple[List[Dict[str, Any]], int]: + """Clear matching tool_result content; return (messages, cleared_count).""" + cleared = 0 + new_messages: List[Dict[str, Any]] = [] + for msg in messages: + content = msg.get("content") + if not isinstance(content, list): + new_messages.append(msg) + continue + + new_blocks: List[Any] = [] + mutated = False + for block in content: + if ( + isinstance(block, dict) + and block.get("type") == "tool_result" + and block.get("tool_use_id") in ids_to_clear + ): + new_block = { + **block, + "content": build_cleared_tool_result_content(block.get("content")), + } + new_blocks.append(new_block) + mutated = True + cleared += 1 + else: + new_blocks.append(block) + + if mutated: + new_messages.append({**msg, "content": new_blocks}) + else: + new_messages.append(msg) + + return new_messages, cleared + + +def apply_clear_tool_uses_20250919( + *, + model: str, + messages: List[Dict[str, Any]], + tools: Optional[List[Dict[str, Any]]], + system: Any, + edit_spec: Dict[str, Any], +) -> Tuple[List[Dict[str, Any]], Optional[AppliedEdit]]: + """Apply clear_tool_uses; return (messages, AppliedEdit or None).""" + ignored_knobs = [ + knob + for knob in ("clear_at_least", "exclude_tools", "clear_tool_inputs") + if knob in edit_spec + ] + for ignored_knob in ignored_knobs: + verbose_logger.warning( + "context_management polyfill: ignoring '%s' on %s " + "(supported only on Anthropic-family forwarding path in v0)", + ignored_knob, + CLEAR_TOOL_USES_EDIT_TYPE, + ) + + trigger = edit_spec.get("trigger") or { + "type": "input_tokens", + "value": DEFAULT_INPUT_TOKENS_TRIGGER, + } + keep = edit_spec.get("keep") or { + "type": "tool_uses", + "value": DEFAULT_KEEP_TOOL_USES, + } + + met, tokens_before = _trigger_met(trigger, model, messages, tools) + if not met: + return messages, None + + keep_count = _resolve_keep_count(keep) + tool_use_ids = _collect_tool_use_ids_in_order(messages) + if len(tool_use_ids) <= keep_count: + return messages, None + + ids_to_clear = set(tool_use_ids[: len(tool_use_ids) - keep_count]) + + # Never clear the latest completed tool_result (reply context). + last_completed_id = _last_completed_tool_use_id(messages) + if last_completed_id is not None: + ids_to_clear.discard(last_completed_id) + + edited, cleared_count = _clear_tool_results(messages, ids_to_clear) + verbose_logger.debug("context_management polyfill: edited: %s", edited) + if cleared_count == 0: + return messages, None + + if tokens_before is None: + tokens_before = litellm.token_counter( + model=model, messages=messages, tools=cast(Any, tools) + ) + tokens_after = litellm.token_counter( + model=model, messages=edited, tools=cast(Any, tools) + ) + cleared_input_tokens = max(tokens_before - tokens_after, 0) + + applied: AppliedEdit = { + "type": CLEAR_TOOL_USES_EDIT_TYPE, + "cleared_tool_uses": cleared_count, + "cleared_input_tokens": cleared_input_tokens, + } + if ignored_knobs: + applied["warnings"] = [f"{knob}_ignored" for knob in ignored_knobs] + return edited, applied diff --git a/litellm/llms/anthropic/experimental_pass_through/context_management/editors/compact.py b/litellm/llms/anthropic/experimental_pass_through/context_management/editors/compact.py new file mode 100644 index 00000000000..4aae85b17fe --- /dev/null +++ b/litellm/llms/anthropic/experimental_pass_through/context_management/editors/compact.py @@ -0,0 +1,1206 @@ +"""``compact_20260112`` polyfill (server-side context compaction). + +Mirrors Anthropic's native ``compact_20260112`` for non-Anthropic providers: + +- Scans the message history for an existing ``compaction`` block; everything + before it is dropped (slice). +- If still over the configured trigger, calls a separately-configured + summarization model and synthesizes a fresh ``compaction`` block. +- The summary is injected as a system-message prefix on the downstream call + (the user/assistant log carries no ``compaction`` block downstream). +- The synthesized ``compaction`` block is returned via ``PolyfillResult`` so + the response adapter can prepend it to the response ``content`` array. +""" + +import re +from typing import Any, Dict, List, Literal, Optional, Tuple, Union, cast + +import litellm +from litellm._logging import verbose_logger +from litellm.types.llms.anthropic import ( + AppliedEdit, + CompactionBlock, + UsageIteration, +) + +from ..constants import ( + COMPACT_DEFAULT_INSTRUCTIONS, + COMPACT_DEFAULT_TRIGGER_TOKENS, + COMPACT_EDIT_TYPE, + COMPACT_MIN_TRIGGER_TOKENS, + COMPACT_NO_TOOL_CALLS_SUFFIX, + COMPACT_SUMMARY_MAX_TOKENS, + COMPACT_SUMMARY_MAX_TOKENS_SETTING_KEY, + COMPACT_SUMMARY_MODEL_SETTING_KEY, + COMPACT_SUMMARY_SYSTEM_PREFIX, + COMPACT_SUMMARY_TIMEOUT_SECONDS, +) +from ..errors import AnthropicContextManagementError +from ..result import PolyfillResult + +# Auth metadata fields propagated from the parent request to the summary call +# so the summary's spend is attributed to the same scopes. The list mirrors the +# fields populated by +# ``LiteLLMProxyRequestSetup.add_user_api_key_auth_to_request_metadata``. +# ``user_api_key_model_max_budget`` / ``user_api_key_end_user_model_max_budget`` +# are what ``_PROXY_VirtualKeyModelMaxBudgetLimiter`` reads post-call to update +# the per-model spend caches, so without them the summary spend would never +# count against the caller's model budget. ``user_api_key_end_user_id`` / +# ``user_api_key_project_id`` are the scope identifiers the post-call spend hook +# and rate limiter key their counters on, and ``user_api_end_user_max_budget`` +# is the end-user budget the cost callback enforces — without these the summary +# tokens escape the caller's end-user/project budgets and counters. +_PROPAGATED_METADATA_KEYS = ( + "user_api_key", + "user_api_key_alias", + "user_api_key_team_id", + "user_api_key_team_alias", + "user_api_key_user_id", + "user_api_key_user_email", + "user_api_key_org_id", + "user_api_key_project_id", + "user_api_key_end_user_id", + "user_api_end_user_max_budget", + "user_api_key_model_max_budget", + "user_api_key_end_user_model_max_budget", + "litellm_call_id", + "litellm_parent_otel_span", +) + +_SUMMARY_TAG_RE = re.compile(r"(.*?)", re.IGNORECASE | re.DOTALL) + + +def _read_summary_model_setting() -> Optional[str]: + """Look up the configured summarization model from proxy general_settings.""" + try: + from litellm.proxy.proxy_server import general_settings + except Exception: + return None + value = general_settings.get(COMPACT_SUMMARY_MODEL_SETTING_KEY) + return value if isinstance(value, str) and value else None + + +def _read_summary_max_tokens_setting() -> int: + """Look up the configured summary ``max_tokens`` from proxy general_settings. + + Falls back to :data:`COMPACT_SUMMARY_MAX_TOKENS` when the setting is + missing or invalid (non-positive int, wrong type). Operators tune this + when the default doesn't fit their chosen summary model's output budget. + """ + try: + from litellm.proxy.proxy_server import general_settings + except Exception: + return COMPACT_SUMMARY_MAX_TOKENS + value = general_settings.get(COMPACT_SUMMARY_MAX_TOKENS_SETTING_KEY) + if isinstance(value, int) and value > 0: + return value + return COMPACT_SUMMARY_MAX_TOKENS + + +async def _check_summary_model_access( # noqa: PLR0915 + user_api_key_auth: Any, + summary_model: str, + llm_router: Any, +) -> bool: + """Return True when every model-allowlist scope on the parent request is + satisfied for ``summary_model``. + + The summary subrequest does not pass through ``user_api_key_auth`` again, + so without this gate a caller whose configured scope at any of these + levels excludes ``context_management_summary_model`` could still get the + proxy to invoke that model and return its ```` output as a + compaction block. Mirrors the model-scope enforcement that + ``litellm.proxy.auth.common_checks`` runs for the client-requested model: + key, team, user (personal), project, and team-member allowlists. + + Returns True (allow) when ``user_api_key_auth`` is not present — SDK + callers and tests run outside the proxy, where no key/team policy exists. + Returns False when any of the active allowlists denies the summary model + (``ProxyException`` from ``_can_object_call_model`` / ``can_*_model``). + Unexpected errors during an access check fail closed but are logged + separately so operators can distinguish them from a real access-denied + response. DB-lookup failures (object missing from cache or DB) skip the + corresponding scope — matching ``common_checks``, which only enforces a + scope when its backing object can be loaded. + """ + if user_api_key_auth is None: + return True + try: + from litellm.proxy._types import ProxyException + from litellm.proxy.auth.auth_checks import ( + _can_object_call_model, + can_project_access_model, + can_user_call_model, + get_project_object, + get_team_membership, + get_user_object, + ) + from litellm.proxy.proxy_server import ( + prisma_client, + proxy_logging_obj, + user_api_key_cache, + ) + except Exception: + return True + + key_models = list(getattr(user_api_key_auth, "models", None) or []) + team_id = getattr(user_api_key_auth, "team_id", None) + team_model_aliases = getattr(user_api_key_auth, "team_model_aliases", None) + team_models = list(getattr(user_api_key_auth, "team_models", None) or []) + user_id = getattr(user_api_key_auth, "user_id", None) + project_id = getattr(user_api_key_auth, "project_id", None) + + checks: Tuple[Tuple[Literal["key", "team"], List[str]], ...] = ( + ("key", key_models), + ("team", team_models), + ) + for object_type, models in checks: + if not models: + continue + try: + _can_object_call_model( + model=summary_model, + llm_router=llm_router, + models=models, + team_model_aliases=team_model_aliases, + team_id=team_id, + object_type=object_type, + ) + except ProxyException: + return False + except Exception as e: + verbose_logger.warning( + "compact_20260112: unexpected error during %s-level access " + "check for summary_model=%s; denying access: %s", + object_type, + summary_model, + e, + ) + return False + + if user_id is not None and prisma_client is not None: + try: + user_obj = await get_user_object( + user_id=user_id, + prisma_client=prisma_client, + user_api_key_cache=user_api_key_cache, + user_id_upsert=False, + proxy_logging_obj=proxy_logging_obj, + ) + except Exception as e: + verbose_logger.debug( + "compact_20260112: user object lookup failed for " + "summary_model=%s access check; skipping user-level scope: %s", + summary_model, + e, + ) + user_obj = None + if user_obj is not None: + try: + await can_user_call_model( + model=summary_model, + llm_router=llm_router, + user_object=user_obj, + ) + except ProxyException: + return False + except Exception as e: + verbose_logger.warning( + "compact_20260112: unexpected error during user-level " + "access check for summary_model=%s; denying access: %s", + summary_model, + e, + ) + return False + + if project_id is not None and prisma_client is not None: + try: + project_obj = await get_project_object( + project_id=project_id, + prisma_client=prisma_client, + user_api_key_cache=user_api_key_cache, + proxy_logging_obj=proxy_logging_obj, + ) + except Exception as e: + verbose_logger.debug( + "compact_20260112: project object lookup failed for " + "summary_model=%s access check; skipping project-level scope: %s", + summary_model, + e, + ) + project_obj = None + if project_obj is not None and project_obj.models: + try: + can_project_access_model( + model=summary_model, + project_object=project_obj, + llm_router=llm_router, + ) + except ProxyException: + return False + except Exception as e: + verbose_logger.warning( + "compact_20260112: unexpected error during project-level " + "access check for summary_model=%s; denying access: %s", + summary_model, + e, + ) + return False + + if user_id is not None and team_id is not None and prisma_client is not None: + try: + team_membership = await get_team_membership( + user_id=user_id, + team_id=team_id, + prisma_client=prisma_client, + user_api_key_cache=user_api_key_cache, + proxy_logging_obj=proxy_logging_obj, + ) + except Exception as e: + verbose_logger.debug( + "compact_20260112: team membership lookup failed for " + "summary_model=%s access check; skipping member-level scope: %s", + summary_model, + e, + ) + team_membership = None + member_allowed_models = ( + team_membership.litellm_budget_table.allowed_models + if team_membership is not None + and team_membership.litellm_budget_table is not None + else None + ) + if member_allowed_models: + try: + _can_object_call_model( + model=summary_model, + llm_router=llm_router, + models=list(member_allowed_models), + team_model_aliases=team_model_aliases, + team_id=team_id, + object_type="team", + ) + except ProxyException: + return False + except Exception as e: + verbose_logger.warning( + "compact_20260112: unexpected error during member-level " + "access check for summary_model=%s; denying access: %s", + summary_model, + e, + ) + return False + + return True + + +async def _check_summary_model_budget( + user_api_key_auth: Any, + summary_model: str, +) -> bool: + """Return True when the caller is within their per-model budget for + ``summary_model``. + + The summary subrequest never passes back through ``user_api_key_auth``, so + without this gate a caller whose ``model_max_budget`` for + ``context_management_summary_model`` is exhausted could keep consuming that + model via compaction. Mirrors the ``model_max_budget`` / + ``end_user_model_max_budget`` enforcement that ``user_api_key_auth`` runs for + the client-requested model. Returns True outside the proxy or when no + per-model budget is configured. + """ + if user_api_key_auth is None: + return True + try: + from litellm.proxy.proxy_server import model_max_budget_limiter + except Exception: + return True + + model_max_budget = getattr(user_api_key_auth, "model_max_budget", None) + token = getattr(user_api_key_auth, "token", None) + if isinstance(model_max_budget, dict) and model_max_budget and token is not None: + try: + await model_max_budget_limiter.is_key_within_model_budget( + user_api_key_dict=user_api_key_auth, + model=summary_model, + ) + except litellm.BudgetExceededError: + return False + except Exception as e: + verbose_logger.warning( + "compact_20260112: unexpected error during key model-budget " + "check for summary_model=%s; denying: %s", + summary_model, + e, + ) + return False + + end_user_model_max_budget = getattr( + user_api_key_auth, "end_user_model_max_budget", None + ) + end_user_id = getattr(user_api_key_auth, "end_user_id", None) + if ( + isinstance(end_user_model_max_budget, dict) + and end_user_model_max_budget + and end_user_id is not None + ): + try: + await model_max_budget_limiter.is_end_user_within_model_budget( + end_user_id=end_user_id, + end_user_model_max_budget=end_user_model_max_budget, + model=summary_model, + ) + except litellm.BudgetExceededError: + return False + except Exception as e: + verbose_logger.warning( + "compact_20260112: unexpected error during end-user model-budget " + "check for summary_model=%s; denying: %s", + summary_model, + e, + ) + return False + + return True + + +async def _check_summary_model_rate_limit( + user_api_key_auth: Any, + summary_model: str, +) -> bool: + """Return True when the caller is within their configured RPM/TPM limits + for ``summary_model``. + + The summary subrequest never passes back through the proxy's pre-call + rate limiter, so without this gate a caller already at their key / team / + user RPM or TPM could still drive an extra summary-model completion per + allowed ``/v1/messages`` request. This mirrors the read side of + ``_PROXY_MaxParallelRequestsHandler_v3.async_pre_call_hook`` for the + summary model: it builds the same descriptor set and runs the check in + ``read_only`` mode so no counter is reserved or incremented — the summary + call's actual usage is still charged exactly once by the limiter's + post-call success hook (via the propagated ``litellm_metadata``). + + Returns True (allow) outside the proxy, when the active limiter does not + expose the read-only descriptor check (legacy limiter), or when the + descriptor set cannot be built — the only deny signal is a definitive + ``OVER_LIMIT`` response, so an internal error here forwards the request + uncompacted rather than blocking every summary. + """ + if user_api_key_auth is None: + return True + try: + from litellm.proxy.proxy_server import proxy_logging_obj + except Exception: + return True + + limiter = getattr(proxy_logging_obj, "max_parallel_request_limiter", None) + if ( + limiter is None + or not hasattr(limiter, "should_rate_limit") + or not hasattr(limiter, "_create_rate_limit_descriptors") + ): + return True + + try: + metadata = getattr(user_api_key_auth, "metadata", None) or {} + data = {"model": summary_model} + descriptors = limiter._create_rate_limit_descriptors( + user_api_key_dict=user_api_key_auth, + data=data, + rpm_limit_type=metadata.get("rpm_limit_type"), + tpm_limit_type=metadata.get("tpm_limit_type"), + model_has_failures=False, + ) + limiter._add_team_model_rate_limit_descriptor_from_metadata( + user_api_key_dict=user_api_key_auth, + requested_model=summary_model, + descriptors=descriptors, + ) + limiter._add_project_model_rate_limit_descriptor_from_metadata( + user_api_key_dict=user_api_key_auth, + requested_model=summary_model, + descriptors=descriptors, + ) + descriptors.extend( + limiter.create_organization_rate_limit_descriptor( + user_api_key_auth, summary_model + ) + ) + if not descriptors: + return True + response = await limiter.should_rate_limit( + descriptors=descriptors, + parent_otel_span=getattr(user_api_key_auth, "parent_otel_span", None), + read_only=True, + ) + except Exception as e: + verbose_logger.warning( + "compact_20260112: unexpected error during rate-limit check for " + "summary_model=%s; allowing: %s", + summary_model, + e, + ) + return True + return response.get("overall_code") != "OVER_LIMIT" + + +def _find_latest_compaction_index( + messages: List[Dict[str, Any]], +) -> Tuple[Optional[int], Optional[int]]: + """Return (message_index, block_index) of the most recent compaction block. + + ``None, None`` if no compaction block is present. Iterates from the end so + only the latest one is considered. + """ + for msg_idx in range(len(messages) - 1, -1, -1): + content = messages[msg_idx].get("content") + if not isinstance(content, list): + continue + for blk_idx in range(len(content) - 1, -1, -1): + block = content[blk_idx] + if isinstance(block, dict) and block.get("type") == "compaction": + return msg_idx, blk_idx + return None, None + + +def _slice_around_compaction_block( + messages: List[Dict[str, Any]], +) -> Tuple[List[Dict[str, Any]], Optional[Dict[str, Any]]]: + """Apply Anthropic's "drop everything before the compaction block" rule. + + Returns ``(sliced_messages_with_compaction_block, compaction_block_dict)`` + if a block was found, else ``(original_messages, None)``. The sliced result + keeps the compaction block in the assistant turn that originally carried + it (in practice it's the only block in that turn) so callers can still + extract the summary text from it. + """ + msg_idx, blk_idx = _find_latest_compaction_index(messages) + if msg_idx is None or blk_idx is None: + return messages, None + + original_msg = messages[msg_idx] + original_content = original_msg["content"] + compaction_block = cast(Dict[str, Any], original_content[blk_idx]) + + # Per Anthropic's contract everything before the compaction block is + # dropped, including earlier blocks within the same assistant message. + sliced_content = list(original_content[blk_idx:]) + sliced_first_msg = {**original_msg, "content": sliced_content} + + sliced_messages: List[Dict[str, Any]] = [sliced_first_msg] + sliced_messages.extend(messages[msg_idx + 1 :]) + return sliced_messages, compaction_block + + +def _strip_compaction_blocks( + messages: List[Dict[str, Any]], +) -> List[Dict[str, Any]]: + """Drop any ``compaction`` content blocks from messages. + + Used to build the downstream-bound message list — the adapter has no + concept of a compaction block, so it must not see one. + """ + cleaned: List[Dict[str, Any]] = [] + for msg in messages: + content = msg.get("content") + if not isinstance(content, list): + cleaned.append(msg) + continue + filtered = [ + block + for block in content + if not (isinstance(block, dict) and block.get("type") == "compaction") + ] + if not filtered: + # The compaction block was the only content; drop the whole turn. + continue + cleaned.append({**msg, "content": filtered}) + return cleaned + + +def _augment_system_with_summary( + system: Optional[Union[str, List[Dict[str, Any]]]], + summary_text: str, +) -> Union[str, List[Dict[str, Any]]]: + """Prepend a "Previous conversation summary: ..." block to ``system``.""" + prefix = f"{COMPACT_SUMMARY_SYSTEM_PREFIX}{summary_text}\n\n" + if system is None: + return prefix.rstrip() + if isinstance(system, str): + return f"{prefix}{system}" + # List of content blocks: prepend the prefix to the first text block, + # otherwise insert a new text block at the head. + for idx, block in enumerate(system): + if isinstance(block, dict) and block.get("type") == "text": + existing = block.get("text", "") or "" + new_block = {**block, "text": f"{prefix}{existing}"} + return [*system[:idx], new_block, *system[idx + 1 :]] + return [{"type": "text", "text": prefix.rstrip()}, *system] + + +def _resolve_trigger_tokens(edit_spec: Dict[str, Any]) -> Tuple[int, List[str]]: + """Validate and resolve ``trigger.value``. + + Raises ``AnthropicContextManagementError`` if the explicitly-supplied value + is below the 50k minimum. Unknown ``trigger.type`` values fall back to + ``input_tokens`` with a warning. + """ + warnings: List[str] = [] + trigger = edit_spec.get("trigger") or {} + if not isinstance(trigger, dict): + warnings.append("trigger_not_a_dict_using_default") + return COMPACT_DEFAULT_TRIGGER_TOKENS, warnings + + trigger_type = trigger.get("type", "input_tokens") + if trigger_type != "input_tokens": + warnings.append(f"unsupported_trigger_type_{trigger_type}_using_input_tokens") + + value = trigger.get("value") + if value is None: + return COMPACT_DEFAULT_TRIGGER_TOKENS, warnings + if not isinstance(value, int): + warnings.append("trigger_value_not_int_using_default") + return COMPACT_DEFAULT_TRIGGER_TOKENS, warnings + if value < COMPACT_MIN_TRIGGER_TOKENS: + raise AnthropicContextManagementError( + status_code=400, + message=( + f"context_management.compact_20260112.trigger.value must be at " + f"least {COMPACT_MIN_TRIGGER_TOKENS} tokens" + ), + ) + return value, warnings + + +def _build_summary_prompt( + edit_spec: Dict[str, Any], tools: Optional[List[Dict[str, Any]]] +) -> str: + custom = edit_spec.get("instructions") + if isinstance(custom, str) and custom.strip(): + return custom + prompt = COMPACT_DEFAULT_INSTRUCTIONS + if tools: + prompt = f"{prompt}{COMPACT_NO_TOOL_CALLS_SUFFIX}" + return prompt + + +def _propagate_metadata( + parent_litellm_metadata: Optional[Dict[str, Any]], +) -> Dict[str, Any]: + """Extract the parent request's auth/spend-attribution fields for the summary subcall. + + The proxy attaches ``user_api_key``, ``user_api_key_team_id`` etc. to + ``data["litellm_metadata"]`` (see + ``LiteLLMProxyRequestSetup.add_user_api_key_auth_to_request_metadata``). + Without these on the summary subrequest, the router's post-call hooks + cannot attribute summary tokens to the caller's key/team budget. + """ + if not parent_litellm_metadata: + return {} + propagated: Dict[str, Any] = {} + for key in _PROPAGATED_METADATA_KEYS: + if key in parent_litellm_metadata: + propagated[key] = parent_litellm_metadata[key] + return propagated + + +def _count_effective_tokens( + model: str, + effective_messages: List[Dict[str, Any]], + compaction_block: Optional[Dict[str, Any]], + tools: Optional[List[Dict[str, Any]]], + system: Optional[Union[str, List[Dict[str, Any]]]] = None, +) -> int: + """Token-count the conversation as it will appear downstream. + + The compaction block (if any) becomes a system prefix on the downstream + call, so its content still counts even though it isn't in ``messages``. + The system prompt (which may already include a prior compaction summary + prepended via ``_augment_system_with_summary``) is also counted so the + threshold check matches the downstream ``input_tokens`` metric. + """ + # Local import to avoid pulling the adapter at module load time. + from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import ( + LiteLLMAnthropicMessagesAdapter, + ) + + messages_without_compaction = _strip_compaction_blocks(effective_messages) + adapter = LiteLLMAnthropicMessagesAdapter() + try: + openai_shape = adapter.translate_anthropic_messages_to_openai( + messages=cast(Any, messages_without_compaction) + ) + except Exception as e: + verbose_logger.debug( + "compact_20260112: anthropic→openai translation failed during token " + "count, falling back to raw messages: %s", + e, + ) + openai_shape = cast(Any, messages_without_compaction) + + # Translate Anthropic-shaped tools (``input_schema``) to OpenAI-shaped + # tools (``{"type": "function", "function": {...}}``) so ``token_counter`` + # gets a consistent format regardless of which counting path it uses. + # An inaccurate tool token count here could cause the polyfill to skip + # needed compaction or trigger unnecessary summarization. + openai_tools: Optional[List[Dict[str, Any]]] = None + if tools: + try: + translated_tools, _ = adapter.translate_anthropic_tools_to_openai( + tools=cast(Any, tools) + ) + openai_tools = cast(List[Dict[str, Any]], translated_tools) + except Exception as e: + verbose_logger.debug( + "compact_20260112: anthropic→openai tools translation failed " + "during token count, falling back to raw tools: %s", + e, + ) + openai_tools = tools + + total = litellm.token_counter( + model=model, + messages=cast(Any, openai_shape), + tools=cast(Any, openai_tools), + ) + if compaction_block is not None: + content = compaction_block.get("content") or "" + if content: + total += litellm.token_counter(model=model, text=content) + system_text = _system_to_text(system) + if system_text: + total += litellm.token_counter(model=model, text=system_text) + return total + + +def _system_to_text( + system: Optional[Union[str, List[Dict[str, Any]]]], +) -> str: + """Flatten an Anthropic-style ``system`` value into a single string for + token counting. Returns ``""`` when ``system`` carries no text.""" + if system is None: + return "" + if isinstance(system, str): + return system + parts: List[str] = [] + for block in system: + if isinstance(block, dict) and block.get("type") == "text": + text = block.get("text") + if isinstance(text, str) and text: + parts.append(text) + return "\n".join(parts) + + +def _select_last_user_question( + messages: List[Dict[str, Any]], +) -> List[Dict[str, Any]]: + """Pick the most recent ``user`` turn that is a real question. + + Returns a one-element message list with any ``tool_result`` blocks + stripped: after compaction the paired ``tool_use`` assistant turn no + longer exists in the downstream context, so forwarding ``tool_result`` + blocks would translate to orphaned ``role=tool`` messages on + non-Anthropic providers (OpenAI, Gemini, …) and cause a 400 error. + + Falls back to a synthetic continuation prompt if no eligible turn + exists (e.g. the conversation only ever contained ``tool_result`` + turns, or contained no user turns at all). The downstream call always + needs a non-empty user message. + """ + for msg in reversed(messages): + if msg.get("role") != "user": + continue + content = msg.get("content") + if isinstance(content, list): + filtered = [ + blk + for blk in content + if not (isinstance(blk, dict) and blk.get("type") == "tool_result") + ] + if not filtered: + # Purely tool_result — skip and look for an earlier turn. + continue + if len(filtered) < len(content): + return [{**msg, "content": filtered}] + return [msg] + return [ + { + "role": "user", + "content": "Please continue based on the conversation summary above.", + } + ] + + +def _extract_summary_text(raw: Optional[str]) -> Optional[str]: + if not raw: + return None + match = _SUMMARY_TAG_RE.search(raw) + if match is None: + return None + summary = match.group(1).strip() + return summary or None + + +def _system_to_openai_message( + system: Optional[Union[str, List[Dict[str, Any]]]], +) -> Optional[Dict[str, Any]]: + """Translate Anthropic-shaped ``system`` to an OpenAI system message. + + Accepts a bare string or a list of Anthropic content blocks; returns + ``None`` if no usable text is present. Only ``type=="text"`` blocks are + carried over — the summary model has no use for ``cache_control`` or + other non-text metadata. + """ + if isinstance(system, str): + return {"role": "system", "content": system} if system else None + if isinstance(system, list): + parts = [ + block.get("text", "") + for block in system + if isinstance(block, dict) and block.get("type") == "text" + ] + joined = "\n\n".join(part for part in parts if part) + return {"role": "system", "content": joined} if joined else None + return None + + +def _build_summary_messages( + effective_messages: List[Dict[str, Any]], + prompt: str, + system: Optional[Union[str, List[Dict[str, Any]]]] = None, +) -> List[Dict[str, Any]]: + """Build the OpenAI-shape message list for the summary call. + + The caller's ``system`` prompt is prepended (the default summarization + instructions reference "the initial task above", which lives in that + system prompt); the conversation history is translated to OpenAI shape; + the summarization prompt is appended as a final user turn. + """ + from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import ( + LiteLLMAnthropicMessagesAdapter, + ) + + stripped = _strip_compaction_blocks(effective_messages) + try: + openai_messages = ( + LiteLLMAnthropicMessagesAdapter().translate_anthropic_messages_to_openai( + messages=cast(Any, stripped) + ) + ) + except Exception as e: + verbose_logger.warning( + "compact_20260112: anthropic→openai translation failed when " + "building summary call; falling back to raw shape: %s", + e, + ) + openai_messages = cast(Any, stripped) + + summary_messages: List[Dict[str, Any]] = [] + system_message = _system_to_openai_message(system) + if system_message is not None: + summary_messages.append(system_message) + summary_messages.extend(openai_messages) + # If the last turn is already a user message, merge the summarization + # prompt into it. Some providers (and strict OpenAI-compatible endpoints) + # reject two consecutive ``role=user`` messages, which would otherwise + # silently fall into the ``summary_call_failed`` error path. + if summary_messages and _is_user_message(summary_messages[-1]): + last_msg = summary_messages[-1] + summary_messages[-1] = { + **last_msg, + "content": _append_text_to_content(last_msg.get("content"), prompt), + } + else: + summary_messages.append({"role": "user", "content": prompt}) + return summary_messages + + +def _is_user_message(msg: Any) -> bool: + return isinstance(msg, dict) and msg.get("role") == "user" + + +def _append_text_to_content(content: Any, extra_text: str) -> Any: + """Append ``extra_text`` to an OpenAI-shape message ``content`` field. + + Handles the two common shapes: ``str`` and ``list`` of content parts. + For unexpected/empty shapes, fall back so the caller gets a usable value. + """ + if content is None or content == "": + return extra_text + if isinstance(content, str): + return f"{content}\n\n{extra_text}" + if isinstance(content, list): + return [*content, {"type": "text", "text": extra_text}] + return [content, {"type": "text", "text": extra_text}] + + +async def _call_summary_model( + *, + summary_model: str, + summary_messages: List[Dict[str, Any]], + metadata: Dict[str, Any], + llm_router: Any, + allowed_model_region: Optional[str] = None, + max_tokens: int = COMPACT_SUMMARY_MAX_TOKENS, +) -> Any: + """Invoke the configured summary model. + + Prefers ``llm_router.acompletion`` so the model alias resolves against the + proxy's ``model_list``; falls back to ``litellm.acompletion`` if no router + is available (e.g. SDK usage outside the proxy). + """ + # ``max_tokens`` is required by providers like Anthropic and silently + # accepted by providers that don't strictly require it (OpenAI etc.). + # Setting a sensible default here means the feature works regardless of + # which model an admin configures as ``context_management_summary_model``; + # operators can override via ``context_management_summary_max_tokens`` in + # ``general_settings`` when the default doesn't fit the chosen model's + # output budget. + # The propagated proxy auth/spend-attribution fields (``user_api_key`` etc.) + # must travel as ``litellm_metadata`` — that is the parameter the proxy's + # post-call spend hooks read for budget attribution. The provider-level + # ``metadata`` kwarg corresponds to the upstream API request body and would + # not flow into spend tracking. + # ``allowed_model_region`` must travel as a top-level kwarg because the + # router enforces region restrictions by reading ``request_kwargs`` directly + # (see ``Router._common_checks_available_deployment``); without this the + # summary subrequest could be routed to a deployment outside the caller's + # permitted region. + # ``timeout`` bounds how long a slow/unresponsive summary model can stall + # the parent ``/v1/messages`` request. On timeout the caller catches the + # exception and surfaces ``applied_edits[0].error = "summary_call_failed"``, + # forwarding the request without compaction rather than hanging. + call_kwargs: Dict[str, Any] = { + "model": summary_model, + "messages": summary_messages, + "max_tokens": max_tokens, + "timeout": COMPACT_SUMMARY_TIMEOUT_SECONDS, + "litellm_metadata": metadata, + } + # The end-user id must also travel as the top-level ``user`` kwarg: legacy + # limiter hooks and prometheus end-user tracking read it from there rather + # than from ``litellm_metadata``, so without it the summary tokens would not + # debit the caller's end-user counters. + end_user_id = metadata.get("user_api_key_end_user_id") + if end_user_id: + call_kwargs["user"] = end_user_id + if allowed_model_region is not None: + call_kwargs["allowed_model_region"] = allowed_model_region + if llm_router is not None and hasattr(llm_router, "acompletion"): + return await llm_router.acompletion(**call_kwargs) + return await litellm.acompletion(**call_kwargs) + + +def _extract_response_text(response: Any) -> Optional[str]: + try: + choice = response.choices[0] + message = choice.message + content = getattr(message, "content", None) + if isinstance(content, str): + return content + # Some providers return a list of content parts. + if isinstance(content, list): + text_parts = [ + part.get("text", "") + for part in content + if isinstance(part, dict) and part.get("type") == "text" + ] + return "".join(text_parts) or None + except (AttributeError, IndexError, KeyError): + return None + return None + + +def _extract_usage(response: Any) -> Tuple[int, int]: + usage = getattr(response, "usage", None) + if usage is None: + return 0, 0 + return ( + int(getattr(usage, "prompt_tokens", 0) or 0), + int(getattr(usage, "completion_tokens", 0) or 0), + ) + + +def apply_client_compaction_block_history( + *, + messages: List[Dict[str, Any]], + system: Optional[Union[str, List[Dict[str, Any]]]], +) -> Optional[PolyfillResult]: + """Honor client-sent compaction blocks without a ``compact_20260112`` edit. + + When the request omits ``context_management`` but the message history already + contains a ``compaction`` content block (e.g. Claude Code client-side + compaction), apply the same slice-only forwarding as the under-threshold + path: the prior summary is prepended to ``system`` and the post-compaction + tail is forwarded unchanged (with compaction blocks stripped) so recent + turns the summary does not cover are preserved. + """ + effective_messages, prior_compaction_block = _slice_around_compaction_block( + messages + ) + if prior_compaction_block is None: + return None + + verbose_logger.info( + "compact_20260112: client compaction block in message history; " + "applying slice-only forwarding (no context_management edit)" + ) + + prior_summary_text = prior_compaction_block.get("content") or "" + augmented_system: Union[str, List[Dict[str, Any]], None] = system + if isinstance(prior_summary_text, str) and prior_summary_text: + augmented_system = _augment_system_with_summary(system, prior_summary_text) + verbose_logger.info( + "compact_20260112: compaction summary added to main call system prefix (%s chars)", + len(prior_summary_text), + ) + + # Post-compaction turns are recent context the prior summary does not cover, + # so forward them unchanged. Only fall back to the last user question if the + # strip leaves the downstream call with nothing to answer. + downstream_messages = _strip_compaction_blocks(effective_messages) + if not downstream_messages: + downstream_messages = _select_last_user_question(effective_messages) + + return PolyfillResult( + messages=downstream_messages, + system=augmented_system, + applied_edits=[], + ) + + +async def apply_compact_20260112( # noqa: PLR0915 + *, + model: str, + messages: List[Dict[str, Any]], + tools: Optional[List[Dict[str, Any]]], + system: Optional[Union[str, List[Dict[str, Any]]]], + edit_spec: Dict[str, Any], + litellm_metadata: Optional[Dict[str, Any]] = None, + llm_router: Any = None, + user_api_key_auth: Any = None, +) -> PolyfillResult: + """Apply ``compact_20260112``; return a ``PolyfillResult``. + + See module docstring for the algorithm. Errors are best-effort: when the + summary call fails or the response is malformed, the editor returns the + pre-summary state (with ``applied_edits[0].error`` populated) so the + original request still proceeds. + """ + # Validation runs first. Raising AnthropicContextManagementError here is + # the only path on which the polyfill aborts the request. + trigger_tokens, warnings = _resolve_trigger_tokens(edit_spec) + verbose_logger.info( + "compact_20260112: request has compaction trigger (input_tokens threshold=%s)", + trigger_tokens, + ) + if edit_spec.get("pause_after_compaction"): + warnings.append("pause_after_compaction_ignored") + + applied: AppliedEdit = {"type": COMPACT_EDIT_TYPE} + if warnings: + applied["warnings"] = warnings + + # Phase A: slice around any existing compaction block. Runs before the + # opt-in gate below so that even when summarization is disabled we still + # strip Anthropic-only ``compaction`` blocks from messages going to + # non-Anthropic backends (which would reject them). + effective_messages, prior_compaction_block = _slice_around_compaction_block( + messages + ) + prior_summary_text = ( + prior_compaction_block.get("content") if prior_compaction_block else None + ) + augmented_system: Union[str, List[Dict[str, Any]], None] = system + if isinstance(prior_summary_text, str) and prior_summary_text: + augmented_system = _augment_system_with_summary(system, prior_summary_text) + verbose_logger.info( + "compact_20260112: compaction summary added to main call system prefix (%s chars)", + len(prior_summary_text), + ) + + downstream_messages = _strip_compaction_blocks(effective_messages) + + # Opt-in gate: no summary model configured → no-op (but still return the + # Phase A-sliced/stripped messages so compaction blocks don't leak). + summary_model = _read_summary_model_setting() + if summary_model is None: + applied["error"] = "summary_model_not_configured" + # Slice-only forwarding: ``augmented_system`` already carries any prior + # compaction summary, and the post-compaction tail in + # ``downstream_messages`` is recent context the summary does not cover, + # so forward it unchanged. Only fall back to the last user question when + # the strip leaves nothing for the downstream call to answer. + if not downstream_messages: + downstream_messages = _select_last_user_question(effective_messages) + return PolyfillResult( + messages=downstream_messages, + system=augmented_system, + applied_edits=[applied], + ) + + # Phase B: threshold check. + try: + current_tokens = _count_effective_tokens( + model=model, + effective_messages=effective_messages, + # ``augmented_system`` already carries the prior compaction summary + # (prepended via ``_augment_system_with_summary``); pass ``None`` + # here so we don't double-count the summary text. + compaction_block=None, + tools=tools, + system=augmented_system, + ) + except Exception as e: + verbose_logger.warning( + "compact_20260112: token_counter failed; assuming under threshold: %s", e + ) + current_tokens = 0 + + verbose_logger.debug( + "compact_20260112: current_tokens=%s trigger=%s", current_tokens, trigger_tokens + ) + + if current_tokens <= trigger_tokens: + # Slice-only path: the prior compaction summary already lives in + # ``augmented_system``. Post-compaction turns are recent context the + # summary does not cover, so forward ``downstream_messages`` (the + # post-compaction tail with compaction blocks stripped) unchanged. + # Only fall back to the last user question when the strip leaves + # nothing for the downstream call to answer. + if not downstream_messages: + downstream_messages = _select_last_user_question(effective_messages) + return PolyfillResult( + messages=downstream_messages, + system=augmented_system, + applied_edits=[applied], + ) + + # Phase C: summarize. ``augmented_system`` carries any prior compaction + # summary so multi-round compaction does not lose accumulated history — + # ``effective_messages`` only contains turns since the last compaction. + if not await _check_summary_model_access( + user_api_key_auth=user_api_key_auth, + summary_model=summary_model, + llm_router=llm_router, + ): + verbose_logger.warning( + "compact_20260112: caller not authorized for summary_model=%s; " + "skipping summary call", + summary_model, + ) + applied["error"] = "summary_model_access_denied" + return PolyfillResult( + messages=downstream_messages, + system=augmented_system, + applied_edits=[applied], + ) + + if not await _check_summary_model_budget( + user_api_key_auth=user_api_key_auth, + summary_model=summary_model, + ): + verbose_logger.warning( + "compact_20260112: caller over model budget for summary_model=%s; " + "skipping summary call", + summary_model, + ) + applied["error"] = "summary_model_budget_exceeded" + return PolyfillResult( + messages=downstream_messages, + system=augmented_system, + applied_edits=[applied], + ) + + if not await _check_summary_model_rate_limit( + user_api_key_auth=user_api_key_auth, + summary_model=summary_model, + ): + verbose_logger.warning( + "compact_20260112: caller over rate limit for summary_model=%s; " + "skipping summary call", + summary_model, + ) + applied["error"] = "summary_model_rate_limit_exceeded" + return PolyfillResult( + messages=downstream_messages, + system=augmented_system, + applied_edits=[applied], + ) + + prompt = _build_summary_prompt(edit_spec, tools) + summary_messages = _build_summary_messages( + effective_messages, prompt, system=augmented_system + ) + propagated_metadata = _propagate_metadata(litellm_metadata) + allowed_model_region = getattr(user_api_key_auth, "allowed_model_region", None) + + try: + response = await _call_summary_model( + summary_model=summary_model, + summary_messages=summary_messages, + metadata=propagated_metadata, + llm_router=llm_router, + allowed_model_region=allowed_model_region, + max_tokens=_read_summary_max_tokens_setting(), + ) + except Exception as e: + verbose_logger.warning("compact_20260112: summary call failed: %s", e) + applied["error"] = "summary_call_failed" + return PolyfillResult( + messages=downstream_messages, + system=augmented_system, + applied_edits=[applied], + ) + + summary_text = _extract_summary_text(_extract_response_text(response)) + if summary_text is None: + applied["error"] = "summary_extraction_failed" + return PolyfillResult( + messages=downstream_messages, + system=augmented_system, + applied_edits=[applied], + ) + + summary_input_tokens, summary_output_tokens = _extract_usage(response) + applied["summary_input_tokens"] = summary_input_tokens + applied["summary_output_tokens"] = summary_output_tokens + + compaction_block: CompactionBlock = { + "type": "compaction", + "content": summary_text, + } + iterations_usage: List[UsageIteration] = [ + { + "type": "compaction", + "input_tokens": summary_input_tokens, + "output_tokens": summary_output_tokens, + } + ] + + # Per Anthropic's contract, everything before the compaction block is + # dropped. Phase D: the user/assistant log goes empty; the summary lives + # on the system message instead. Anthropic requires a non-empty messages + # array, so keep the most recent original user *question* turn so the + # model has something to answer. Skip ``tool_result``-only user turns: + # in Anthropic's format those are role=user but represent the response + # from a tool, and surfacing one as the sole downstream message would + # produce an orphaned ``tool``-role message on non-Anthropic providers + # with no matching ``tool_calls`` in the prior assistant history. If no + # eligible turn exists, fall back to a synthetic continuation prompt so + # the downstream call still has a non-empty user message. + summarized_system = _augment_system_with_summary(system, summary_text) + verbose_logger.info( + "compact_20260112: compaction summary added to main call system prefix (%s chars)", + len(summary_text), + ) + downstream_messages_after_summary = _select_last_user_question(effective_messages) + + return PolyfillResult( + messages=downstream_messages_after_summary, + system=summarized_system, + applied_edits=[applied], + compaction_block=compaction_block, + iterations_usage=iterations_usage, + ) diff --git a/litellm/llms/anthropic/experimental_pass_through/context_management/errors.py b/litellm/llms/anthropic/experimental_pass_through/context_management/errors.py new file mode 100644 index 00000000000..1b14089a451 --- /dev/null +++ b/litellm/llms/anthropic/experimental_pass_through/context_management/errors.py @@ -0,0 +1,14 @@ +"""Exceptions raised by the context_management polyfill.""" + + +class AnthropicContextManagementError(Exception): + """Validation error from the polyfill, surfaced as an Anthropic-format 4xx. + + The `/v1/messages` endpoint catches this in its exception handler and + emits an Anthropic-shaped error body instead of the default OpenAI shape. + """ + + def __init__(self, *, status_code: int, message: str) -> None: + super().__init__(message) + self.status_code = status_code + self.message = message diff --git a/litellm/llms/anthropic/experimental_pass_through/context_management/placeholders.py b/litellm/llms/anthropic/experimental_pass_through/context_management/placeholders.py new file mode 100644 index 00000000000..f684d970df4 --- /dev/null +++ b/litellm/llms/anthropic/experimental_pass_through/context_management/placeholders.py @@ -0,0 +1,14 @@ +"""Placeholder content for cleared ``tool_result`` blocks (string or block list).""" + +from typing import Any, List, Union + +from .constants import CLEARED_TOOL_RESULT_PLACEHOLDER + + +def build_cleared_tool_result_content( + original_content: Any, +) -> Union[str, List[dict]]: + """Return a string or single text block list, matching ``original_content`` shape.""" + if isinstance(original_content, list): + return [{"type": "text", "text": CLEARED_TOOL_RESULT_PLACEHOLDER}] + return CLEARED_TOOL_RESULT_PLACEHOLDER diff --git a/litellm/llms/anthropic/experimental_pass_through/context_management/result.py b/litellm/llms/anthropic/experimental_pass_through/context_management/result.py new file mode 100644 index 00000000000..36bcde98d0c --- /dev/null +++ b/litellm/llms/anthropic/experimental_pass_through/context_management/result.py @@ -0,0 +1,53 @@ +"""``PolyfillResult`` — the shape returned by the context-management dispatcher. + +Threaded from the dispatcher through ``async_anthropic_messages_handler`` into +the adapter so it can prepend the ``compaction`` block to the response and +attach ``iterations`` to ``usage``. +""" + +from dataclasses import dataclass, field +from typing import Any, Dict, List, Optional, Union + +from litellm.types.llms.anthropic import ( + AppliedEdit, + CompactionBlock, + UsageIteration, +) + +from .constants import COMPACT_EDIT_TYPE + + +@dataclass +class PolyfillResult: + messages: List[Dict[str, Any]] + system: Optional[Union[str, List[Dict[str, Any]]]] + applied_edits: List[AppliedEdit] = field(default_factory=list) + compaction_block: Optional[CompactionBlock] = None + iterations_usage: Optional[List[UsageIteration]] = None + + def applied_edits_for_response(self) -> Optional[List[AppliedEdit]]: + """``applied_edits`` to attach on the client-visible response. + + ``compact_20260112`` is included when a new compaction block was + synthesized (success), when the edit carries an ``error`` field + (``summary_model_not_configured``, ``summary_call_failed``, + ``summary_extraction_failed``), or when the edit carries + ``warnings`` (e.g. ``unsupported_trigger_type_X_using_input_tokens``, + ``pause_after_compaction_ignored``) — operators and clients need to + see why compaction was requested but not applied as expected. + Slice-only / under-threshold paths that produced no edit at all + (no block, no error, no warnings) are omitted. Other edit types are + included when the editor returned an ``AppliedEdit``. + """ + visible: List[AppliedEdit] = [] + for edit in self.applied_edits: + if edit.get("type") == COMPACT_EDIT_TYPE: + if ( + self.compaction_block is not None + or edit.get("error") + or edit.get("warnings") + ): + visible.append(edit) + else: + visible.append(edit) + return visible or None diff --git a/litellm/llms/anthropic/experimental_pass_through/messages/agentic_streaming_iterator.py b/litellm/llms/anthropic/experimental_pass_through/messages/agentic_streaming_iterator.py index d0780c82d06..d693d50b8e5 100644 --- a/litellm/llms/anthropic/experimental_pass_through/messages/agentic_streaming_iterator.py +++ b/litellm/llms/anthropic/experimental_pass_through/messages/agentic_streaming_iterator.py @@ -13,7 +13,6 @@ from typing import Any, AsyncIterator, Dict, List, Optional, cast from litellm._logging import verbose_logger - # --------------------------------------------------------------------------- # SSE parsing helpers (module-level to keep the class lean) # --------------------------------------------------------------------------- diff --git a/litellm/llms/anthropic/experimental_pass_through/messages/handler.py b/litellm/llms/anthropic/experimental_pass_through/messages/handler.py index 0c59e812e0b..62eced8e6f0 100644 --- a/litellm/llms/anthropic/experimental_pass_through/messages/handler.py +++ b/litellm/llms/anthropic/experimental_pass_through/messages/handler.py @@ -8,10 +8,23 @@ import asyncio import contextvars from functools import partial -from typing import Any, AsyncIterator, Coroutine, Dict, List, Optional, Union, cast +from typing import ( + Any, + AsyncIterator, + Coroutine, + Dict, + Iterator, + List, + Optional, + Union, + cast, +) import litellm from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.llms.anthropic.common_utils import ( + strip_empty_text_blocks_from_anthropic_messages, +) from litellm.llms.base_llm.anthropic_messages.transformation import ( BaseAnthropicMessagesConfig, ) @@ -186,10 +199,22 @@ async def anthropic_messages( client: Optional[AsyncHTTPHandler] = None, custom_llm_provider: Optional[str] = None, **kwargs, -) -> Union[AnthropicMessagesResponse, AsyncIterator]: +) -> Union[AnthropicMessagesResponse, Iterator[bytes], AsyncIterator[Any]]: """ - Async: Make llm api request in Anthropic /messages API spec + Async: Make llm api request in Anthropic /messages API spec. + + Runs the empty-text-block sanitizer before any backend dispatch. """ + # Anthropic's API rejects requests containing empty / whitespace-only + # text content blocks with "messages: text content blocks must be + # non-empty". Multi-turn tool-use clients (e.g. Claude Code) routinely + # loop assistant responses that contain {"type": "text", "text": ""} + # alongside tool_use blocks back as conversation history, which then + # causes the next /v1/messages call to 400. /v1/chat/completions + # already handles this in anthropic_messages_pt; sanitize the native + # Anthropic Messages path here for the same guarantee. See #22930. + messages = strip_empty_text_blocks_from_anthropic_messages(messages) + original_stream = stream or kwargs.get( "_websearch_interception_converted_stream", False ) @@ -278,6 +303,12 @@ async def anthropic_messages( api_base=api_base, client=client, custom_llm_provider=custom_llm_provider, + # messages were already empty-text-block sanitized at the top of this + # function and are NOT reassigned before this dispatch, so the handler + # can skip its (otherwise redundant) second full-messages scan. Passed + # explicitly (not via **kwargs) so it only affects this direct + # dispatch -- interceptor / sync entry points still sanitize. + _litellm_messages_presanitized=True, **kwargs, ) ctx = contextvars.copy_context() @@ -325,8 +356,11 @@ def anthropic_messages_handler( **kwargs, ) -> Union[ AnthropicMessagesResponse, + Iterator[bytes], AsyncIterator[Any], - Coroutine[Any, Any, Union[AnthropicMessagesResponse, AsyncIterator[Any]]], + Coroutine[ + Any, Any, Union[AnthropicMessagesResponse, AsyncIterator[Any], Iterator[bytes]] + ], ]: """ Makes Anthropic `/v1/messages` API calls In the Anthropic API Spec @@ -336,6 +370,15 @@ def anthropic_messages_handler( """ from litellm.types.utils import LlmProviders + # Sanitize empty text blocks so the sync entry point + # (litellm.messages.create -> anthropic_messages_handler) gets the same + # protection as the async wrapper. The async wrapper already sanitized and + # does not reassign messages before dispatch, so it sets + # ``_litellm_messages_presanitized`` to skip this redundant second + # full-messages scan. Pop it so it never leaks into provider params. + if not kwargs.pop("_litellm_messages_presanitized", False): + messages = strip_empty_text_blocks_from_anthropic_messages(messages) + metadata = validate_anthropic_api_metadata(metadata) local_vars = locals() @@ -426,9 +469,14 @@ def anthropic_messages_handler( return LiteLLMMessagesToResponsesAPIHandler.anthropic_messages_handler( **_shared_kwargs ) + + # The in-gateway context_management polyfill runs inside + # ``async_anthropic_messages_handler`` so it can ``await`` the + # summarization model for ``compact_20260112``. ``context_management`` + # is passed through as a regular kwarg. return ( LiteLLMMessagesToCompletionTransformationHandler.anthropic_messages_handler( - **_shared_kwargs + **_shared_kwargs, ) ) diff --git a/litellm/llms/anthropic/experimental_pass_through/messages/transformation.py b/litellm/llms/anthropic/experimental_pass_through/messages/transformation.py index 35495d59610..3a2c09f2183 100644 --- a/litellm/llms/anthropic/experimental_pass_through/messages/transformation.py +++ b/litellm/llms/anthropic/experimental_pass_through/messages/transformation.py @@ -230,6 +230,8 @@ class AnthropicMessagesConfig(BaseAnthropicMessagesConfig): """Translate legacy ``thinking.type=enabled`` to adaptive for 4.6/4.7. Caller-provided ``output_config.effort`` is never overridden. """ + from litellm.llms.anthropic.chat.transformation import AnthropicConfig + if not AnthropicModelInfo._is_adaptive_thinking_model(model): return thinking = optional_params.get("thinking") @@ -237,7 +239,7 @@ class AnthropicMessagesConfig(BaseAnthropicMessagesConfig): return budget = int(thinking.get("budget_tokens") or 0) - if budget >= 24000: + if budget >= 24000 and AnthropicConfig._supports_effort_level(model, "xhigh"): effort = "xhigh" elif budget >= 10000: effort = "high" @@ -312,7 +314,10 @@ class AnthropicMessagesConfig(BaseAnthropicMessagesConfig): ) ####### get required params for all anthropic messages requests ###### - verbose_logger.debug(f"TRANSFORMATION DEBUG - Messages: {messages}") + # Lazy %s: the f-string previously stringified the entire messages + # payload on every request regardless of log level (a full scan of the + # request body on the hot path). Defer it to when DEBUG is enabled. + verbose_logger.debug("TRANSFORMATION DEBUG - Messages: %s", messages) # Auto-strip advisor blocks from history if advisor tool is absent. # Prevents Anthropic 400: advisor_tool_result in history requires advisor tool. @@ -424,8 +429,13 @@ class AnthropicMessagesConfig(BaseAnthropicMessagesConfig): ANTHROPIC_BETA_HEADER_VALUES.CONTEXT_MANAGEMENT_2025_06_27.value ) - # Check for structured outputs - if optional_params.get("output_format") is not None: + # Check for structured outputs. Anthropic's newer request shape nests + # the schema under output_config.format; the older top-level + # output_format remains supported for backwards compatibility. + output_config = optional_params.get("output_config") + if optional_params.get("output_format") is not None or ( + isinstance(output_config, dict) and output_config.get("format") is not None + ): beta_values.add( ANTHROPIC_BETA_HEADER_VALUES.STRUCTURED_OUTPUT_2025_09_25.value ) diff --git a/litellm/llms/anthropic/experimental_pass_through/messages/utils.py b/litellm/llms/anthropic/experimental_pass_through/messages/utils.py index fa951ebd2e5..88832fb3f63 100644 --- a/litellm/llms/anthropic/experimental_pass_through/messages/utils.py +++ b/litellm/llms/anthropic/experimental_pass_through/messages/utils.py @@ -1,4 +1,5 @@ -from typing import Any, Dict, List, cast, get_type_hints +from functools import lru_cache +from typing import Any, Dict, FrozenSet, List, cast, get_type_hints from litellm.types.llms.anthropic import AnthropicMessagesRequestOptionalParams from litellm.types.llms.anthropic_messages.anthropic_response import ( @@ -6,6 +7,18 @@ from litellm.types.llms.anthropic_messages.anthropic_response import ( ) +@lru_cache(maxsize=1) +def _anthropic_messages_optional_param_keys() -> FrozenSet[str]: + """ + Valid AnthropicMessagesRequestOptionalParams keys. + + ``typing.get_type_hints`` is ~80us/call and this TypedDict is static, so + resolving it once per process instead of once per request removes a fixed + full-pass cost from the /v1/messages request-parse path. + """ + return frozenset(get_type_hints(AnthropicMessagesRequestOptionalParams).keys()) + + class AnthropicMessagesRequestUtils: @staticmethod def get_requested_anthropic_messages_optional_param( @@ -20,7 +33,7 @@ class AnthropicMessagesRequestUtils: Returns: AnthropicMessagesRequestOptionalParams instance with only the valid parameters """ - valid_keys = get_type_hints(AnthropicMessagesRequestOptionalParams).keys() + valid_keys = _anthropic_messages_optional_param_keys() filtered_params = { k: v for k, v in params.items() if k in valid_keys and v is not None } diff --git a/litellm/llms/azure/audio_transcription/__init__.py b/litellm/llms/azure/audio_transcription/__init__.py new file mode 100644 index 00000000000..cedd0c6dbeb --- /dev/null +++ b/litellm/llms/azure/audio_transcription/__init__.py @@ -0,0 +1,3 @@ +from .transformation import AzureSpeechAudioTranscriptionConfig + +__all__ = ["AzureSpeechAudioTranscriptionConfig"] diff --git a/litellm/llms/azure/audio_transcription/transformation.py b/litellm/llms/azure/audio_transcription/transformation.py new file mode 100644 index 00000000000..e478c8ebf35 --- /dev/null +++ b/litellm/llms/azure/audio_transcription/transformation.py @@ -0,0 +1,224 @@ +""" +Azure AI Speech (Cognitive Services) speech-to-text transformation. + +Maps OpenAI-compatible audio transcription calls to Azure Speech REST +recognition for short audio. +""" + +from typing import Any, Dict, List, Optional, Union +from urllib.parse import urlencode, urlparse + +import httpx + +from litellm.litellm_core_utils.audio_utils.utils import process_audio_file +from litellm.llms.base_llm.audio_transcription.transformation import ( + AudioTranscriptionRequestData, + BaseAudioTranscriptionConfig, +) +from litellm.llms.base_llm.chat.transformation import BaseLLMException +from litellm.types.llms.openai import ( + AllMessageValues, + OpenAIAudioTranscriptionOptionalParams, +) +from litellm.secret_managers.main import get_secret_str +from litellm.types.utils import FileTypes, TranscriptionResponse + + +class AzureSpeechAudioTranscriptionException(BaseLLMException): + pass + + +class AzureSpeechAudioTranscriptionConfig(BaseAudioTranscriptionConfig): + """ + Configuration for Azure AI Speech (Cognitive Services) STT. + + Reference: + https://learn.microsoft.com/en-us/azure/ai-services/speech-service/rest-speech-to-text-short + """ + + COGNITIVE_SERVICES_DOMAIN = "api.cognitive.microsoft.com" + STT_SPEECH_DOMAIN = "stt.speech.microsoft.com" + STT_ENDPOINT_PATH = "/speech/recognition/conversation/cognitiveservices/v1" + DEFAULT_LANGUAGE = "en-US" + + def get_supported_openai_params( + self, model: str + ) -> List[OpenAIAudioTranscriptionOptionalParams]: + return ["language", "response_format"] + + def map_openai_params( + self, + non_default_params: dict, + optional_params: dict, + model: str, + drop_params: bool, + ) -> dict: + supported_params = self.get_supported_openai_params(model=model) + for key, value in non_default_params.items(): + if key in supported_params: + optional_params[key] = value + return optional_params + + def validate_environment( + self, + headers: dict, + model: str, + messages: List[AllMessageValues], + optional_params: dict, + litellm_params: dict, + api_key: Optional[str] = None, + api_base: Optional[str] = None, + ) -> dict: + api_key = api_key or get_secret_str("AZURE_SPEECH_API_KEY") + if not api_key: + raise AzureSpeechAudioTranscriptionException( + message="api_key is required for Azure AI Speech transcription.", + status_code=401, + ) + + validated_headers = headers.copy() + validated_headers["Ocp-Apim-Subscription-Key"] = api_key + validated_headers["Content-Type"] = validated_headers.get( + "Content-Type", "audio/wav" + ) + validated_headers["Accept"] = "application/json" + return validated_headers + + def get_complete_url( + self, + api_base: Optional[str], + api_key: Optional[str], + model: str, + optional_params: dict, + litellm_params: dict, + stream: Optional[bool] = None, + ) -> str: + api_base = api_base or get_secret_str("AZURE_SPEECH_API_BASE") + if api_base is None: + raise AzureSpeechAudioTranscriptionException( + message=( + "api_base is required for Azure AI Speech transcription. " + "Use a Cognitive Services endpoint like " + "https://{region}.api.cognitive.microsoft.com or an STT " + "endpoint like https://{region}.stt.speech.microsoft.com." + ), + status_code=400, + ) + + base_url = self._resolve_stt_base_url(api_base=api_base) + query_params = { + "language": optional_params.get("language", self.DEFAULT_LANGUAGE), + "format": self._get_azure_response_format( + optional_params.get("response_format") + ), + } + return f"{base_url}{self.STT_ENDPOINT_PATH}?{urlencode(query_params)}" + + def transform_audio_transcription_request( + self, + model: str, + audio_file: FileTypes, + optional_params: dict, + litellm_params: dict, + ) -> AudioTranscriptionRequestData: + processed_audio = process_audio_file(audio_file) + return AudioTranscriptionRequestData( + data=processed_audio.file_content, + files=None, + content_type=processed_audio.content_type, + ) + + def transform_audio_transcription_response( + self, + raw_response: httpx.Response, + ) -> TranscriptionResponse: + response_json = raw_response.json() + recognition_status = response_json.get("RecognitionStatus") + if recognition_status is not None and recognition_status != "Success": + raise AzureSpeechAudioTranscriptionException( + message=( + "Azure AI Speech transcription failed with " + f"RecognitionStatus={recognition_status}." + ), + status_code=raw_response.status_code, + headers=raw_response.headers, + ) + + text = self._extract_text(response_json) + response = TranscriptionResponse(text=text) + response._hidden_params = response_json + return response + + def get_error_class( + self, error_message: str, status_code: int, headers: Union[dict, httpx.Headers] + ) -> BaseLLMException: + return AzureSpeechAudioTranscriptionException( + message=error_message, + status_code=status_code, + headers=headers, + ) + + def _resolve_stt_base_url(self, api_base: str) -> str: + api_base = api_base.rstrip("/") + parsed_url = urlparse(api_base) + hostname = parsed_url.hostname or "" + + if self._is_cognitive_services_endpoint(hostname=hostname): + region = self._extract_region_from_hostname( + hostname=hostname, domain=self.COGNITIVE_SERVICES_DOMAIN + ) + return self._build_stt_base_url(region=region) + + if self._is_stt_endpoint(hostname=hostname): + return f"{parsed_url.scheme}://{hostname}" + + if self._is_azure_openai_endpoint(hostname=hostname): + raise AzureSpeechAudioTranscriptionException( + message=( + "Azure AI Speech transcription requires a Cognitive Services " + "or STT Speech endpoint, not an Azure OpenAI endpoint." + ), + status_code=400, + ) + + return api_base + + def _is_cognitive_services_endpoint(self, hostname: str) -> bool: + return hostname == self.COGNITIVE_SERVICES_DOMAIN or hostname.endswith( + f".{self.COGNITIVE_SERVICES_DOMAIN}" + ) + + def _is_stt_endpoint(self, hostname: str) -> bool: + return hostname == self.STT_SPEECH_DOMAIN or hostname.endswith( + f".{self.STT_SPEECH_DOMAIN}" + ) + + def _is_azure_openai_endpoint(self, hostname: str) -> bool: + return hostname.endswith(".openai.azure.com") + + def _extract_region_from_hostname(self, hostname: str, domain: str) -> str: + if hostname.endswith(f".{domain}"): + return hostname[: -len(f".{domain}")] + return "" + + def _build_stt_base_url(self, region: str) -> str: + if region: + return f"https://{region}.{self.STT_SPEECH_DOMAIN}" + return f"https://{self.STT_SPEECH_DOMAIN}" + + def _get_azure_response_format(self, response_format: Optional[str]) -> str: + if response_format == "verbose_json": + return "detailed" + return "simple" + + def _extract_text(self, response_json: Dict[str, Any]) -> str: + if isinstance(response_json.get("DisplayText"), str): + return response_json["DisplayText"] + + nbest = response_json.get("NBest") + if isinstance(nbest, list) and nbest: + best = nbest[0] + if isinstance(best, dict): + return best.get("Display") or best.get("Lexical") or "" + + return "" diff --git a/litellm/llms/azure/azure.py b/litellm/llms/azure/azure.py index 9291269d153..734b8ecef16 100644 --- a/litellm/llms/azure/azure.py +++ b/litellm/llms/azure/azure.py @@ -239,7 +239,9 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): ) data = {"model": None, "messages": messages, **optional_params} - elif litellm.AzureOpenAIGPT5Config.is_model_gpt_5_model(model=model): + elif litellm.AzureOpenAIGPT5Config.is_model_gpt_5_model( + model=litellm_params.get("base_model") or model + ): data = litellm.AzureOpenAIGPT5Config().transform_request( model=model, messages=messages, diff --git a/litellm/llms/azure/chat/o_series_transformation.py b/litellm/llms/azure/chat/o_series_transformation.py index cae7513245c..0a73597a4e4 100644 --- a/litellm/llms/azure/chat/o_series_transformation.py +++ b/litellm/llms/azure/chat/o_series_transformation.py @@ -4,10 +4,10 @@ Support for o1 and o3 model families https://platform.openai.com/docs/guides/reasoning Translations handled by LiteLLM: -- modalities: image => drop param (if user opts in to dropping param) -- role: system ==> translate to role 'user' -- streaming => faked by LiteLLM -- Tools, response_format => drop param (if user opts in to dropping param) +- modalities: image => drop param (if user opts in to dropping param) +- role: system ==> translate to role 'user' +- streaming => faked by LiteLLM +- Tools, response_format => drop param (if user opts in to dropping param) - Logprobs => drop param (if user opts in to dropping param) - Temperature => drop param (if user opts in to dropping param) """ diff --git a/litellm/llms/azure/common_utils.py b/litellm/llms/azure/common_utils.py index 4fc1ae960b8..e1ac1858912 100644 --- a/litellm/llms/azure/common_utils.py +++ b/litellm/llms/azure/common_utils.py @@ -1,3 +1,5 @@ +import asyncio +import hashlib import json import os from typing import Any, Callable, Dict, Literal, NamedTuple, Optional, Union, cast @@ -449,6 +451,25 @@ class BaseAzureLLM(BaseOpenAILLM): ] = None client_initialization_params: dict = locals() client_initialization_params["is_async"] = _is_async + _lp = litellm_params or {} + _ad_provider = _lp.get("azure_ad_token_provider") + _ad_token = _lp.get("azure_ad_token") + _client_secret = _lp.get("client_secret") + _azure_password = _lp.get("azure_password") + client_initialization_params["azure_ad_token"] = ( + hashlib.sha256(_ad_token.encode()).hexdigest() + if isinstance(_ad_token, str) + else None + ) + client_initialization_params["azure_ad_token_provider"] = ( + f"provider_id={id(_ad_provider) if callable(_ad_provider) else None}" + f"|tenant_id={_lp.get('tenant_id')}" + f"|client_id={_lp.get('client_id')}" + f"|client_secret={hashlib.sha256(_client_secret.encode()).hexdigest() if isinstance(_client_secret, str) else None}" + f"|azure_username={_lp.get('azure_username')}" + f"|azure_password={hashlib.sha256(_azure_password.encode()).hexdigest() if isinstance(_azure_password, str) else None}" + f"|azure_scope={_lp.get('azure_scope')}" + ) if client is None: cached_client = self.get_cached_openai_client( client_initialization_params=client_initialization_params, @@ -474,8 +495,29 @@ class BaseAzureLLM(BaseOpenAILLM): if self._is_azure_v1_api_version(api_version): # Extract only params that OpenAI client accepts # Always use /openai/v1/ regardless of whether user passed "v1", "latest", or "preview" - v1_params = { - "api_key": azure_client_params.get("api_key"), + # The OpenAI client accepts a callable for `api_key` and re-invokes it + # on every request (via `_refresh_api_key`), so passing + # `azure_ad_token_provider` directly preserves Azure AD token refresh + # behavior that the regular AzureOpenAI client provides. + v1_api_key: Optional[Union[str, Callable[[], Any]]] = ( + azure_client_params.get("api_key") + or azure_client_params.get("azure_ad_token_provider") + or azure_client_params.get("azure_ad_token") + ) + if _is_async is True and callable(v1_api_key): + # AsyncOpenAI expects an async provider; wrap the sync provider + # returned by azure-identity. Offload to a thread so a token + # refresh (blocking HTTP call to AAD on cache miss) does not + # stall the event loop. + _sync_provider = v1_api_key + + async def _async_v1_api_key() -> str: + return await asyncio.to_thread(_sync_provider) + + v1_api_key = _async_v1_api_key + + v1_params: Dict[str, Any] = { + "api_key": v1_api_key, "base_url": f"{api_base}/openai/v1/", } if "timeout" in azure_client_params: diff --git a/litellm/llms/azure/containers/transformation.py b/litellm/llms/azure/containers/transformation.py index 586b2e379a0..cd897511585 100644 --- a/litellm/llms/azure/containers/transformation.py +++ b/litellm/llms/azure/containers/transformation.py @@ -1,9 +1,16 @@ from typing import Optional +from urllib.parse import parse_qs, urlparse, urlunparse from litellm.llms.azure.common_utils import BaseAzureLLM from litellm.llms.openai.containers.transformation import OpenAIContainerConfig from litellm.types.router import GenericLiteLLMParams +# Endpoint-specific path suffixes that may appear in a deployment's api_base +# (e.g. the responses endpoint URL is stored as api_base for Azure models). +# Strip these before building the containers URL so we always start from the +# resource root (https://resource.cognitiveservices.azure.com). +_AZURE_ENDPOINT_PATHS = ("/openai/responses",) + class AzureContainerConfig(OpenAIContainerConfig): """ @@ -27,6 +34,27 @@ class AzureContainerConfig(OpenAIContainerConfig): litellm_params=GenericLiteLLMParams(api_key=api_key), ) + @staticmethod + def _normalize_api_base(api_base: Optional[str]) -> Optional[str]: + """Strip endpoint-specific path suffixes from api_base to get the resource root.""" + if not api_base: + return api_base + parsed = urlparse(api_base) + path = parsed.path.rstrip("/") + for ep in _AZURE_ENDPOINT_PATHS: + if path.endswith(ep): + return urlunparse( + (parsed.scheme, parsed.netloc, path[: -len(ep)], "", "", "") + ) + return api_base + + @staticmethod + def _extract_api_version(api_base: Optional[str]) -> Optional[str]: + """Return the api-version query param from api_base if present.""" + if not api_base: + return None + return parse_qs(urlparse(api_base).query).get("api-version", [None])[0] + def get_complete_url( self, api_base: Optional[str], @@ -39,10 +67,19 @@ class AzureContainerConfig(OpenAIContainerConfig): {endpoint}/openai/v1/containers when api_version is 'v1', 'latest', or 'preview'; otherwise: {endpoint}/openai/containers + + The deployment's api_base may be the responses endpoint URL + (e.g. .../openai/responses?api-version=2025-04-01-preview). We + prefer the api-version embedded there over the deployment's + api_version field, which may point to an older chat API version. """ + effective_params = dict(litellm_params) + api_version_from_base = self._extract_api_version(api_base) + if api_version_from_base: + effective_params["api_version"] = api_version_from_base return BaseAzureLLM._get_base_azure_url( - api_base=api_base, - litellm_params=litellm_params, + api_base=self._normalize_api_base(api_base), + litellm_params=effective_params, route="/openai/containers", default_api_version="v1", ) diff --git a/litellm/llms/azure/image_edit/transformation.py b/litellm/llms/azure/image_edit/transformation.py index 0b6ecfb0767..a450ee0b217 100644 --- a/litellm/llms/azure/image_edit/transformation.py +++ b/litellm/llms/azure/image_edit/transformation.py @@ -3,8 +3,10 @@ from typing import Optional, cast import httpx import litellm +from litellm.llms.azure.common_utils import BaseAzureLLM from litellm.llms.openai.image_edit.transformation import OpenAIImageEditConfig from litellm.secret_managers.main import get_secret_str +from litellm.types.router import GenericLiteLLMParams from litellm.utils import _add_path_to_api_base @@ -30,20 +32,42 @@ class AzureImageEditConfig(OpenAIImageEditConfig): litellm_params: Optional[dict] = None, api_base: Optional[str] = None, ) -> dict: - api_key = ( - api_key - or litellm.api_key - or litellm.azure_key - or get_secret_str("AZURE_OPENAI_API_KEY") - or get_secret_str("AZURE_API_KEY") - ) + """ + Validate Azure environment and set up authentication headers. - headers.update( - { - "Authorization": f"Bearer {api_key}", - } + Delegates to ``BaseAzureLLM._base_validate_azure_environment`` so the + Azure image-edit route uses the same auth resolution as every other + Azure provider (videos, vector_stores, responses, containers, ...): + + - prefers the Azure-style ``api-key`` header when an API key is available + - falls back to ``Authorization: Bearer `` only when AAD + auth is configured + + The previous implementation unconditionally set + ``Authorization: Bearer ``, which is correct for OpenAI direct + but not for Azure OpenAI / API Management gateways that expect the + ``api-key`` header. Subscription-key-based deployments (e.g., behind + Azure APIM) responded with ``401 "Access denied due to missing + subscription key"``. + + API-key precedence (matches ``AzureVideosConfig``): + + - ``litellm_params["api_key"]`` is the source of truth. + - The positional ``api_key`` kwarg only fills in when + ``litellm_params["api_key"]`` is empty. + - This is a deliberate change from the old ``or`` chain (where the + positional ``api_key`` argument won) so behavior matches every other + Azure ``validate_environment`` implementation. In production the only + caller (``llm_http_handler.image_edit``) sources both values from + the same ``litellm_params.api_key``, so the precedence only matters + for direct callers of this method. + """ + params = GenericLiteLLMParams(**(litellm_params or {})) + if api_key is not None and params.api_key is None: + params.api_key = api_key + return BaseAzureLLM._base_validate_azure_environment( + headers=headers, litellm_params=params ) - return headers def get_complete_url( self, diff --git a/litellm/llms/azure_ai/embed/cohere_transformation.py b/litellm/llms/azure_ai/embed/cohere_transformation.py index 64433c21b61..bbbfb60fbde 100644 --- a/litellm/llms/azure_ai/embed/cohere_transformation.py +++ b/litellm/llms/azure_ai/embed/cohere_transformation.py @@ -1,5 +1,5 @@ """ -Transformation logic from OpenAI /v1/embeddings format to Azure AI Cohere's /v1/embed. +Transformation logic from OpenAI /v1/embeddings format to Azure AI Cohere's /v1/embed. Why separate file? Make it easy to see how transformation works diff --git a/litellm/llms/azure_ai/rerank/transformation.py b/litellm/llms/azure_ai/rerank/transformation.py index b5993040ea0..f64133afa8b 100644 --- a/litellm/llms/azure_ai/rerank/transformation.py +++ b/litellm/llms/azure_ai/rerank/transformation.py @@ -1,5 +1,5 @@ """ -Translate between Cohere's `/rerank` format and Azure AI's `/rerank` format. +Translate between Cohere's `/rerank` format and Azure AI's `/rerank` format. """ from typing import Optional diff --git a/litellm/llms/base_llm/agents/__init__.py b/litellm/llms/base_llm/agents/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/litellm/llms/base_llm/agents/transformation.py b/litellm/llms/base_llm/agents/transformation.py new file mode 100644 index 00000000000..508e54cb7ab --- /dev/null +++ b/litellm/llms/base_llm/agents/transformation.py @@ -0,0 +1,165 @@ +""" +Base transformation class for provider-side Agents API. + +Providers that have a native agents CRUD API (e.g. Gemini v1beta/agents) +subclass BaseAgentsAPIConfig and implement the abstract methods. + +The HTTP calls are handled by AgentsHTTPHandler — this class is pure +transform logic (same separation as BaseInteractionsAPIConfig / +InteractionsHTTPHandler). +""" + +from abc import ABC, abstractmethod +from typing import Any, Dict, Optional, Tuple, Union + +import httpx + +from litellm.types.agents import ( + AgentCreateResponse, + AgentDeleteResult, + AgentListResponse, + AgentVersionsResponse, +) + + +class BaseAgentsAPIConfig(ABC): + """ + Minimal interface for providers that expose a native agents CRUD API. + """ + + # ------------------------------------------------------------------ # + # CREATE # + # ------------------------------------------------------------------ # + + @abstractmethod + def get_complete_url( + self, + api_base: Optional[str], + litellm_params: Dict[str, Any], + ) -> str: + """Return the full URL for POST /agents (create).""" + + @abstractmethod + def validate_environment( + self, + headers: Dict[str, str], + litellm_params: Dict[str, Any], + ) -> Dict[str, str]: + """Validate credentials and return auth headers.""" + + @abstractmethod + def transform_create_request( + self, + name: str, + litellm_params: Dict[str, Any], + ) -> Dict[str, Any]: + """Map name + litellm_params to the provider's create-agent body.""" + + @abstractmethod + def transform_create_response( + self, + raw_response: httpx.Response, + name: str, + ) -> AgentCreateResponse: + """Parse create response. Raise on non-2xx.""" + + # ------------------------------------------------------------------ # + # LIST # + # ------------------------------------------------------------------ # + + @abstractmethod + def transform_list_request( + self, + api_base: Optional[str], + litellm_params: Dict[str, Any], + ) -> Tuple[str, Dict[str, Any]]: + """Return (url, query_params) for GET /agents.""" + + @abstractmethod + def transform_list_response( + self, + raw_response: httpx.Response, + ) -> AgentListResponse: + """Parse list-agents response. Raise on non-2xx.""" + + # ------------------------------------------------------------------ # + # GET # + # ------------------------------------------------------------------ # + + @abstractmethod + def transform_get_request( + self, + name: str, + api_base: Optional[str], + litellm_params: Dict[str, Any], + ) -> Tuple[str, Dict[str, Any]]: + """Return (url, query_params) for GET /agents/{name}.""" + + @abstractmethod + def transform_get_response( + self, + raw_response: httpx.Response, + name: str, + ) -> AgentCreateResponse: + """Parse get-agent response. Raise on non-2xx.""" + + # ------------------------------------------------------------------ # + # DELETE # + # ------------------------------------------------------------------ # + + @abstractmethod + def transform_delete_request( + self, + name: str, + api_base: Optional[str], + litellm_params: Dict[str, Any], + ) -> str: + """Return the URL for DELETE /agents/{name}.""" + + @abstractmethod + def transform_delete_response( + self, + raw_response: httpx.Response, + name: str, + ) -> AgentDeleteResult: + """Parse delete-agent response. Raise on non-2xx.""" + + # ------------------------------------------------------------------ # + # LIST VERSIONS # + # ------------------------------------------------------------------ # + + @abstractmethod + def transform_list_versions_request( + self, + name: str, + api_base: Optional[str], + litellm_params: Dict[str, Any], + ) -> Tuple[str, Dict[str, Any]]: + """Return (url, query_params) for GET /agents/{name}/versions.""" + + @abstractmethod + def transform_list_versions_response( + self, + raw_response: httpx.Response, + name: str, + ) -> AgentVersionsResponse: + """Parse list-versions response. Raise on non-2xx.""" + + # ------------------------------------------------------------------ # + # ERROR HANDLING # + # ------------------------------------------------------------------ # + + def get_error_class( + self, + error_message: str, + status_code: int, + headers: Union[dict, httpx.Headers], + ) -> Exception: + """Map HTTP error status codes to provider-specific exceptions.""" + from litellm.llms.base_llm.chat.transformation import BaseLLMException + + return BaseLLMException( + status_code=status_code, + message=error_message, + headers=headers, + ) diff --git a/litellm/llms/base_llm/chat/transformation.py b/litellm/llms/base_llm/chat/transformation.py index bec25916c4b..5f35a58ce1f 100644 --- a/litellm/llms/base_llm/chat/transformation.py +++ b/litellm/llms/base_llm/chat/transformation.py @@ -108,10 +108,9 @@ class BaseConfig(ABC): return type_to_response_format_param(response_format=response_format) def is_thinking_enabled(self, non_default_params: dict) -> bool: - return ( - non_default_params.get("thinking", {}).get("type") == "enabled" - or non_default_params.get("reasoning_effort") is not None - ) + return (non_default_params.get("thinking") or {}).get( + "type" + ) == "enabled" or non_default_params.get("reasoning_effort") is not None def is_max_tokens_in_request(self, non_default_params: dict) -> bool: """ diff --git a/litellm/llms/base_llm/managed_resources/__init__.py b/litellm/llms/base_llm/managed_resources/__init__.py index 5eb9b46f89f..a5543e631c0 100644 --- a/litellm/llms/base_llm/managed_resources/__init__.py +++ b/litellm/llms/base_llm/managed_resources/__init__.py @@ -24,10 +24,12 @@ from .utils import ( generate_unified_id_string, is_base64_encoded_unified_id, parse_unified_id, + resolve_passthrough_managed_id_provider, ) __all__ = [ "BaseManagedResource", + "resolve_passthrough_managed_id_provider", "is_base64_encoded_unified_id", "extract_target_model_names_from_unified_id", "extract_resource_type_from_unified_id", diff --git a/litellm/llms/base_llm/managed_resources/utils.py b/litellm/llms/base_llm/managed_resources/utils.py index 59f5ff0d845..e9a6aef689e 100644 --- a/litellm/llms/base_llm/managed_resources/utils.py +++ b/litellm/llms/base_llm/managed_resources/utils.py @@ -7,7 +7,40 @@ different managed resource types (files, vector stores, etc.). import base64 import re -from typing import List, Optional, Union, Literal +from typing import Any, List, Literal, Optional, Union + +PASSTHROUGH_MANAGED_ID_AZURE_PROVIDERS = ("azure", "azure_ai") + + +def resolve_passthrough_managed_id_provider( + custom_llm_provider: Any, +) -> Optional[str]: + """Map a pass-through ``custom_llm_provider`` to the provider scope that + namespaces passthrough managed object IDs, or ``None`` when the route is not + an OpenAI/Azure pass-through and managed IDs must not apply. + + Scoping is keyed on the explicit provider that the pass-through route + forwards (``openai``, ``azure``, ``azure_ai``), not on the upstream URL, so + a third-party OpenAI-compatible endpoint never triggers managed-ID minting. + + ``azure`` and ``azure_ai`` deliberately collapse to one ``"azure"`` scope: + they expose the same Azure OpenAI files/batches surface, so an ID minted + while routing as one must still resolve while routing as the other. + Splitting them would make a managed ID minted on ``azure`` fail to resolve + when replayed on ``azure_ai`` and vice versa. + """ + provider = str( + getattr(custom_llm_provider, "value", custom_llm_provider) or "" + ).lower() + if not provider: + return None + if provider in PASSTHROUGH_MANAGED_ID_AZURE_PROVIDERS or provider.endswith( + (".azure", ".azure_ai") + ): + return "azure" + if provider == "openai" or provider.endswith(".openai"): + return "openai" + return None def is_base64_encoded_unified_id( @@ -177,8 +210,14 @@ def extract_model_id_from_unified_id( if decoded_id: unified_id = decoded_id - # Extract model ID - match = re.search(r"model_id,([^;]+)", unified_id) + # Extract model ID. Anchor to a field boundary (start of string or + # after `;`) so this regex doesn't substring-match the `model_id,` + # inside file_id encodings' `llm_output_file_model_id,` + # field — that would feed the deployment UUID as a model candidate + # into the team-access check and 403 every team-BYOK file attach + # with `Tried to access ` (LIT-3244 patch/1.86.0 second-order + # finding). + match = re.search(r"(?:^|;)model_id,([^;]+)", unified_id) if match: return match.group(1).strip() diff --git a/litellm/llms/base_llm/ocr/transformation.py b/litellm/llms/base_llm/ocr/transformation.py index b7f4d8e3b2d..263e0c094ce 100644 --- a/litellm/llms/base_llm/ocr/transformation.py +++ b/litellm/llms/base_llm/ocr/transformation.py @@ -54,6 +54,7 @@ class OCRUsageInfo(LiteLLMPydanticObjectBase): """Usage information from OCR response.""" pages_processed: Optional[int] = None + credits: Optional[float] = None doc_size_bytes: Optional[int] = None model_config = {"extra": "allow"} diff --git a/litellm/llms/base_llm/realtime/transformation.py b/litellm/llms/base_llm/realtime/transformation.py index d5531a532b9..0f239b4ad45 100644 --- a/litellm/llms/base_llm/realtime/transformation.py +++ b/litellm/llms/base_llm/realtime/transformation.py @@ -3,6 +3,7 @@ from typing import TYPE_CHECKING, Any, List, Optional, Union import httpx +from litellm.types.llms.openai import OpenAIRealtimeStreamSessionEvents from litellm.types.realtime import ( RealtimeResponseTransformInput, RealtimeResponseTypedDict, @@ -69,6 +70,20 @@ class BaseRealtimeConfig(ABC): ) -> Optional[str]: # message sent to setup the realtime session return None + def transform_session_created_event( + self, + model: str, + logging_session_id: str, + session_configuration_request: Optional[str] = None, + ) -> Optional[Union[dict, OpenAIRealtimeStreamSessionEvents]]: + """ + Optional hook for providers that defer session setup until client `session.update`. + + Return an OpenAI-compatible `session.created` payload when the proxy should + emit a synthetic event immediately after backend websocket connection. + """ + return None + @abstractmethod def transform_realtime_response( self, diff --git a/litellm/llms/base_llm/videos/transformation.py b/litellm/llms/base_llm/videos/transformation.py index 87289ad6a0c..9b4cf777280 100644 --- a/litellm/llms/base_llm/videos/transformation.py +++ b/litellm/llms/base_llm/videos/transformation.py @@ -321,6 +321,23 @@ class BaseVideoConfig(ABC): "video get character is not supported for this provider" ) + def get_video_edit_prefetch_params( + self, + video_id: str, + api_base: str, + litellm_params: GenericLiteLLMParams, + headers: dict, + ) -> Optional[Tuple[str, Dict]]: + """ + Return (url, body) for a pre-fetch HTTP call that must be made before + transform_video_edit_request, or None if no pre-fetch is required. + + Providers that need to retrieve the source video before constructing the + edit request (e.g. Vertex AI) should override this method. The handler + uses the existing shared httpx client so the call is properly async. + """ + return None + def transform_video_edit_request( self, prompt: str, @@ -329,6 +346,7 @@ class BaseVideoConfig(ABC): litellm_params: GenericLiteLLMParams, headers: dict, extra_body: Optional[Dict[str, Any]] = None, + prefetched_source_data: Optional[Dict[str, Any]] = None, ) -> Tuple[str, Dict]: """ Transform the video edit request into a URL and JSON data. @@ -343,6 +361,7 @@ class BaseVideoConfig(ABC): raw_response: httpx.Response, logging_obj: LiteLLMLoggingObj, custom_llm_provider: Optional[str] = None, + request_data: Optional[Dict] = None, ) -> VideoObject: raise NotImplementedError("video edit is not supported for this provider") diff --git a/litellm/llms/bedrock/base_aws_llm.py b/litellm/llms/bedrock/base_aws_llm.py index 9dd2b055a12..b659c1b0a0a 100644 --- a/litellm/llms/bedrock/base_aws_llm.py +++ b/litellm/llms/bedrock/base_aws_llm.py @@ -44,6 +44,12 @@ else: # (e.g. "us-east-1", "eu-west-2", "us-gov-west-1", "cn-north-1"). _VALID_AWS_REGION_PATTERN = re.compile(r"\A[a-z0-9-]+\Z") +# Regional STS hostnames, e.g. sts.eu-west-1.amazonaws.com or +# vpce-xxx.sts.eu-west-1.vpce.amazonaws.com +_STS_REGION_FROM_ENDPOINT_PATTERN = re.compile( + r"(?:^|\.)sts(?:-fips)?\.([a-z0-9-]+)\.(?:amazonaws\.com(?:\.cn)?|vpce\.amazonaws\.com)" +) + class Boto3CredentialsInfo(BaseModel): credentials: Credentials @@ -450,6 +456,24 @@ class BaseAWSLLM: model_id = BaseAWSLLM.encode_model_id(model_id=model_id) else: model_id = model + # Strip LiteLLM routing prefixes (e.g. "bedrock/", "invoke/", + # "bedrock/invoke/", "bedrock/converse/") that are not part of the + # actual Bedrock model ID. The converse path already does this; the + # invoke path must do the same so that ARN models such as + # bedrock/arn:aws:bedrock:…:inference-profile/global.anthropic.… + # are not forwarded verbatim to the Bedrock API, which would produce + # a malformed URL and cause botocore's EventStreamBuffer to receive + # a JSON error body instead of a binary event-stream — surfaced as a + # misleading ChecksumMismatch (0x223a7b22 == ':{"'). + # Use strip_bedrock_routing_prefix (no break) so compound prefixes + # like "bedrock/invoke/arn:..." are fully stripped in one call. + from litellm.llms.bedrock.common_utils import strip_bedrock_routing_prefix + + model_id = strip_bedrock_routing_prefix(model_id) + # URL-encode ARNs so colons and slashes are safe in the URL path. + if model_id.startswith("arn:"): + model_id = BaseAWSLLM.encode_model_id(model_id=model_id) + return model_id model_id = model_id.replace("invoke/", "", 1) if provider == "llama" and "llama/" in model_id: @@ -633,6 +657,40 @@ class BaseAWSLLM: "Region names must contain only lowercase letters, digits, and hyphens." ) + @staticmethod + def _parse_sts_region_from_endpoint( + aws_sts_endpoint: Optional[str], + ) -> Optional[str]: + """Extract region from sts.{region}.amazonaws.com or vpce-x.sts.{region}.vpce.amazonaws.com.""" + if not aws_sts_endpoint: + return None + host = urllib.parse.urlparse(aws_sts_endpoint).hostname or "" + match = _STS_REGION_FROM_ENDPOINT_PATTERN.search(host) + return match.group(1) if match else None + + @staticmethod + def _resolve_sts_region(aws_sts_endpoint: Optional[str] = None) -> Optional[str]: + """STS signing region: parsed from aws_sts_endpoint else AWS_REGION / AWS_DEFAULT_REGION.""" + return ( + BaseAWSLLM._parse_sts_region_from_endpoint(aws_sts_endpoint) + or os.getenv("AWS_REGION") + or os.getenv("AWS_DEFAULT_REGION") + ) + + def _build_sts_client_kwargs( + self, + aws_sts_endpoint: Optional[str] = None, + ssl_verify: Optional[Union[bool, str]] = None, + ) -> dict: + """STS client kwargs with aligned endpoint_url and region_name (SigV4).""" + kwargs: dict = {"verify": self._get_ssl_verify(ssl_verify)} + if aws_sts_endpoint is not None: + kwargs["endpoint_url"] = aws_sts_endpoint + sts_region = self._resolve_sts_region(aws_sts_endpoint) + if sts_region is not None: + kwargs["region_name"] = sts_region + return kwargs + def get_aws_region_name_for_non_llm_api_calls( self, aws_region_name: Optional[str] = None, @@ -787,11 +845,6 @@ class BaseAWSLLM: f"IN Web Identity Token: {aws_web_identity_token} | Role Name: {aws_role_name} | Session Name: {aws_session_name}" ) - if aws_sts_endpoint is None: - sts_endpoint = f"https://sts.{aws_region_name}.amazonaws.com" - else: - sts_endpoint = aws_sts_endpoint - oidc_token = get_secret(aws_web_identity_token) if oidc_token is None: @@ -800,13 +853,13 @@ class BaseAWSLLM: status_code=401, ) + sts_client_kwargs = self._build_sts_client_kwargs( + aws_sts_endpoint=aws_sts_endpoint, + ssl_verify=ssl_verify, + ) + with tracer.trace("boto3.client(sts)"): - sts_client = boto3.client( - "sts", - region_name=aws_region_name, - endpoint_url=sts_endpoint, - verify=self._get_ssl_verify(ssl_verify), - ) + sts_client = boto3.client("sts", **sts_client_kwargs) # https://docs.aws.amazon.com/STS/latest/APIReference/API_AssumeRoleWithWebIdentity.html # https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/sts/client/assume_role_with_web_identity.html @@ -847,7 +900,6 @@ class BaseAWSLLM: irsa_role_arn: str, aws_role_name: str, aws_session_name: str, - region: str, web_identity_token_file: str, aws_external_id: Optional[str] = None, aws_sts_endpoint: Optional[str] = None, @@ -862,12 +914,10 @@ class BaseAWSLLM: with open(web_identity_token_file, "r") as f: web_identity_token = f.read().strip() - irsa_sts_kwargs: dict = { - "region_name": region, - "verify": self._get_ssl_verify(ssl_verify), - } - if aws_sts_endpoint is not None: - irsa_sts_kwargs["endpoint_url"] = aws_sts_endpoint + irsa_sts_kwargs = self._build_sts_client_kwargs( + aws_sts_endpoint=aws_sts_endpoint, + ssl_verify=ssl_verify, + ) # Create an STS client without credentials with tracer.trace("boto3.client(sts) for manual IRSA"): @@ -924,7 +974,6 @@ class BaseAWSLLM: self, aws_role_name: str, aws_session_name: str, - region: str, aws_external_id: Optional[str] = None, aws_sts_endpoint: Optional[str] = None, ssl_verify: Optional[Union[bool, str]] = None, @@ -932,12 +981,10 @@ class BaseAWSLLM: """Handle same-account role assumption for IRSA.""" import boto3 - irsa_sts_kwargs: dict = { - "region_name": region, - "verify": self._get_ssl_verify(ssl_verify), - } - if aws_sts_endpoint is not None: - irsa_sts_kwargs["endpoint_url"] = aws_sts_endpoint + irsa_sts_kwargs = self._build_sts_client_kwargs( + aws_sts_endpoint=aws_sts_endpoint, + ssl_verify=ssl_verify, + ) verbose_logger.debug("Same account role assumption, using automatic IRSA") with tracer.trace("boto3.client(sts) with automatic IRSA"): @@ -1010,12 +1057,6 @@ class BaseAWSLLM: web_identity_token_file = os.getenv("AWS_WEB_IDENTITY_TOKEN_FILE") irsa_role_arn = os.getenv("AWS_ROLE_ARN") - region = ( - aws_region_name - or os.getenv("AWS_REGION") - or os.getenv("AWS_DEFAULT_REGION") - ) - # If we have IRSA environment variables and no explicit credentials, # we need to use the web identity token flow if ( @@ -1031,16 +1072,12 @@ class BaseAWSLLM: ) try: - # Use passed-in region when set, else env, else default (align with AssumeRole path) - region = region or "us-east-1" - # Check if we need to do cross-account role assumption if aws_role_name != irsa_role_arn: sts_response = self._handle_irsa_cross_account( irsa_role_arn, aws_role_name, aws_session_name, - region, web_identity_token_file, aws_external_id, aws_sts_endpoint=aws_sts_endpoint, @@ -1050,7 +1087,6 @@ class BaseAWSLLM: sts_response = self._handle_irsa_same_account( aws_role_name, aws_session_name, - region, aws_external_id, aws_sts_endpoint=aws_sts_endpoint, ssl_verify=ssl_verify, @@ -1074,11 +1110,10 @@ class BaseAWSLLM: # In EKS/IRSA environments, use ambient credentials (no explicit keys needed) # This allows the web identity token to work automatically - sts_client_kwargs: dict = {"verify": self._get_ssl_verify(ssl_verify)} - if region is not None: - sts_client_kwargs["region_name"] = region - if aws_sts_endpoint is not None: - sts_client_kwargs["endpoint_url"] = aws_sts_endpoint + sts_client_kwargs = self._build_sts_client_kwargs( + aws_sts_endpoint=aws_sts_endpoint, + ssl_verify=ssl_verify, + ) if aws_access_key_id is None and aws_secret_access_key is None: with tracer.trace("boto3.client(sts)"): sts_client = boto3.client("sts", **sts_client_kwargs) diff --git a/litellm/llms/bedrock/batches/transformation.py b/litellm/llms/bedrock/batches/transformation.py index 0602b1c2f62..620bc91732d 100644 --- a/litellm/llms/bedrock/batches/transformation.py +++ b/litellm/llms/bedrock/batches/transformation.py @@ -5,6 +5,7 @@ from typing import Any, Dict, List, Literal, Optional, Union, cast from httpx import Headers, Response +from litellm.litellm_core_utils.safe_json_dumps import safe_dumps from litellm.llms.base_llm.batches.transformation import BaseBatchesConfig from litellm.llms.base_llm.chat.transformation import BaseLLMException from litellm.secret_managers.main import get_secret_str @@ -263,9 +264,32 @@ class BedrockBatchesConfig(BaseAWSLLM, BaseBatchesConfig): cancelling_at=None, cancelled_at=None, request_counts=None, - metadata=original_request.get("metadata", {}), + metadata=self._get_openai_compatible_batch_metadata( + original_request.get("metadata", {}) + ), ) + @staticmethod + def _get_openai_compatible_batch_metadata(metadata: Any) -> Dict[str, str]: + """ + OpenAI Batch metadata only accepts string values. + """ + if not isinstance(metadata, dict): + return {} + + sanitized_metadata: Dict[str, str] = {} + for key, value in metadata.items(): + if key == "standard_logging_guardrail_information" or value is None: + continue + + str_key = str(key) + if isinstance(value, str): + sanitized_metadata[str_key] = value + else: + sanitized_metadata[str_key] = safe_dumps(value) + + return sanitized_metadata + def transform_retrieve_batch_request( self, batch_id: str, diff --git a/litellm/llms/bedrock/chat/converse_transformation.py b/litellm/llms/bedrock/chat/converse_transformation.py index efc890d9ee2..90dfa13e938 100644 --- a/litellm/llms/bedrock/chat/converse_transformation.py +++ b/litellm/llms/bedrock/chat/converse_transformation.py @@ -30,6 +30,7 @@ from litellm.litellm_core_utils.prompt_templates.factory import ( BedrockConverseMessagesProcessor, _bedrock_converse_messages_pt, _bedrock_tools_pt, + make_valid_bedrock_tool_name, ) from litellm.llms.anthropic.chat.transformation import ( DROP_UNSUPPORTED_OUTPUT_CONFIG_WARNING, @@ -40,6 +41,7 @@ from litellm.llms.base_llm.chat.transformation import BaseConfig, BaseLLMExcepti from litellm.types.llms.bedrock import * from litellm.types.llms.openai import ( AllMessageValues, + ChatCompletionAnnotation, ChatCompletionAssistantMessage, ChatCompletionRedactedThinkingBlock, ChatCompletionResponseMessage, @@ -77,6 +79,7 @@ from ..common_utils import ( get_anthropic_beta_from_headers, get_bedrock_tool_name, is_claude_4_5_on_bedrock, + normalize_bedrock_opus_output_config_effort, ) # Computer use tool prefixes supported by Bedrock @@ -447,10 +450,20 @@ class AmazonConverseConfig(BaseConfig): value=reasoning_effort, llm_provider="bedrock_converse", ) + existing_output_config = optional_params.get("output_config") + if not isinstance(existing_output_config, dict): + existing_output_config = {} + existing_output_config.setdefault("effort", mapped_effort) + normalize_bedrock_opus_output_config_effort( + model=model, + output_config=existing_output_config, + ) + mapped_effort = existing_output_config["effort"] self._validate_anthropic_adaptive_effort( model=model, effort=mapped_effort ) - optional_params["output_config"] = {"effort": mapped_effort} + optional_params["output_config"] = existing_output_config + optional_params["_output_config_normalized"] = True @staticmethod def _validate_anthropic_adaptive_effort(model: str, effort: str) -> None: @@ -573,6 +586,9 @@ class AmazonConverseConfig(BaseConfig): ): supported_params.append("thinking") supported_params.append("reasoning_effort") + + if base_model.startswith("anthropic"): + supported_params.append("context_management") return supported_params def map_tool_choice_values( @@ -595,7 +611,9 @@ class AmazonConverseConfig(BaseConfig): elif isinstance(tool_choice, dict): # only supported for anthropic + mistral models - https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ToolChoice.html specific_tool = SpecificToolChoiceBlock( - name=tool_choice.get("function", {}).get("name", "") + name=make_valid_bedrock_tool_name( + tool_choice.get("function", {}).get("name", "") + ) ) return ToolChoiceValuesBlock(tool=specific_tool) else: @@ -932,10 +950,10 @@ class AmazonConverseConfig(BaseConfig): self._handle_reasoning_effort_parameter( model=model, reasoning_effort=value, optional_params=optional_params ) + elif param == "context_management" and isinstance(value, (dict, list)): + self._map_context_management_param(value, optional_params) if param == "requestMetadata": - if value is not None and isinstance(value, dict): - self._validate_request_metadata(value) # type: ignore - optional_params["requestMetadata"] = value + self._map_request_metadata_param(value, optional_params) if param == "service_tier" and isinstance(value, str): self._map_service_tier_param(value, optional_params) @@ -968,6 +986,32 @@ class AmazonConverseConfig(BaseConfig): return optional_params + def _map_request_metadata_param(self, value: Any, optional_params: dict) -> None: + if value is not None and isinstance(value, dict): + self._validate_request_metadata(value) # type: ignore + optional_params["requestMetadata"] = value + + def _map_context_management_param( + self, value: Union[dict, list], optional_params: dict + ) -> None: + # Match the dispatcher's ``_normalize_spec`` behavior: only run the + # OpenAI→Anthropic mapper for list inputs. Dict inputs are already in + # Anthropic-native shape (``{"edits": [...]}``) and should pass + # through unchanged so an Anthropic-format ``context_management`` + # value isn't silently dropped when the mapper can't classify it. + if isinstance(value, list): + mapped = AnthropicConfig.map_openai_context_management_to_anthropic( + cast(Union[dict, list], value) + ) + else: + mapped = value + # Skip when the mapper returned None for malformed input — leaving the + # key out is safer than passing `context_management: null` downstream, + # which Bedrock would reject and which can confuse intermediate checks + # before the final _filter_context_management_for_bedrock_converse step. + if mapped is not None: + optional_params["context_management"] = mapped + def _map_service_tier_param(self, value: str, optional_params: dict) -> None: """Map OpenAI service_tier (string) to Bedrock serviceTier (object). @@ -1198,6 +1242,12 @@ class AmazonConverseConfig(BaseConfig): self, optional_params: dict, model: str ) -> Tuple[dict, dict, dict, Optional[OutputConfigBlock]]: """Prepare and separate request parameters.""" + # Consume the internal ``_output_config_normalized`` marker set by + # ``_handle_reasoning_effort_parameter`` so it does not linger on the + # caller's ``optional_params`` after the transformation returns. + anthropic_output_config_already_normalized = bool( + optional_params.pop("_output_config_normalized", False) + ) # Filter out exception objects before deepcopy to prevent deepcopy failures # Exceptions should not be stored in optional_params (this is a defensive fix) cleaned_params = filter_exceptions_from_params(optional_params) @@ -1216,8 +1266,17 @@ class AmazonConverseConfig(BaseConfig): # Anthropic-only ``output_config`` (snake_case) — re-attached to # ``additionalModelRequestFields`` for Anthropic models below. The - # Bedrock-native ``outputConfig`` (camelCase) is handled separately. + # structured-output ``format`` subfield is consumed into Bedrock's + # native ``outputConfig`` (camelCase), which is handled separately. anthropic_output_config = inference_params.pop("output_config", None) + output_config_format = None + if isinstance(anthropic_output_config, dict): + anthropic_output_config = dict(anthropic_output_config) + candidate_output_config_format = anthropic_output_config.pop("format", None) + if isinstance(candidate_output_config_format, dict): + output_config_format = candidate_output_config_format + if not anthropic_output_config: + anthropic_output_config = None # Extract requestMetadata before processing other parameters request_metadata = inference_params.pop("requestMetadata", None) @@ -1227,6 +1286,30 @@ class AmazonConverseConfig(BaseConfig): output_config: Optional[OutputConfigBlock] = inference_params.pop( "outputConfig", None ) + base_model = BedrockModelInfo.get_base_model(model) + if ( + output_config is None + and output_config_format is not None + and output_config_format.get("type") == "json_schema" + and base_model.startswith("anthropic") + and self._supports_native_structured_outputs( + model, self.custom_llm_provider + ) + ): + output_config = self._create_output_config_for_response_format( + json_schema=output_config_format.get("schema"), + name=output_config_format.get("name"), + description=output_config_format.get("description"), + ) + elif output_config is None and output_config_format is not None: + litellm.verbose_logger.warning( + "Bedrock Converse: dropping `output_config.format` for model=%s — " + "model does not advertise `supports_native_structured_output` in " + "model_prices_and_context_window.json. The schema will not be " + "enforced; pass `response_format` to use the synthetic tool-call " + "fallback.", + model, + ) # keep supported params in 'inference_params', and set all model-specific params in 'additional_request_params' additional_request_params = { @@ -1272,7 +1355,6 @@ class AmazonConverseConfig(BaseConfig): if anthropic_output_config is not None and isinstance( anthropic_output_config, dict ): - base_model = BedrockModelInfo.get_base_model(model) if base_model.startswith("anthropic"): if ( litellm.drop_params is True @@ -1283,6 +1365,11 @@ class AmazonConverseConfig(BaseConfig): model, ) else: + if not anthropic_output_config_already_normalized: + normalize_bedrock_opus_output_config_effort( + model=model, + output_config=anthropic_output_config, + ) effort = anthropic_output_config.get("effort") if effort is not None: self._validate_anthropic_adaptive_effort( @@ -1430,6 +1517,11 @@ class AmazonConverseConfig(BaseConfig): if ANTHROPIC_EFFORT_BETA_HEADER not in anthropic_beta_list: anthropic_beta_list.append(ANTHROPIC_EFFORT_BETA_HEADER) + # Bedrock Converse: compact_20260112 edits only (+ beta header). + AmazonConverseConfig._filter_context_management_for_bedrock_converse( + additional_request_params, anthropic_beta_list + ) + # Set anthropic_beta in additional_request_params if we have any beta features # ONLY apply to Anthropic/Claude models - other models (e.g., Qwen, Llama) don't support this field if anthropic_beta_list and base_model.startswith("anthropic"): @@ -1437,6 +1529,42 @@ class AmazonConverseConfig(BaseConfig): return bedrock_tools, anthropic_beta_list + @staticmethod + def _filter_context_management_for_bedrock_converse( + additional_request_params: dict, + anthropic_beta_list: list, + ) -> None: + """Keep only compact_20260112 edits for Bedrock; add beta header or drop field.""" + from litellm.llms.anthropic.experimental_pass_through.context_management.constants import ( + COMPACT_EDIT_TYPE, + ) + from litellm.types.llms.anthropic import ANTHROPIC_BETA_HEADER_VALUES + + cm = additional_request_params.get("context_management") + if not isinstance(cm, dict): + additional_request_params.pop("context_management", None) + return + edits = cm.get("edits") + if not isinstance(edits, list): + additional_request_params.pop("context_management", None) + return + + compact_edits = [ + e + for e in edits + if isinstance(e, dict) and e.get("type") == COMPACT_EDIT_TYPE + ] + if compact_edits: + compact_beta = ANTHROPIC_BETA_HEADER_VALUES.COMPACT_2026_01_12.value + if compact_beta not in anthropic_beta_list: + anthropic_beta_list.append(compact_beta) + additional_request_params["context_management"] = { + **cm, + "edits": compact_edits, + } + else: + additional_request_params.pop("context_management", None) + def _transform_request_helper( self, model: str, @@ -1887,6 +2015,75 @@ class AmazonConverseConfig(BaseConfig): return content_str, tools, reasoningContentBlocks, citationsContentBlocks + @staticmethod + def _transform_citations_to_annotations( + citations_content_blocks: Optional[List[CitationsContentBlock]], + ) -> Tuple[Optional[str], Optional[List[ChatCompletionAnnotation]]]: + """ + Convert Bedrock citationsContent blocks into OpenAI-style annotations. + + Returns: + citations_text: concatenated text from citationsContent.content + annotations: OpenAI URL citation annotations + """ + if not citations_content_blocks: + return None, None + + annotations: List[ChatCompletionAnnotation] = [] + citations_text_parts: List[str] = [] + content_offset = 0 + + for citations_block in citations_content_blocks: + block_text = "" + raw_content = citations_block.get("content") + if isinstance(raw_content, list): + for content_part in raw_content: + if isinstance(content_part, dict): + _text = content_part.get("text") + if isinstance(_text, str): + block_text += _text + + block_offset = content_offset + if block_text: + citations_text_parts.append(block_text) + content_offset += len(block_text) + + raw_citations = citations_block.get("citations") + if not isinstance(raw_citations, list): + continue + + for citation in raw_citations: + if not isinstance(citation, dict): + continue + + location = citation.get("location") + if not isinstance(location, dict): + continue + + search_location = location.get("searchResultLocation") + if not isinstance(search_location, dict): + continue + + start = search_location.get("start") + end = search_location.get("end") + if not isinstance(start, int) or not isinstance(end, int): + continue + + annotations.append( + ChatCompletionAnnotation( + type="url_citation", + url_citation={ + "start_index": block_offset + start, + "end_index": block_offset + end, + "title": str(citation.get("title") or ""), + "url": str(citation.get("source") or ""), + }, + ) + ) + + citations_text = "".join(citations_text_parts) if citations_text_parts else None + return citations_text, annotations or None + @staticmethod def _unwrap_bedrock_properties(json_str: str) -> str: """ @@ -2069,6 +2266,24 @@ class AmazonConverseConfig(BaseConfig): provider_specific_fields ) + citations_text, annotations = self._transform_citations_to_annotations( + citationsContentBlocks + ) + citations_included_in_content = False + if citations_text: + stripped_content = content_str.strip() + if not stripped_content: + content_str = citations_text + citations_included_in_content = True + elif not any(char.isalnum() for char in stripped_content): + # Bedrock may emit the cited sentence in citationsContent and only + # punctuation in the text blocks; stitch citations_text in front so + # its annotation span indices stay aligned with the final content. + content_str = citations_text + content_str + citations_included_in_content = True + if annotations and citations_included_in_content: + chat_completion_message["annotations"] = annotations + if reasoningContentBlocks is not None: chat_completion_message["reasoning_content"] = ( self._transform_reasoning_content(reasoningContentBlocks) diff --git a/litellm/llms/bedrock/chat/invoke_agent/transformation.py b/litellm/llms/bedrock/chat/invoke_agent/transformation.py index e4072c24557..c88fa32b6a0 100644 --- a/litellm/llms/bedrock/chat/invoke_agent/transformation.py +++ b/litellm/llms/bedrock/chat/invoke_agent/transformation.py @@ -299,9 +299,9 @@ class AmazonInvokeAgentConfig(BaseConfig, BaseAWSLLM): ) def _get_response_stream_shape(self): - from litellm.llms.bedrock.common_utils import BEDROCK_RESPONSE_STREAM_SHAPE + from litellm.llms.bedrock.common_utils import get_bedrock_response_stream_shape - return BEDROCK_RESPONSE_STREAM_SHAPE + return get_bedrock_response_stream_shape() def _extract_response_content(self, events: InvokeAgentEventList) -> str: """Extract the final response content from parsed events.""" diff --git a/litellm/llms/bedrock/chat/invoke_handler.py b/litellm/llms/bedrock/chat/invoke_handler.py index 92ca75db95b..7a9916f1f31 100644 --- a/litellm/llms/bedrock/chat/invoke_handler.py +++ b/litellm/llms/bedrock/chat/invoke_handler.py @@ -68,9 +68,9 @@ from litellm.utils import CustomStreamWrapper, get_secret from ..base_aws_llm import BaseAWSLLM from ..common_utils import ( - BEDROCK_RESPONSE_STREAM_SHAPE, BedrockError, ModelResponseIterator, + get_bedrock_response_stream_shape, get_bedrock_tool_name, ) @@ -1828,7 +1828,8 @@ class AWSEventStreamDecoder: yield self._chunk_parser(chunk_data=_data) def _parse_message_from_event(self, event) -> Optional[str]: - if BEDROCK_RESPONSE_STREAM_SHAPE is None: + response_stream_shape = get_bedrock_response_stream_shape() + if response_stream_shape is None: raise BedrockError( status_code=500, message=( @@ -1837,9 +1838,7 @@ class AWSEventStreamDecoder: ), ) response_dict = event.to_response_dict() - parsed_response = self.parser.parse( - response_dict, BEDROCK_RESPONSE_STREAM_SHAPE - ) + parsed_response = self.parser.parse(response_dict, response_stream_shape) if response_dict["status_code"] != 200: decoded_body = response_dict["body"].decode() diff --git a/litellm/llms/bedrock/chat/invoke_transformations/anthropic_claude3_transformation.py b/litellm/llms/bedrock/chat/invoke_transformations/anthropic_claude3_transformation.py index c883ab68dff..a13336b6c88 100644 --- a/litellm/llms/bedrock/chat/invoke_transformations/anthropic_claude3_transformation.py +++ b/litellm/llms/bedrock/chat/invoke_transformations/anthropic_claude3_transformation.py @@ -3,6 +3,7 @@ from typing import TYPE_CHECKING, Any, List, Optional import httpx from litellm.anthropic_beta_headers_manager import filter_and_transform_beta_headers +from litellm.litellm_core_utils.litellm_logging import verbose_logger from litellm.litellm_core_utils.prompt_templates.factory import ( convert_to_anthropic_image_obj, ) @@ -15,13 +16,17 @@ from litellm.llms.bedrock.chat.invoke_transformations.base_invoke_transformation AmazonInvokeConfig, ) from litellm.llms.bedrock.common_utils import ( + convert_bedrock_invoke_output_format_to_inline_schema, get_anthropic_beta_from_headers, + normalize_bedrock_opus_output_config_effort, normalize_tool_input_schema_types_for_bedrock_invoke, + pop_bedrock_invoke_output_config_format, remove_custom_field_from_tools, ) from litellm.types.llms.anthropic import ANTHROPIC_TOOL_SEARCH_BETA_HEADER from litellm.types.llms.openai import AllMessageValues from litellm.types.utils import ModelResponse +from litellm.utils import _supports_factory if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj @@ -73,6 +78,17 @@ class AmazonAnthropicClaudeConfig(AmazonInvokeConfig, AnthropicConfig): # Use a model name that forces tool-based approach model = "claude-3-sonnet-20240229" + # Clamp ``reasoning_effort`` to the Bedrock effort ceiling before the + # parent mapping converts it to ``output_config.effort`` and the + # downstream effort gate runs. Mirrors the converse path's + # ``_handle_reasoning_effort_parameter`` and the messages path's + # ``_clamp_adaptive_reasoning_effort_for_bedrock`` so adaptive Claude + # requests degrade ``xhigh`` -> ``max`` rather than 400-ing on + # models like Opus 4.6 that don't natively advertise xhigh. + self._clamp_adaptive_reasoning_effort_for_bedrock( + model=original_model, params=non_default_params + ) + optional_params = AnthropicConfig.map_openai_params( self, non_default_params, @@ -86,6 +102,27 @@ class AmazonAnthropicClaudeConfig(AmazonInvokeConfig, AnthropicConfig): return optional_params + @staticmethod + def _clamp_adaptive_reasoning_effort_for_bedrock(model: str, params: dict) -> None: + """Lower ``reasoning_effort`` to the Bedrock effort ceiling before mapping. + + Bedrock's adaptive Claude models accept the OpenAI-style + ``reasoning_effort`` tier, but the request validator can reject tiers + the model does not natively advertise (e.g. ``xhigh`` on Opus 4.6). + Clamp the raw tier to the model's + ``bedrock_output_config_effort_ceiling`` so Claude Code "goal mode" + keeps working. Non-adaptive models and models without a ceiling are + left untouched. + """ + if not AnthropicConfig._is_adaptive_thinking_model(model): + return + effort = params.get("reasoning_effort") + if not isinstance(effort, str): + return + clamped = {"effort": effort} + normalize_bedrock_opus_output_config_effort(model=model, output_config=clamped) + params["reasoning_effort"] = clamped["effort"] + def transform_request( self, model: str, @@ -155,6 +192,13 @@ class AmazonAnthropicClaudeConfig(AmazonInvokeConfig, AnthropicConfig): for k, v in optional_params.items() if k not in self.aws_authentication_params } + output_config = filtered_params.get("output_config") + if isinstance(output_config, dict): + filtered_params["output_config"] = dict(output_config) + normalize_bedrock_opus_output_config_effort( + model=model, + output_config=filtered_params["output_config"], + ) filtered_params = self._normalize_bedrock_tool_search_tools(filtered_params) anthropic_request = AnthropicConfig.transform_request( @@ -168,7 +212,38 @@ class AmazonAnthropicClaudeConfig(AmazonInvokeConfig, AnthropicConfig): anthropic_request.pop("model", None) anthropic_request.pop("stream", None) - anthropic_request.pop("output_format", None) + output_format = anthropic_request.pop("output_format", None) + output_config_format = pop_bedrock_invoke_output_config_format( + anthropic_request + ) + if output_format: + convert_bedrock_invoke_output_format_to_inline_schema( + output_format=output_format, + request_body=anthropic_request, + ) + elif output_config_format: + convert_bedrock_invoke_output_format_to_inline_schema( + output_format=output_config_format, + request_body=anthropic_request, + ) + if not ( + _supports_factory( + model=model, + custom_llm_provider="bedrock", + key="supports_output_config", + ) + or AnthropicConfig._model_supports_effort_param(model) + ): + if anthropic_request.pop("output_config", None) is not None: + verbose_logger.warning( + "Bedrock Invoke: stripping unsupported `output_config` for " + "model=%s — neither `supports_output_config` nor any " + "`supports_*_reasoning_effort` flag is set in " + "model_prices_and_context_window.json. Add the capability " + "flag to the model JSON entry if this model accepts " + "`output_config`.", + model, + ) if "anthropic_version" not in anthropic_request: anthropic_request["anthropic_version"] = self.anthropic_version diff --git a/litellm/llms/bedrock/chat/mantle/transformation.py b/litellm/llms/bedrock/chat/mantle/transformation.py index b9bea77c118..ef0199031af 100644 --- a/litellm/llms/bedrock/chat/mantle/transformation.py +++ b/litellm/llms/bedrock/chat/mantle/transformation.py @@ -21,7 +21,9 @@ if TYPE_CHECKING: else: LiteLLMLoggingObj = Any -MANTLE_ENDPOINT_TEMPLATE = "https://bedrock-mantle.{region}.api.aws/v1/messages" +MANTLE_ENDPOINT_TEMPLATE = ( + "https://bedrock-mantle.{region}.api.aws/anthropic/v1/messages" +) class AmazonMantleConfig(AmazonAnthropicClaudeConfig): diff --git a/litellm/llms/bedrock/claude_platform/common_utils.py b/litellm/llms/bedrock/claude_platform/common_utils.py index 121221518c8..3abb8710de7 100644 --- a/litellm/llms/bedrock/claude_platform/common_utils.py +++ b/litellm/llms/bedrock/claude_platform/common_utils.py @@ -4,7 +4,6 @@ import litellm from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM from litellm.secret_managers.main import get_secret_str - CLAUDE_PLATFORM_SERVICE_NAME: Literal["aws-external-anthropic"] = ( "aws-external-anthropic" ) diff --git a/litellm/llms/bedrock/common_utils.py b/litellm/llms/bedrock/common_utils.py index 0256d5d4b95..bdc5da321c6 100644 --- a/litellm/llms/bedrock/common_utils.py +++ b/litellm/llms/bedrock/common_utils.py @@ -4,6 +4,7 @@ from __future__ import annotations Common utilities used across bedrock chat/embedding/image generation """ +import functools import json import os from typing import TYPE_CHECKING, Any, Dict, List, Literal, Optional, Union @@ -33,6 +34,15 @@ class BedrockError(BaseLLMException): # Lazy import cache to avoid circular imports and performance impact _get_model_info = None +BedrockOutputConfigEffort = Literal["low", "medium", "high", "max", "xhigh"] +_BEDROCK_OUTPUT_CONFIG_EFFORT_ORDER: Dict[BedrockOutputConfigEffort, int] = { + "low": 0, + "medium": 1, + "high": 2, + "max": 3, + "xhigh": 4, +} + def get_cached_model_info(): """ @@ -50,6 +60,79 @@ def get_cached_model_info(): return _get_model_info +@functools.lru_cache(maxsize=1) +def _get_local_model_cost_map() -> Dict: + from litellm.litellm_core_utils.get_model_cost_map import GetModelCostMap + + return GetModelCostMap.load_local_model_cost_map() + + +def pop_bedrock_invoke_output_config_format(request_body: Dict) -> Optional[Dict]: + """ + Remove and return Anthropic's nested ``output_config.format`` field. + + Bedrock Invoke paths convert the schema to inline message text. Any remaining + ``output_config`` keys, such as ``effort``, are left in place. + """ + output_config = request_body.get("output_config") + if not isinstance(output_config, dict): + return None + + output_format = output_config.pop("format", None) + if not output_config: + request_body.pop("output_config", None) + + if isinstance(output_format, dict): + return output_format + return None + + +def convert_bedrock_invoke_output_format_to_inline_schema( + output_format: Dict, + request_body: Dict, +) -> None: + """ + Embed an Anthropic structured-output schema into the last user message. + + Bedrock Invoke does not support ``output_format`` directly, so the schema is + appended to the final user message for prompt-engineered structured output. + The caller's ``messages`` list, message dict, and content list are not + mutated; a fresh ``messages`` list with a copied final user message is + written back to ``request_body``. + """ + schema = output_format.get("schema") + if not schema: + return + + messages = request_body.get("messages") + if not isinstance(messages, list) or not messages: + return + + last_user_idx = None + for i in range(len(messages) - 1, -1, -1): + message = messages[i] + if isinstance(message, dict) and message.get("role") == "user": + last_user_idx = i + break + + if last_user_idx is None: + return + + original = messages[last_user_idx] + content = original.get("content", []) + schema_block = {"type": "text", "text": json.dumps(schema)} + if isinstance(content, str): + new_content = [{"type": "text", "text": content}, schema_block] + elif isinstance(content, list): + new_content = [*content, schema_block] + else: + return + + new_messages = list(messages) + new_messages[last_user_idx] = {**original, "content": new_content} + request_body["messages"] = new_messages + + def remove_custom_field_from_tools(request_body: dict) -> None: """ Remove ``custom`` field from each tool in the request body. @@ -602,6 +685,62 @@ def is_claude_4_5_on_bedrock(model: str) -> bool: return any(pattern in model_lower for pattern in claude_4_5_patterns) +def normalize_bedrock_opus_output_config_effort(model: str, output_config: Any) -> None: + """ + Normalize Anthropic ``output_config.effort`` values for Bedrock Opus ids. + + Bedrock's Claude Opus request validator can accept a narrower effort + vocabulary than Anthropic's compatibility surface. The Bedrock ceiling is + read from ``model_prices_and_context_window.json`` via + ``bedrock_output_config_effort_ceiling``. + + Mutates ``output_config`` in place so callers can accept Claude Code's + ``xhigh`` input without forwarding a provider-invalid value. + """ + if not isinstance(output_config, dict): + return + + effort = output_config.get("effort") + if effort not in _BEDROCK_OUTPUT_CONFIG_EFFORT_ORDER: + return + + ceiling = _get_bedrock_output_config_effort_ceiling(model) + if ceiling is None: + return + + if ( + _BEDROCK_OUTPUT_CONFIG_EFFORT_ORDER[effort] + > _BEDROCK_OUTPUT_CONFIG_EFFORT_ORDER[ceiling] + ): + output_config["effort"] = ceiling + + +def _get_bedrock_output_config_effort_ceiling( + model: str, +) -> Optional[BedrockOutputConfigEffort]: + try: + model_info = get_cached_model_info()( + model=model, + custom_llm_provider="bedrock", + ) + except Exception: + return None + + ceiling = model_info.get("bedrock_output_config_effort_ceiling") + if isinstance(ceiling, str) and ceiling in _BEDROCK_OUTPUT_CONFIG_EFFORT_ORDER: + return ceiling # type: ignore[return-value] + + model_cost_key = model_info.get("key") + if not isinstance(model_cost_key, str): + return None + + local_model_info = _get_local_model_cost_map().get(model_cost_key, {}) + ceiling = local_model_info.get("bedrock_output_config_effort_ceiling") + if isinstance(ceiling, str) and ceiling in _BEDROCK_OUTPUT_CONFIG_EFFORT_ORDER: + return ceiling # type: ignore[return-value] + return None + + # Import after standalone functions to avoid circular imports from litellm.llms.bedrock.count_tokens.bedrock_token_counter import BedrockTokenCounter @@ -963,10 +1102,8 @@ def _load_bedrock_response_stream_shape(): """ Load the ResponseStream shape from botocore's bundled bedrock-runtime schema. - Called once at module import time; the result is stored in - ``BEDROCK_RESPONSE_STREAM_SHAPE`` and reused for the process lifetime. Returns ``None`` if botocore is unavailable or the service model cannot be - loaded, so the module still imports cleanly. + loaded. """ try: from botocore.loaders import Loader @@ -977,15 +1114,22 @@ def _load_bedrock_response_stream_shape(): return ServiceModel(service_dict).shape_for("ResponseStream") except Exception as e: verbose_logger.warning( - "litellm: could not pre-load bedrock-runtime response stream shape " + "litellm: could not load bedrock-runtime response stream shape " "— Bedrock event-stream decoding will be unavailable. Error: %s", e, ) return None -# Eagerly resolved once per process — avoids per-instance or per-request disk I/O. -BEDROCK_RESPONSE_STREAM_SHAPE = _load_bedrock_response_stream_shape() +@functools.lru_cache(maxsize=1) +def get_bedrock_response_stream_shape(): + """ + Lazily load and cache the bedrock-runtime ResponseStream shape for the process. + + Avoids importing botocore (and logging warnings) unless Bedrock event-stream + decoding is actually needed. + """ + return _load_bedrock_response_stream_shape() class BedrockEventStreamDecoderBase: @@ -999,7 +1143,8 @@ class BedrockEventStreamDecoderBase: self.parser = EventStreamJSONParser() def _parse_message_from_event(self, event) -> Optional[str]: - if BEDROCK_RESPONSE_STREAM_SHAPE is None: + response_stream_shape = get_bedrock_response_stream_shape() + if response_stream_shape is None: raise BedrockError( status_code=500, message=( @@ -1008,9 +1153,7 @@ class BedrockEventStreamDecoderBase: ), ) response_dict = event.to_response_dict() - parsed_response = self.parser.parse( - response_dict, BEDROCK_RESPONSE_STREAM_SHAPE - ) + parsed_response = self.parser.parse(response_dict, response_stream_shape) if response_dict["status_code"] != 200: decoded_body = response_dict["body"].decode() diff --git a/litellm/llms/bedrock/embed/amazon_titan_g1_transformation.py b/litellm/llms/bedrock/embed/amazon_titan_g1_transformation.py index 2747551af81..64a79b73273 100644 --- a/litellm/llms/bedrock/embed/amazon_titan_g1_transformation.py +++ b/litellm/llms/bedrock/embed/amazon_titan_g1_transformation.py @@ -1,5 +1,5 @@ """ -Transformation logic from OpenAI /v1/embeddings format to Bedrock Amazon Titan G1 /invoke format. +Transformation logic from OpenAI /v1/embeddings format to Bedrock Amazon Titan G1 /invoke format. Why separate file? Make it easy to see how transformation works diff --git a/litellm/llms/bedrock/embed/cohere_transformation.py b/litellm/llms/bedrock/embed/cohere_transformation.py index d00cb74aae0..9570ff1a14c 100644 --- a/litellm/llms/bedrock/embed/cohere_transformation.py +++ b/litellm/llms/bedrock/embed/cohere_transformation.py @@ -1,5 +1,5 @@ """ -Transformation logic from OpenAI /v1/embeddings format to Bedrock Cohere /invoke format. +Transformation logic from OpenAI /v1/embeddings format to Bedrock Cohere /invoke format. Why separate file? Make it easy to see how transformation works """ @@ -22,7 +22,7 @@ class BedrockCohereEmbeddingConfig: ) -> dict: for k, v in non_default_params.items(): if k == "encoding_format": - optional_params["embedding_types"] = v + optional_params["embedding_types"] = v if isinstance(v, list) else [v] elif k == "dimensions": optional_params["output_dimension"] = v return optional_params diff --git a/litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py b/litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py index 151e0e404a0..42c3bd517a9 100644 --- a/litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py +++ b/litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py @@ -32,19 +32,26 @@ from litellm.llms.bedrock.chat.invoke_transformations.base_invoke_transformation AmazonInvokeConfig, ) from litellm.llms.bedrock.common_utils import ( + convert_bedrock_invoke_output_format_to_inline_schema, ensure_bedrock_anthropic_messages_tool_names, get_anthropic_beta_from_headers, is_claude_4_5_on_bedrock, + normalize_bedrock_opus_output_config_effort, normalize_tool_input_schema_types_for_bedrock_invoke, + pop_bedrock_invoke_output_config_format, remove_custom_field_from_tools, ) -from litellm.types.llms.anthropic import ANTHROPIC_TOOL_SEARCH_BETA_HEADER +from litellm.types.llms.anthropic import ( + ANTHROPIC_BETA_HEADER_VALUES, + ANTHROPIC_TOOL_SEARCH_BETA_HEADER, +) from litellm.types.llms.bedrock import BedrockInvokeAnthropicMessagesRequest from litellm.types.llms.openai import AllMessageValues from litellm.types.router import GenericLiteLLMParams from litellm.types.utils import GenericStreamingChunk from litellm.types.utils import GenericStreamingChunk as GChunk from litellm.types.utils import ModelResponseStream +from litellm.utils import _supports_factory if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj @@ -441,7 +448,7 @@ class AmazonAnthropicClaudeMessagesConfig( if isinstance(e, dict) and e.get("type") == "compact_20260112" ] if compact_edits: - beta_set.add("compact-2026-01-12") + beta_set.add(ANTHROPIC_BETA_HEADER_VALUES.COMPACT_2026_01_12.value) anthropic_messages_request["context_management"] = { **cm, "edits": compact_edits, @@ -449,123 +456,15 @@ class AmazonAnthropicClaudeMessagesConfig( else: anthropic_messages_request.pop("context_management", None) - def _convert_output_format_to_inline_schema( - self, - output_format: Dict, - anthropic_messages_request: Dict, - ) -> None: - """ - Convert Anthropic output_format to inline schema in message content. - - Bedrock Invoke doesn't support the output_format parameter, so we embed - the schema directly into the user message content as text instructions. - - This approach adds the schema to the last user message, instructing the model - to respond in the specified JSON format. - - Args: - output_format: The output_format dict with 'type' and 'schema' - anthropic_messages_request: The request dict to modify in-place - - Ref: https://aws.amazon.com/blogs/machine-learning/structured-data-response-with-amazon-bedrock-prompt-engineering-and-tool-use/ - """ - import json - - # Extract schema from output_format - schema = output_format.get("schema") - if not schema: - return - - # Get messages from the request - messages = anthropic_messages_request.get("messages", []) - if not messages: - return - - # Find the last user message - last_user_message_idx = None - for idx in range(len(messages) - 1, -1, -1): - if messages[idx].get("role") == "user": - last_user_message_idx = idx - break - - if last_user_message_idx is None: - return - - last_user_message = messages[last_user_message_idx] - content = last_user_message.get("content", []) - - # Ensure content is a list - if isinstance(content, str): - content = [{"type": "text", "text": content}] - last_user_message["content"] = content - - # Add schema as text content to the message - schema_text = {"type": "text", "text": json.dumps(schema)} - content.append(schema_text) - - def transform_anthropic_messages_request( + def _get_bedrock_invoke_anthropic_beta_headers( self, model: str, messages: List[Dict], anthropic_messages_optional_request_params: Dict, - litellm_params: GenericLiteLLMParams, headers: dict, - ) -> Dict: - anthropic_messages_request = AnthropicMessagesConfig.transform_anthropic_messages_request( - self=self, - model=model, - messages=messages, - anthropic_messages_optional_request_params=anthropic_messages_optional_request_params, - litellm_params=litellm_params, - headers=headers, - ) - ######################################################### - ############## BEDROCK Invoke SPECIFIC TRANSFORMATION ### - ######################################################### - - # 1. anthropic_version is required for all claude models - if "anthropic_version" not in anthropic_messages_request: - anthropic_messages_request["anthropic_version"] = ( - self.DEFAULT_BEDROCK_ANTHROPIC_API_VERSION - ) - - # 2. `stream` is not allowed in request body for bedrock invoke - if "stream" in anthropic_messages_request: - anthropic_messages_request.pop("stream", None) - - # 3. `model` is not allowed in request body for bedrock invoke - if "model" in anthropic_messages_request: - anthropic_messages_request.pop("model", None) - - injected_thinking_for_clear_thinking = ( - self._ensure_thinking_for_clear_thinking_context_management( - anthropic_messages_request=anthropic_messages_request, - model=model, - ) - ) - - # 4. Remove `ttl` field from cache_control in messages (Bedrock doesn't support it for older models) - self._remove_ttl_from_cache_control( - anthropic_messages_request=anthropic_messages_request, model=model - ) - - # 5. Convert `output_format` to inline schema (Bedrock invoke doesn't support output_format) - output_format = anthropic_messages_request.pop("output_format", None) - if output_format: - self._convert_output_format_to_inline_schema( - output_format=output_format, - anthropic_messages_request=anthropic_messages_request, - ) - - # 5a. Remove `custom` field from tools (Bedrock doesn't support it) - # Claude Code sends `custom: {defer_loading: true}` on tool definitions, - # which causes Bedrock to reject the request with "Extra inputs are not permitted" - # Ref: https://github.com/BerriAI/litellm/issues/22847 - remove_custom_field_from_tools(anthropic_messages_request) - normalize_tool_input_schema_types_for_bedrock_invoke(anthropic_messages_request) - ensure_bedrock_anthropic_messages_tool_names(anthropic_messages_request) - - # 6. AUTO-INJECT beta headers based on features used + anthropic_messages_request: Dict, + injected_thinking_for_clear_thinking: bool, + ) -> List[str]: anthropic_model_info = AnthropicModelInfo() tools = anthropic_messages_optional_request_params.get("tools") messages_typed = cast(List[AllMessageValues], messages) @@ -628,6 +527,160 @@ class AmazonAnthropicClaudeMessagesConfig( dropped_user_betas, ) + return filtered_betas + + def _strip_unsupported_bedrock_invoke_fields( + self, + anthropic_messages_request: Dict, + ) -> Dict: + allowed = self.BEDROCK_INVOKE_ALLOWED_TOP_LEVEL_FIELDS + stripped = sorted(k for k in anthropic_messages_request if k not in allowed) + if stripped: + verbose_logger.debug( + "Bedrock Invoke: stripping unsupported top-level request fields: %s", + stripped, + ) + return {k: v for k, v in anthropic_messages_request.items() if k in allowed} + + @staticmethod + def _clamp_adaptive_reasoning_effort_for_bedrock( + model: str, optional_params: Dict + ) -> None: + """Lower ``reasoning_effort`` to the Bedrock effort ceiling before validation. + + The shared ``/v1/messages`` effort gate rejects tiers a model does not + natively support (e.g. ``xhigh`` on Opus 4.6). Bedrock's chat paths instead + clamp the tier to the model's ``bedrock_output_config_effort_ceiling`` so + Claude Code "goal mode" keeps working; mirror that here so the messages + path degrades ``xhigh`` -> ``max`` rather than 400-ing. Non-adaptive models + and models without a ceiling are left untouched. + """ + if not AnthropicModelInfo._is_adaptive_thinking_model(model): + return + effort = optional_params.get("reasoning_effort") + if not isinstance(effort, str): + return + clamped = {"effort": effort} + normalize_bedrock_opus_output_config_effort(model=model, output_config=clamped) + optional_params["reasoning_effort"] = clamped["effort"] + + def transform_anthropic_messages_request( + self, + model: str, + messages: List[Dict], + anthropic_messages_optional_request_params: Dict, + litellm_params: GenericLiteLLMParams, + headers: dict, + ) -> Dict: + self._clamp_adaptive_reasoning_effort_for_bedrock( + model=model, + optional_params=anthropic_messages_optional_request_params, + ) + anthropic_messages_request = AnthropicMessagesConfig.transform_anthropic_messages_request( + self=self, + model=model, + messages=messages, + anthropic_messages_optional_request_params=anthropic_messages_optional_request_params, + litellm_params=litellm_params, + headers=headers, + ) + ######################################################### + ############## BEDROCK Invoke SPECIFIC TRANSFORMATION ### + ######################################################### + + # 1. anthropic_version is required for all claude models + if "anthropic_version" not in anthropic_messages_request: + anthropic_messages_request["anthropic_version"] = ( + self.DEFAULT_BEDROCK_ANTHROPIC_API_VERSION + ) + + # 2. `stream` is not allowed in request body for bedrock invoke + if "stream" in anthropic_messages_request: + anthropic_messages_request.pop("stream", None) + + # 3. `model` is not allowed in request body for bedrock invoke + if "model" in anthropic_messages_request: + anthropic_messages_request.pop("model", None) + + injected_thinking_for_clear_thinking = ( + self._ensure_thinking_for_clear_thinking_context_management( + anthropic_messages_request=anthropic_messages_request, + model=model, + ) + ) + + # 4. Remove `ttl` field from cache_control in messages (Bedrock doesn't support it for older models) + self._remove_ttl_from_cache_control( + anthropic_messages_request=anthropic_messages_request, model=model + ) + + # 5. Convert structured-output params to inline schema. + # Bedrock Invoke doesn't support top-level `output_format`; its + # accepted `output_config` subset is also narrower than Anthropic's, so + # consume the newer `output_config.format` shape here instead of + # forwarding it as an unknown nested key. + existing_output_config = anthropic_messages_request.get("output_config") + if isinstance(existing_output_config, dict): + anthropic_messages_request["output_config"] = dict(existing_output_config) + output_format = anthropic_messages_request.pop("output_format", None) + output_config_format = pop_bedrock_invoke_output_config_format( + anthropic_messages_request + ) + if output_format: + convert_bedrock_invoke_output_format_to_inline_schema( + output_format=output_format, + request_body=anthropic_messages_request, + ) + elif output_config_format: + convert_bedrock_invoke_output_format_to_inline_schema( + output_format=output_config_format, + request_body=anthropic_messages_request, + ) + normalize_bedrock_opus_output_config_effort( + model=model, + output_config=anthropic_messages_request.get("output_config"), + ) + + # 5a. Bedrock Invoke supports output_config (effort) for Claude 4.6+ models, + # but older models do not — strip it to avoid request rejection. + # Ref: https://github.com/BerriAI/litellm/issues/22797 + if not ( + _supports_factory( + model=model, + custom_llm_provider="bedrock", + key="supports_output_config", + ) + or AnthropicConfig._model_supports_effort_param(model) + ): + if anthropic_messages_request.pop("output_config", None) is not None: + verbose_logger.warning( + "Bedrock Invoke: stripping unsupported `output_config` for " + "model=%s — neither `supports_output_config` nor any " + "`supports_*_reasoning_effort` flag is set in " + "model_prices_and_context_window.json. Add the capability " + "flag to the model JSON entry if this model accepts " + "`output_config`.", + model, + ) + + # 5b. Remove `custom` field from tools (Bedrock doesn't support it) + # Claude Code sends `custom: {defer_loading: true}` on tool definitions, + # which causes Bedrock to reject the request with "Extra inputs are not permitted" + # Ref: https://github.com/BerriAI/litellm/issues/22847 + remove_custom_field_from_tools(anthropic_messages_request) + normalize_tool_input_schema_types_for_bedrock_invoke(anthropic_messages_request) + ensure_bedrock_anthropic_messages_tool_names(anthropic_messages_request) + + # 6. AUTO-INJECT beta headers based on features used + filtered_betas = self._get_bedrock_invoke_anthropic_beta_headers( + model=model, + messages=messages, + anthropic_messages_optional_request_params=anthropic_messages_optional_request_params, + headers=headers, + anthropic_messages_request=anthropic_messages_request, + injected_thinking_for_clear_thinking=injected_thinking_for_clear_thinking, + ) + if filtered_betas: anthropic_messages_request["anthropic_beta"] = filtered_betas @@ -646,16 +699,9 @@ class AmazonAnthropicClaudeMessagesConfig( # Catches Anthropic-only extensions (output_config, speed, mcp_servers, ...) # and any future additions Claude Code may start sending. ``context_management`` # has already been pre-filtered to its Bedrock-supported subset above. - allowed = self.BEDROCK_INVOKE_ALLOWED_TOP_LEVEL_FIELDS - stripped = sorted(k for k in anthropic_messages_request if k not in allowed) - if stripped: - verbose_logger.debug( - "Bedrock Invoke: stripping unsupported top-level request fields: %s", - stripped, - ) - anthropic_messages_request = { - k: v for k, v in anthropic_messages_request.items() if k in allowed - } + anthropic_messages_request = self._strip_unsupported_bedrock_invoke_fields( + anthropic_messages_request + ) return anthropic_messages_request diff --git a/litellm/llms/bedrock/messages/mantle_transformation.py b/litellm/llms/bedrock/messages/mantle_transformation.py index 3f04c8a3052..a78f696a057 100644 --- a/litellm/llms/bedrock/messages/mantle_transformation.py +++ b/litellm/llms/bedrock/messages/mantle_transformation.py @@ -20,7 +20,9 @@ if TYPE_CHECKING: else: LiteLLMLoggingObj = Any -MANTLE_ENDPOINT_TEMPLATE = "https://bedrock-mantle.{region}.api.aws/v1/messages" +MANTLE_ENDPOINT_TEMPLATE = ( + "https://bedrock-mantle.{region}.api.aws/anthropic/v1/messages" +) class AmazonMantleMessagesConfig(AmazonAnthropicClaudeMessagesConfig): diff --git a/litellm/llms/bedrock_mantle/chat/transformation.py b/litellm/llms/bedrock_mantle/chat/transformation.py index e413bb22b2d..81a56030a5c 100644 --- a/litellm/llms/bedrock_mantle/chat/transformation.py +++ b/litellm/llms/bedrock_mantle/chat/transformation.py @@ -16,7 +16,6 @@ from litellm.secret_managers.main import get_secret_str from ...openai_like.chat.transformation import OpenAILikeChatConfig - BEDROCK_MANTLE_DEFAULT_REGION = "us-east-1" diff --git a/litellm/llms/chatgpt/responses/transformation.py b/litellm/llms/chatgpt/responses/transformation.py index 66acd933416..56b61b66c84 100644 --- a/litellm/llms/chatgpt/responses/transformation.py +++ b/litellm/llms/chatgpt/responses/transformation.py @@ -1,7 +1,5 @@ -import json -from typing import Any, Optional +from typing import Any, Dict, Optional -from litellm.constants import STREAM_SSE_DONE_STRING from litellm.exceptions import AuthenticationError from litellm.litellm_core_utils.core_helpers import process_response_headers from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response import ( @@ -9,13 +7,17 @@ from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response impo ) from litellm.llms.openai.common_utils import OpenAIError from litellm.llms.openai.responses.transformation import OpenAIResponsesAPIConfig +from litellm.responses.sse_output_recovery import ( + parse_sse_json_chunk, + record_output_item_chunk, + record_output_text_chunk, +) from litellm.types.llms.openai import ( ResponsesAPIResponse, ResponsesAPIStreamEvents, ) from litellm.types.router import GenericLiteLLMParams from litellm.types.utils import LlmProviders -from litellm.utils import CustomStreamWrapper from ..authenticator import Authenticator from ..common_utils import ( @@ -111,86 +113,139 @@ class ChatGPTResponsesAPIConfig(OpenAIResponsesAPIConfig): raw_response: Any, logging_obj: Any, ): - content_type = (raw_response.headers or {}).get("content-type", "") body_text = raw_response.text or "" - if "text/event-stream" not in content_type.lower(): - trimmed_body = body_text.lstrip() - if not ( - trimmed_body.startswith("event:") - or trimmed_body.startswith("data:") - or "\nevent:" in body_text - or "\ndata:" in body_text - ): - return super().transform_response_api_response( - model=model, - raw_response=raw_response, - logging_obj=logging_obj, - ) + if not self._should_parse_as_sse( + raw_response=raw_response, body_text=body_text + ): + return super().transform_response_api_response( + model=model, + raw_response=raw_response, + logging_obj=logging_obj, + ) logging_obj.post_call( original_response=raw_response.text, additional_args={"complete_input_dict": {}}, ) - completed_response = None - error_message = None - for chunk in body_text.splitlines(): - stripped_chunk = CustomStreamWrapper._strip_sse_data_from_chunk(chunk) - if not stripped_chunk: - continue - stripped_chunk = stripped_chunk.strip() - if not stripped_chunk: - continue - if stripped_chunk == STREAM_SSE_DONE_STRING: - break - try: - parsed_chunk = json.loads(stripped_chunk) - except json.JSONDecodeError: - continue - if not isinstance(parsed_chunk, dict): - continue - event_type = parsed_chunk.get("type") - if event_type == ResponsesAPIStreamEvents.RESPONSE_COMPLETED: - response_payload = parsed_chunk.get("response") - if isinstance(response_payload, dict): - response_payload = dict(response_payload) - if "created_at" in response_payload: - response_payload["created_at"] = _safe_convert_created_field( - response_payload["created_at"] - ) - try: - completed_response = ResponsesAPIResponse(**response_payload) - except Exception: - completed_response = ResponsesAPIResponse.model_construct( - **response_payload - ) - break - if event_type in ( - ResponsesAPIStreamEvents.RESPONSE_FAILED, - ResponsesAPIStreamEvents.ERROR, - ): - error_obj = parsed_chunk.get("error") or ( - parsed_chunk.get("response") or {} - ).get("error") - if error_obj is not None: - if isinstance(error_obj, dict): - error_message = error_obj.get("message") or str(error_obj) - else: - error_message = str(error_obj) - + completed_response, error_message = self._extract_completed_response_from_sse( + body_text=body_text + ) if completed_response is None: raise OpenAIError( message=error_message or raw_response.text, status_code=raw_response.status_code, ) + self._attach_response_headers( + completed_response=completed_response, raw_response=raw_response + ) + return completed_response + + def _should_parse_as_sse(self, raw_response: Any, body_text: str) -> bool: + content_type = (raw_response.headers or {}).get("content-type", "") + if "text/event-stream" in content_type.lower(): + return True + trimmed_body = body_text.lstrip() + return bool( + trimmed_body.startswith("event:") + or trimmed_body.startswith("data:") + or "\nevent:" in body_text + or "\ndata:" in body_text + ) + + def _extract_completed_response_from_sse( + self, body_text: str + ) -> tuple[Optional[ResponsesAPIResponse], Optional[str]]: + completed_response = None + error_message = None + streamed_output_items: Dict[int, dict] = {} + text_only_output_items: Dict[int, dict] = {} + for chunk in body_text.splitlines(): + parsed_chunk = parse_sse_json_chunk(chunk) + if parsed_chunk is None: + continue + + event_type = parsed_chunk.get("type") + if event_type == ResponsesAPIStreamEvents.OUTPUT_ITEM_DONE: + record_output_item_chunk( + parsed_chunk=parsed_chunk, + output_items=streamed_output_items, + ) + continue + + if event_type == ResponsesAPIStreamEvents.OUTPUT_TEXT_DONE: + record_output_text_chunk( + parsed_chunk=parsed_chunk, + output_items=streamed_output_items, + text_only_items=text_only_output_items, + ) + continue + + if event_type == ResponsesAPIStreamEvents.RESPONSE_COMPLETED: + # Real OUTPUT_ITEM_DONE events take precedence at any given + # output_index, but text-only items at indices without a + # matching OUTPUT_ITEM_DONE must still be preserved (e.g. + # providers that emit only OUTPUT_TEXT_DONE for some indices). + merged_items: Dict[int, dict] = {**text_only_output_items} + merged_items.update(streamed_output_items) + completed_response = self._build_completed_response_from_chunk( + parsed_chunk=parsed_chunk, + streamed_output_items=merged_items, + ) + break + + if event_type in ( + ResponsesAPIStreamEvents.RESPONSE_FAILED, + ResponsesAPIStreamEvents.ERROR, + ): + extracted_error = self._extract_error_message(parsed_chunk) + if extracted_error is not None: + error_message = extracted_error + + return completed_response, error_message + + def _build_completed_response_from_chunk( + self, parsed_chunk: Dict[str, Any], streamed_output_items: Dict[int, dict] + ) -> Optional[ResponsesAPIResponse]: + response_payload = parsed_chunk.get("response") + if not isinstance(response_payload, dict): + return None + response_payload = dict(response_payload) + if not response_payload.get("output") and streamed_output_items: + response_payload["output"] = [ + item for _, item in sorted(streamed_output_items.items()) + ] + if "created_at" in response_payload: + response_payload["created_at"] = _safe_convert_created_field( + response_payload["created_at"] + ) + try: + return ResponsesAPIResponse(**response_payload) + except Exception: + return ResponsesAPIResponse.model_construct(**response_payload) + + def _extract_error_message(self, parsed_chunk: Dict[str, Any]) -> Optional[str]: + error_obj = parsed_chunk.get("error") or ( + parsed_chunk.get("response") or {} + ).get("error") + if error_obj is None: + return None + if isinstance(error_obj, dict): + return error_obj.get("message") or str(error_obj) + return str(error_obj) + + def _attach_response_headers( + self, + completed_response: ResponsesAPIResponse, + raw_response: Any, + ) -> None: raw_headers = dict(raw_response.headers) processed_headers = process_response_headers(raw_headers) if not hasattr(completed_response, "_hidden_params"): setattr(completed_response, "_hidden_params", {}) completed_response._hidden_params["additional_headers"] = processed_headers completed_response._hidden_params["headers"] = raw_headers - return completed_response def get_complete_url( self, diff --git a/litellm/llms/cohere/embed/handler.py b/litellm/llms/cohere/embed/handler.py index 3ab8baf7ba8..81b6a1c7aec 100644 --- a/litellm/llms/cohere/embed/handler.py +++ b/litellm/llms/cohere/embed/handler.py @@ -1,5 +1,5 @@ """ -Legacy /v1/embedding handler for Bedrock Cohere. +Legacy /v1/embedding handler for Bedrock Cohere. """ import json diff --git a/litellm/llms/cohere/embed/v1_transformation.py b/litellm/llms/cohere/embed/v1_transformation.py index feca9cb5b88..82c901e7eca 100644 --- a/litellm/llms/cohere/embed/v1_transformation.py +++ b/litellm/llms/cohere/embed/v1_transformation.py @@ -110,15 +110,35 @@ class CohereEmbeddingConfig: additional_args={"complete_input_dict": data}, original_response=response_json, ) + return self._populate_embedding_response( + response_json=response_json, + model_response=model_response, + model=model, + encoding=encoding, + input=input, + ) + + def _populate_embedding_response( + self, + response_json: dict, + model_response: EmbeddingResponse, + model: str, + encoding: Any, + input: list, + ) -> EmbeddingResponse: """ - response + Parse a Cohere embed response body into an OpenAI-style EmbeddingResponse. + + Split out from `_transform_response` so callers that already log + `post_call` themselves (e.g. SageMaker's embedding handler) can reuse + the parsing without triggering a second `post_call`. + + Response shape: { 'object': "list", - 'data': [ - - ] - 'model', - 'usage' + 'data': [...], + 'model', + 'usage', } """ embeddings = response_json["embeddings"] @@ -149,9 +169,6 @@ class CohereEmbeddingConfig: model_response.object = "list" model_response.data = output_data model_response.model = model - input_tokens = 0 - for text in input: - input_tokens += len(encoding.encode(text)) setattr( model_response, diff --git a/litellm/llms/custom_httpx/container_handler.py b/litellm/llms/custom_httpx/container_handler.py index 599cd705ebf..501390d840b 100644 --- a/litellm/llms/custom_httpx/container_handler.py +++ b/litellm/llms/custom_httpx/container_handler.py @@ -257,14 +257,19 @@ class GenericContainerHandler: returns_binary = endpoint_config.get("returns_binary", False) is_multipart = endpoint_config.get("is_multipart", False) + # An empty dict passed as `params` to httpx strips any existing query + # string from the URL (e.g. ?api-version=...). Use None instead so + # httpx leaves the URL's own query string intact. + effective_params = query_params or None + try: if method == "GET": response = http_client.get( - url=url, headers=headers, params=query_params + url=url, headers=headers, params=effective_params ) elif method == "DELETE": response = http_client.delete( - url=url, headers=headers, params=query_params + url=url, headers=headers, params=effective_params ) elif method == "POST": if is_multipart and "file" in kwargs: @@ -272,11 +277,11 @@ class GenericContainerHandler: kwargs["file"], headers ) response = http_client.post( - url=url, headers=headers, params=query_params, files=files + url=url, headers=headers, params=effective_params, files=files ) else: response = http_client.post( - url=url, headers=headers, params=query_params + url=url, headers=headers, params=effective_params ) else: raise ValueError(f"Unsupported HTTP method: {method}") @@ -376,14 +381,19 @@ class GenericContainerHandler: returns_binary = endpoint_config.get("returns_binary", False) is_multipart = endpoint_config.get("is_multipart", False) + # An empty dict passed as `params` to httpx strips any existing query + # string from the URL (e.g. ?api-version=...). Use None instead so + # httpx leaves the URL's own query string intact. + effective_params = query_params or None + try: if method == "GET": response = await http_client.get( - url=url, headers=headers, params=query_params + url=url, headers=headers, params=effective_params ) elif method == "DELETE": response = await http_client.delete( - url=url, headers=headers, params=query_params + url=url, headers=headers, params=effective_params ) elif method == "POST": if is_multipart and "file" in kwargs: @@ -391,11 +401,11 @@ class GenericContainerHandler: kwargs["file"], headers ) response = await http_client.post( - url=url, headers=headers, params=query_params, files=files + url=url, headers=headers, params=effective_params, files=files ) else: response = await http_client.post( - url=url, headers=headers, params=query_params + url=url, headers=headers, params=effective_params ) else: raise ValueError(f"Unsupported HTTP method: {method}") diff --git a/litellm/llms/custom_httpx/http_handler.py b/litellm/llms/custom_httpx/http_handler.py index af18c666679..e11d8532dbf 100644 --- a/litellm/llms/custom_httpx/http_handler.py +++ b/litellm/llms/custom_httpx/http_handler.py @@ -485,11 +485,16 @@ class MaskedHTTPStatusError(httpx.HTTPStatusError): if k.lower() not in ("content-encoding", "content-length") } + try: + request_content = original_error.request.content + except httpx.RequestNotRead: + request_content = b"" + masked_request = httpx.Request( method=original_error.request.method, url=masked_url, headers=original_error.request.headers, - content=original_error.request.content, + content=request_content, ) super().__init__( diff --git a/litellm/llms/custom_httpx/llm_http_handler.py b/litellm/llms/custom_httpx/llm_http_handler.py index fa1253d9005..941fe59e825 100644 --- a/litellm/llms/custom_httpx/llm_http_handler.py +++ b/litellm/llms/custom_httpx/llm_http_handler.py @@ -890,6 +890,18 @@ class BaseLLMHTTPHandler: headers=headers, ) + # Some providers (e.g. OCI) require request signing after the body is built. + # The default BaseConfig.sign_request returns (headers, None) — a no-op for + # providers that don't need signing. + headers, signed_body = provider_config.sign_request( + headers=headers, + optional_params=optional_params, + request_data=data, + api_base=api_base, + api_key=api_key, + model=model, + ) + ## LOGGING logging_obj.pre_call( input=input, @@ -916,6 +928,7 @@ class BaseLLMHTTPHandler: client=client, optional_params=optional_params, litellm_params=litellm_params, + signed_body=signed_body, ) if client is None or not isinstance(client, HTTPHandler): @@ -926,12 +939,20 @@ class BaseLLMHTTPHandler: sync_httpx_client = client try: - response = sync_httpx_client.post( - url=api_base, - headers=headers, - data=json.dumps(data), - timeout=timeout, - ) + if signed_body is not None: + response = sync_httpx_client.post( + url=api_base, + headers=headers, + data=signed_body, + timeout=timeout, + ) + else: + response = sync_httpx_client.post( + url=api_base, + headers=headers, + data=json.dumps(data), + timeout=timeout, + ) except Exception as e: raise self._handle_error( e=e, @@ -964,6 +985,7 @@ class BaseLLMHTTPHandler: api_key: Optional[str] = None, timeout: Optional[Union[float, httpx.Timeout]] = None, client: Optional[Union[HTTPHandler, AsyncHTTPHandler]] = None, + signed_body: Optional[bytes] = None, ) -> EmbeddingResponse: if client is None or not isinstance(client, AsyncHTTPHandler): async_httpx_client = get_async_httpx_client( @@ -974,12 +996,20 @@ class BaseLLMHTTPHandler: async_httpx_client = client try: - response = await async_httpx_client.post( - url=api_base, - headers=headers, - json=request_data, - timeout=timeout, - ) + if signed_body is not None: + response = await async_httpx_client.post( + url=api_base, + headers=headers, + data=signed_body, + timeout=timeout, + ) + else: + response = await async_httpx_client.post( + url=api_base, + headers=headers, + json=request_data, + timeout=timeout, + ) except Exception as e: raise self._handle_error(e=e, provider_config=provider_config) @@ -1177,6 +1207,8 @@ class BaseLLMHTTPHandler: data = transformed_result.data files = transformed_result.files + if transformed_result.content_type is not None: + headers["Content-Type"] = transformed_result.content_type ## LOGGING logging_obj.pre_call( @@ -1409,6 +1441,8 @@ class BaseLLMHTTPHandler: document=document, optional_params=optional_params, headers=headers, + api_key=api_key, + api_base=api_base, ) # All providers return OCRRequestData @@ -1477,6 +1511,8 @@ class BaseLLMHTTPHandler: document=document, optional_params=optional_params, headers=headers, + api_key=api_key, + api_base=api_base, ) # All providers return OCRRequestData @@ -1852,7 +1888,9 @@ class BaseLLMHTTPHandler: async_httpx_client: AsyncHTTPHandler, request_url: str, headers: dict, - signed_json_body: Optional[bytes], + # str when the caller passes a pre-serialized (unsigned) body to avoid + # re-dumping; bytes when a provider signed the request (e.g. Bedrock). + signed_json_body: Optional[Union[str, bytes]], request_body: dict, stream: bool, logging_obj: LiteLLMLoggingObj, @@ -2043,8 +2081,18 @@ class BaseLLMHTTPHandler: model=model, ) + # The request body was serialized once for the pre-call log input and + # again for the wire (json.dumps is O(payload), large for long-context + # Claude Code history). Serialize once and reuse for both. Only when + # the provider didn't sign the request (sign_request no-op for the + # native anthropic path -> signed_json_body is None); signed providers + # (e.g. Bedrock) keep their signed body untouched. The HTTP-error + # retry path mutates + re-signs the body, so it still re-serializes + # internally -- this only deduplicates the success path. + request_body_json = json.dumps(request_body) + logging_obj.pre_call( - input=[{"role": "user", "content": json.dumps(request_body)}], + input=[{"role": "user", "content": request_body_json}], api_key="", additional_args={ "complete_input_dict": request_body, @@ -2057,7 +2105,9 @@ class BaseLLMHTTPHandler: async_httpx_client=async_httpx_client, request_url=request_url, headers=headers, - signed_json_body=signed_json_body, + signed_json_body=( + signed_json_body if signed_json_body is not None else request_body_json + ), request_body=request_body, stream=stream or False, logging_obj=logging_obj, @@ -2079,6 +2129,14 @@ class BaseLLMHTTPHandler: litellm_logging_obj=logging_obj, ) + if not self._has_agentic_completion_hook(logging_obj): + # No callback overrides async_should_run_agentic_loop, so the + # agentic wrapper's only effect would be buffering every chunk + # and rebuilding the response from SSE at end-of-stream to call + # hooks that all return (False, {}). Stream through directly and + # skip that per-chunk + end-of-stream overhead. + return completion_stream + from litellm.llms.anthropic.experimental_pass_through.messages.agentic_streaming_iterator import ( AgenticAnthropicStreamingIterator, ) @@ -4586,6 +4644,51 @@ class BaseLLMHTTPHandler: fingerprints = list(kwargs.get("_agentic_loop_fingerprints", []) or []) return depth, max(max_loops, 1), fingerprints + @staticmethod + def _has_agentic_completion_hook(logging_obj: Any) -> bool: + """ + True if any registered callback actually overrides + ``async_should_run_agentic_loop`` (the gate every agentic hook goes + through). The base ``CustomLogger`` implementation returns + ``(False, {})``, so when nothing overrides it the agentic + post-processing is a guaranteed no-op and the streaming wrapper that + buffers + rebuilds the whole response from SSE just to call it can be + skipped entirely. + + Function-identity comparison (not a leaf ``__dict__`` check) so an + override inherited through any intermediate class is still detected -- + a false negative here would silently disable agentic features. + + String entries in ``litellm.callbacks`` (e.g. ``"datadog"``) are + resolved to their ``CustomLogger`` instance via + ``get_custom_logger_compatible_class`` -- same pattern as + ``ProxyLogging._callback_capabilities`` -- so a string-registered + agentic callback is detected too. + """ + from litellm.integrations.custom_logger import CustomLogger + from litellm.litellm_core_utils.litellm_logging import ( + get_custom_logger_compatible_class, + ) + + base_func = CustomLogger.async_should_run_agentic_loop + callbacks = litellm.callbacks + ( + getattr(logging_obj, "dynamic_success_callbacks", None) or [] + ) + for cb in callbacks: + if isinstance(cb, str): + resolved = get_custom_logger_compatible_class(cb) # type: ignore[arg-type] + if resolved is None: + continue + cb = resolved + if not isinstance(cb, CustomLogger): + continue + cb_func = getattr(type(cb), "async_should_run_agentic_loop", base_func) + if getattr(cb_func, "__func__", cb_func) is not getattr( + base_func, "__func__", base_func + ): + return True + return False + @staticmethod def _check_agentic_loop_safety( tool_calls: Any, @@ -4634,6 +4737,7 @@ class BaseLLMHTTPHandler: fingerprints: List[str], fingerprint: str, stream: bool = False, + callback: Optional[Any] = None, ) -> Any: from litellm.anthropic_interface import messages as anthropic_messages @@ -4675,7 +4779,7 @@ class BaseLLMHTTPHandler: kwargs_for_followup["max_agentic_loops"] = max_loops kwargs_for_followup["_agentic_loop_fingerprints"] = fingerprints + [fingerprint] - return await anthropic_messages.acreate( + response = await anthropic_messages.acreate( **{ "max_tokens": max_tokens, "messages": patch.messages, @@ -4686,6 +4790,23 @@ class BaseLLMHTTPHandler: } ) + if callback is not None: + try: + response = await callback.async_post_agentic_loop_response_hook( + response=response, plan=plan, kwargs=kwargs + ) + except Exception as e: + _call_id = getattr(logging_obj, "litellm_call_id", "unknown") + verbose_logger.exception( + "LiteLLM.AgenticHookError: Exception in " + "async_post_agentic_loop_response_hook [call_id=%s model=%s]: %s", + _call_id, + model, + str(e), + ) + + return response + async def _execute_chat_completion_agentic_plan( self, plan: AgenticLoopPlan, @@ -4869,6 +4990,7 @@ class BaseLLMHTTPHandler: fingerprints=fingerprints, fingerprint=fingerprint, stream=stream, + callback=callback, ) except Exception as e: _call_id = getattr(logging_obj, "litellm_call_id", "unknown") @@ -5194,6 +5316,28 @@ class BaseLLMHTTPHandler: ) if _session_config: realtime_streaming.session_configuration_request = _session_config + + # For providers that defer setup until client session.update, optionally + # send synthetic session.created to unblock clients waiting on connect. + if not provider_config.requires_session_configuration(): + synthetic_session = provider_config.transform_session_created_event( + model=model, + logging_session_id=logging_obj.litellm_trace_id, + session_configuration_request=None, + ) + if synthetic_session is not None: + synthetic_session_str = json.dumps(synthetic_session) + # Record before sending so the synthetic session.created is + # captured in the session log alongside provider-driven + # events; without this it would be silently absent from + # success_handler / async_success_handler payloads. + realtime_streaming.store_message(synthetic_session_str) + await websocket.send_text(synthetic_session_str) + realtime_streaming._session_created_sent_to_client = True + verbose_logger.debug( + "Sent synthetic session.created to client to unblock connection" + ) + await realtime_streaming.bidirectional_forward() except websockets.exceptions.InvalidStatusCode as e: # type: ignore @@ -6416,6 +6560,7 @@ class BaseLLMHTTPHandler: api_key=api_key or litellm_params.get("api_key", None), headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: @@ -6498,6 +6643,7 @@ class BaseLLMHTTPHandler: api_key=api_key or litellm_params.get("api_key", None), headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: @@ -6590,6 +6736,7 @@ class BaseLLMHTTPHandler: api_key=api_key or litellm_params.get("api_key", None), headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: headers.update(extra_headers) @@ -6661,6 +6808,7 @@ class BaseLLMHTTPHandler: api_key=api_key or litellm_params.get("api_key", None), headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: headers.update(extra_headers) @@ -6744,6 +6892,7 @@ class BaseLLMHTTPHandler: api_key=api_key or litellm_params.get("api_key", None), headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: headers.update(extra_headers) @@ -6801,6 +6950,7 @@ class BaseLLMHTTPHandler: api_key=api_key or litellm_params.get("api_key", None), headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: headers.update(extra_headers) @@ -6877,6 +7027,7 @@ class BaseLLMHTTPHandler: api_key=api_key or litellm_params.get("api_key", None), headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: headers.update(extra_headers) @@ -6887,27 +7038,49 @@ class BaseLLMHTTPHandler: litellm_params=dict(litellm_params), ) - url, data = video_provider_config.transform_video_edit_request( - prompt=prompt, + prefetched_source_data = None + prefetch_params = video_provider_config.get_video_edit_prefetch_params( video_id=video_id, api_base=api_base, litellm_params=litellm_params, headers=headers, - extra_body=extra_body, - ) - - logging_obj.pre_call( - input=prompt, - api_key="", - additional_args={ - "complete_input_dict": data, - "api_base": url, - "headers": headers, - "video_id": video_id, - }, ) + if prefetch_params is not None: + prefetch_url, prefetch_body = prefetch_params + try: + prefetch_resp = sync_httpx_client.post( + url=prefetch_url, + headers=headers, + json=prefetch_body, + timeout=timeout, + ) + prefetch_resp.raise_for_status() + except Exception as e: + raise self._handle_error(e=e, provider_config=video_provider_config) + prefetched_source_data = prefetch_resp.json() try: + url, data = video_provider_config.transform_video_edit_request( + prompt=prompt, + video_id=video_id, + api_base=api_base, + litellm_params=litellm_params, + headers=headers, + extra_body=extra_body, + prefetched_source_data=prefetched_source_data, + ) + + logging_obj.pre_call( + input=prompt, + api_key="", + additional_args={ + "complete_input_dict": data, + "api_base": url, + "headers": headers, + "video_id": video_id, + }, + ) + response = sync_httpx_client.post( url=url, headers=headers, @@ -6919,6 +7092,7 @@ class BaseLLMHTTPHandler: raw_response=response, logging_obj=logging_obj, custom_llm_provider=custom_llm_provider, + request_data=data, ) except Exception as e: raise self._handle_error(e=e, provider_config=video_provider_config) @@ -6949,6 +7123,7 @@ class BaseLLMHTTPHandler: api_key=api_key or litellm_params.get("api_key", None), headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: headers.update(extra_headers) @@ -6959,27 +7134,49 @@ class BaseLLMHTTPHandler: litellm_params=dict(litellm_params), ) - url, data = video_provider_config.transform_video_edit_request( - prompt=prompt, + prefetched_source_data = None + prefetch_params = video_provider_config.get_video_edit_prefetch_params( video_id=video_id, api_base=api_base, litellm_params=litellm_params, headers=headers, - extra_body=extra_body, - ) - - logging_obj.pre_call( - input=prompt, - api_key="", - additional_args={ - "complete_input_dict": data, - "api_base": url, - "headers": headers, - "video_id": video_id, - }, ) + if prefetch_params is not None: + prefetch_url, prefetch_body = prefetch_params + try: + prefetch_resp = await async_httpx_client.post( + url=prefetch_url, + headers=headers, + json=prefetch_body, + timeout=timeout, + ) + prefetch_resp.raise_for_status() + except Exception as e: + raise self._handle_error(e=e, provider_config=video_provider_config) + prefetched_source_data = prefetch_resp.json() try: + url, data = video_provider_config.transform_video_edit_request( + prompt=prompt, + video_id=video_id, + api_base=api_base, + litellm_params=litellm_params, + headers=headers, + extra_body=extra_body, + prefetched_source_data=prefetched_source_data, + ) + + logging_obj.pre_call( + input=prompt, + api_key="", + additional_args={ + "complete_input_dict": data, + "api_base": url, + "headers": headers, + "video_id": video_id, + }, + ) + response = await async_httpx_client.post( url=url, headers=headers, @@ -6991,6 +7188,7 @@ class BaseLLMHTTPHandler: raw_response=response, logging_obj=logging_obj, custom_llm_provider=custom_llm_provider, + request_data=data, ) except Exception as e: raise self._handle_error(e=e, provider_config=video_provider_config) @@ -7038,6 +7236,7 @@ class BaseLLMHTTPHandler: api_key=api_key or litellm_params.get("api_key", None), headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: headers.update(extra_headers) @@ -7112,6 +7311,7 @@ class BaseLLMHTTPHandler: api_key=api_key or litellm_params.get("api_key", None), headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: headers.update(extra_headers) @@ -7323,6 +7523,7 @@ class BaseLLMHTTPHandler: api_key=api_key, headers=extra_headers or {}, model="", + litellm_params=litellm_params, ) if extra_headers: @@ -7815,7 +8016,7 @@ class BaseLLMHTTPHandler: response = sync_httpx_client.get( url=url, headers=headers, - params=params, + params=params or None, ) return container_provider_config.transform_container_list_response( @@ -7892,7 +8093,7 @@ class BaseLLMHTTPHandler: response = await async_httpx_client.get( url=url, headers=headers, - params=params, + params=params or None, ) return container_provider_config.transform_container_list_response( @@ -7982,7 +8183,7 @@ class BaseLLMHTTPHandler: response = sync_httpx_client.get( url=url, headers=headers, - params=params, + params=params or None, ) return container_provider_config.transform_container_retrieve_response( @@ -8059,7 +8260,7 @@ class BaseLLMHTTPHandler: response = await async_httpx_client.get( url=url, headers=headers, - params=params, + params=params or None, ) return container_provider_config.transform_container_retrieve_response( @@ -8149,7 +8350,7 @@ class BaseLLMHTTPHandler: response = sync_httpx_client.delete( url=url, headers=headers, - params=params, + params=params or None, ) return container_provider_config.transform_container_delete_response( @@ -8226,7 +8427,7 @@ class BaseLLMHTTPHandler: response = await async_httpx_client.delete( url=url, headers=headers, - params=params, + params=params or None, ) return container_provider_config.transform_container_delete_response( @@ -8322,7 +8523,7 @@ class BaseLLMHTTPHandler: response = sync_httpx_client.get( url=url, headers=headers, - params=params, + params=params or None, ) return container_provider_config.transform_container_file_list_response( @@ -8401,7 +8602,7 @@ class BaseLLMHTTPHandler: response = await async_httpx_client.get( url=url, headers=headers, - params=params, + params=params or None, ) return container_provider_config.transform_container_file_list_response( @@ -8489,7 +8690,7 @@ class BaseLLMHTTPHandler: response = sync_httpx_client.get( url=url, headers=headers, - params=params, + params=params or None, ) return container_provider_config.transform_container_file_content_response( @@ -8565,7 +8766,7 @@ class BaseLLMHTTPHandler: response = await async_httpx_client.get( url=url, headers=headers, - params=params, + params=params or None, ) return container_provider_config.transform_container_file_content_response( diff --git a/litellm/llms/custom_httpx/mock_transport.py b/litellm/llms/custom_httpx/mock_transport.py index c9844753e0e..ad93cc134ee 100644 --- a/litellm/llms/custom_httpx/mock_transport.py +++ b/litellm/llms/custom_httpx/mock_transport.py @@ -13,7 +13,6 @@ from typing import Tuple import httpx - # --------------------------------------------------------------------------- # Pre-built response templates # --------------------------------------------------------------------------- diff --git a/litellm/llms/dashscope/common_utils.py b/litellm/llms/dashscope/common_utils.py new file mode 100644 index 00000000000..b3b89cbbebf --- /dev/null +++ b/litellm/llms/dashscope/common_utils.py @@ -0,0 +1,28 @@ +""" +Common utilities for the DashScope LLM provider. +""" + +from typing import Optional + +import httpx + +from litellm.llms.base_llm.chat.transformation import BaseLLMException + + +class DashScopeError(BaseLLMException): + """Exception class for DashScope provider errors.""" + + def __init__( + self, + status_code: int, + message: str, + headers: Optional[httpx.Headers] = None, + ): + self.status_code = status_code + self.message = message + self.headers = headers or httpx.Headers() + super().__init__( + status_code=status_code, + message=message, + headers=dict(self.headers), + ) diff --git a/litellm/llms/dashscope/cost_calculator.py b/litellm/llms/dashscope/cost_calculator.py index 9b3e3851162..8bb7f605b82 100644 --- a/litellm/llms/dashscope/cost_calculator.py +++ b/litellm/llms/dashscope/cost_calculator.py @@ -1,5 +1,5 @@ """ -Cost calculator for Dashscope Chat models. +Cost calculator for Dashscope Chat models. Handles tiered pricing and prompt caching scenarios. """ diff --git a/litellm/llms/dashscope/embed/__init__.py b/litellm/llms/dashscope/embed/__init__.py new file mode 100644 index 00000000000..4962b1f3251 --- /dev/null +++ b/litellm/llms/dashscope/embed/__init__.py @@ -0,0 +1,7 @@ +""" +DashScope Embedding Module +""" + +from .transformation import DashScopeEmbeddingConfig + +__all__ = ["DashScopeEmbeddingConfig"] diff --git a/litellm/llms/dashscope/embed/transformation.py b/litellm/llms/dashscope/embed/transformation.py new file mode 100644 index 00000000000..5bc0e5ca817 --- /dev/null +++ b/litellm/llms/dashscope/embed/transformation.py @@ -0,0 +1,191 @@ +""" +Transformation logic from OpenAI /v1/embeddings format to DashScope's /v1/embeddings format. + +Supports +- text-embedding-v4 +- text-embedding-v3 + +Endpoint +- https://dashscope.aliyuncs.com/compatible-mode/v1/embeddings + +Docs - https://help.aliyun.com/zh/model-studio/text-embedding-synchronous-api +""" + +from typing import List, Optional, Union + +import httpx + +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.llms.base_llm.chat.transformation import BaseLLMException +from litellm.llms.base_llm.embedding.transformation import BaseEmbeddingConfig +from litellm.secret_managers.main import get_secret_str +from litellm.types.llms.openai import AllEmbeddingInputValues, AllMessageValues +from litellm.types.utils import EmbeddingResponse, Usage + +from ..common_utils import DashScopeError + +DEFAULT_API_BASE = "https://dashscope.aliyuncs.com/compatible-mode/v1" + + +class DashScopeEmbeddingConfig(BaseEmbeddingConfig): + """ + Reference: https://help.aliyun.com/zh/model-studio/text-embedding-synchronous-api + + DashScope exposes an OpenAI-compatible /v1/embeddings endpoint, so the + request and response shapes are nearly identical to OpenAI's. + """ + + def __init__(self) -> None: + pass + + def get_supported_openai_params(self, model: str) -> List[str]: + # DashScope's compatible-mode embeddings API accepts the same params as OpenAI. + # `dimensions` / `encoding_format` are only honored by text-embedding-v3 / v4; + # earlier versions silently ignore them server-side. + return ["dimensions", "encoding_format", "user"] + + def map_openai_params( + self, + non_default_params: dict, + optional_params: dict, + model: str, + drop_params: bool = False, + ) -> dict: + supported = self.get_supported_openai_params(model) + for k, v in non_default_params.items(): + if v is None: + continue + if k in supported: + optional_params[k] = v + # unsupported params are dropped when drop_params=True; + # the upstream _check_valid_arg already raised UnsupportedParamsError + # for drop_params=False before this method is called. + return optional_params + + def validate_environment( + self, + headers: dict, + model: str, + messages: List[AllMessageValues], + optional_params: dict, + litellm_params: dict, + api_key: Optional[str] = None, + api_base: Optional[str] = None, + ) -> dict: + if api_key is None: + api_key = get_secret_str("DASHSCOPE_API_KEY") + if api_key is None: + raise ValueError( + "DashScope API key is required. Set 'DASHSCOPE_API_KEY' env var or pass api_key explicitly." + ) + default_headers = { + "Content-Type": "application/json", + "Authorization": f"Bearer {api_key}", + } + return {**default_headers, **headers} + + def get_complete_url( + self, + api_base: Optional[str], + api_key: Optional[str], + model: str, + optional_params: dict, + litellm_params: dict, + stream: Optional[bool] = None, + ) -> str: + base = api_base or get_secret_str("DASHSCOPE_API_BASE") or DEFAULT_API_BASE + base = base.rstrip("/") + if base.endswith("/embeddings"): + return base + return f"{base}/embeddings" + + def transform_embedding_request( + self, + model: str, + input: AllEmbeddingInputValues, + optional_params: dict, + headers: dict, + ) -> dict: + data: dict = { + "model": model, + "input": input, + } + for key in ("dimensions", "encoding_format", "user"): + value = optional_params.get(key) + if value is not None: + data[key] = value + return data + + def transform_embedding_response( + self, + model: str, + raw_response: httpx.Response, + model_response: EmbeddingResponse, + logging_obj: LiteLLMLoggingObj, + api_key: Optional[str], + request_data: dict, + optional_params: dict, + litellm_params: dict, + ) -> EmbeddingResponse: + try: + response_json = raw_response.json() + except Exception as e: + raise DashScopeError( + status_code=raw_response.status_code, + message=f"Failed to parse DashScope response as JSON: {str(e)}", + ) + + logging_obj.post_call( + input=request_data.get("input"), + api_key=api_key, + additional_args={"complete_input_dict": request_data}, + original_response=response_json, + ) + + if "error" in response_json: + error = response_json["error"] + message = ( + error.get("message", str(error)) + if isinstance(error, dict) + else str(error) + ) + raise DashScopeError( + status_code=raw_response.status_code, + message=message, + ) + + model_response.object = "list" + model_response.data = response_json.get("data", []) + model_response.model = response_json.get("model", model) + + usage = response_json.get("usage") or {} + prompt_tokens = usage.get("prompt_tokens", 0) + total_tokens = usage.get("total_tokens", prompt_tokens) + setattr( + model_response, + "usage", + Usage( + prompt_tokens=prompt_tokens, + completion_tokens=0, + total_tokens=total_tokens, + ), + ) + + if "id" in response_json: + setattr(model_response, "id", response_json["id"]) + + return model_response + + def get_error_class( + self, + error_message: str, + status_code: int, + headers: Union[dict, httpx.Headers], + ) -> BaseLLMException: + if isinstance(headers, dict): + headers = httpx.Headers(headers) + return DashScopeError( + status_code=status_code, + message=error_message, + headers=headers, + ) diff --git a/litellm/llms/dashscope/rerank/__init__.py b/litellm/llms/dashscope/rerank/__init__.py new file mode 100644 index 00000000000..2a1401f6dc0 --- /dev/null +++ b/litellm/llms/dashscope/rerank/__init__.py @@ -0,0 +1,7 @@ +""" +DashScope Rerank Module +""" + +from .transformation import DashScopeRerankConfig + +__all__ = ["DashScopeRerankConfig"] diff --git a/litellm/llms/dashscope/rerank/transformation.py b/litellm/llms/dashscope/rerank/transformation.py new file mode 100644 index 00000000000..629f3cf4af7 --- /dev/null +++ b/litellm/llms/dashscope/rerank/transformation.py @@ -0,0 +1,241 @@ +""" +Transformation logic for DashScope's OpenAI-compatible /v1/reranks API. + +Supports +- qwen3-rerank + +(Other DashScope rerankers — gte-rerank-v2 / qwen3-vl-rerank — share the same +endpoint but have not been validated against this transformer. Behavior with +those models is undefined.) + +Endpoint +- https://dashscope.aliyuncs.com/compatible-api/v1/reranks + +Note: chat/embed live under `/compatible-mode/v1/`, but DashScope's rerank +route is exposed under `/compatible-api/v1/reranks` per the docs. Override +with `DASHSCOPE_API_BASE_RERANK` to point at a different host or path. + +Empirically, qwen3-rerank accepts `return_documents=true` and echoes +`results[].document.text` back, even though the public docs list the flag +as supported only for gte-rerank-v2 / qwen3-vl-rerank. + +Docs - https://help.aliyun.com/zh/model-studio/text-rerank-api +""" + +from typing import Any, Dict, List, Optional, Union + +import httpx + +from litellm._uuid import uuid +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.llms.base_llm.chat.transformation import BaseLLMException +from litellm.llms.base_llm.rerank.transformation import BaseRerankConfig +from litellm.secret_managers.main import get_secret_str +from litellm.types.rerank import ( + OptionalRerankParams, + RerankBilledUnits, + RerankResponse, + RerankResponseMeta, + RerankTokens, +) + +from ..common_utils import DashScopeError + +DEFAULT_RERANK_URL = "https://dashscope.aliyuncs.com/compatible-api/v1/reranks" + + +class DashScopeRerankConfig(BaseRerankConfig): + """ + Reference: https://help.aliyun.com/zh/model-studio/text-rerank-api + + Targets DashScope's qwen3-rerank model. Request fields: model, query, + documents, top_n, return_documents. Response: results[].index, + results[].relevance_score, optionally results[].document.text (when + return_documents=true), plus a top-level usage.total_tokens counter. + """ + + def __init__(self) -> None: + pass + + def get_complete_url( + self, + api_base: Optional[str], + model: str, + optional_params: Optional[dict] = None, + ) -> str: + if api_base is None: + api_base = get_secret_str("DASHSCOPE_API_BASE_RERANK") or DEFAULT_RERANK_URL + + if api_base == DEFAULT_RERANK_URL: + return DEFAULT_RERANK_URL + + cleaned = api_base.rstrip("/") + if cleaned.endswith("/reranks") or cleaned.endswith("/rerank"): + return cleaned + + if cleaned.endswith("/v1"): + return f"{cleaned}/reranks" + + # Unknown base: append /reranks rather than silently ignoring the caller's api_base. + return f"{cleaned}/reranks" + + def validate_environment( + self, + headers: dict, + model: str, + api_key: Optional[str] = None, + optional_params: Optional[dict] = None, + ) -> dict: + if api_key is None: + api_key = get_secret_str("DASHSCOPE_API_KEY") + if api_key is None: + raise ValueError( + "DashScope API key is required. Set 'DASHSCOPE_API_KEY' env var or pass api_key explicitly." + ) + + default_headers = { + "Authorization": f"Bearer {api_key}", + "accept": "application/json", + "content-type": "application/json", + } + return {**default_headers, **headers} + + def get_supported_cohere_rerank_params(self, model: str) -> list: + return ["query", "documents", "top_n", "return_documents"] + + def map_cohere_rerank_params( + self, + non_default_params: Optional[dict], + model: str, + drop_params: bool, + query: str, + documents: List[Union[str, Dict[str, Any]]], + custom_llm_provider: Optional[str] = None, + top_n: Optional[int] = None, + rank_fields: Optional[List[str]] = None, + return_documents: Optional[bool] = True, + max_chunks_per_doc: Optional[int] = None, + max_tokens_per_doc: Optional[int] = None, + ) -> Dict: + # qwen3-rerank accepts query/documents/top_n/return_documents. The + # rest (rank_fields, max_*_per_doc) are silently dropped. + params: OptionalRerankParams = OptionalRerankParams( + query=query, + documents=documents, + ) + if top_n is not None: + params["top_n"] = top_n + if return_documents is not None: + params["return_documents"] = return_documents + return dict(params) + + def transform_rerank_request( + self, + model: str, + optional_rerank_params: Dict, + headers: dict, + litellm_params: Optional[dict] = None, + ) -> dict: + if "query" not in optional_rerank_params: + raise ValueError("query is required for DashScope rerank") + if "documents" not in optional_rerank_params: + raise ValueError("documents is required for DashScope rerank") + + request: Dict[str, Any] = { + "model": model, + "query": optional_rerank_params["query"], + "documents": optional_rerank_params["documents"], + } + if optional_rerank_params.get("top_n") is not None: + request["top_n"] = optional_rerank_params["top_n"] + if optional_rerank_params.get("return_documents") is not None: + request["return_documents"] = optional_rerank_params["return_documents"] + return request + + def transform_rerank_response( + self, + model: str, + raw_response: httpx.Response, + model_response: RerankResponse, + logging_obj: LiteLLMLoggingObj, + api_key: Optional[str] = None, + request_data: Optional[dict] = None, + optional_params: Optional[dict] = None, + litellm_params: Optional[dict] = None, + ) -> RerankResponse: + request_data = request_data or {} + optional_params = optional_params or {} + litellm_params = litellm_params or {} + try: + response_json = raw_response.json() + except Exception: + raise DashScopeError( + status_code=raw_response.status_code, + message=raw_response.text, + ) + + logging_obj.post_call( + input=request_data.get("query"), + api_key=api_key, + additional_args={"complete_input_dict": request_data}, + original_response=response_json, + ) + + # DashScope error envelope: {"code": "...", "message": "...", "request_id": "..."} + if "code" in response_json and "results" not in response_json: + raise DashScopeError( + status_code=raw_response.status_code, + message=response_json.get("message", str(response_json)), + ) + + results = response_json.get("results") + if results is None: + raise DashScopeError( + status_code=raw_response.status_code, + message=f"No results in DashScope rerank response: {response_json}", + ) + + # qwen3-rerank returns: + # {"index": int, "relevance_score": float} + # plus, when return_documents=true was sent: + # "document": {"text": "..."} + # which already matches LiteLLM's RerankResponseDocument shape. + transformed_results: List[dict] = [] + for r in results: + item: Dict[str, Any] = { + "index": r["index"], + "relevance_score": r["relevance_score"], + } + doc = r.get("document") + if isinstance(doc, dict): + item["document"] = doc + elif isinstance(doc, str): + # Defensive: spec says dict, but normalize string-shaped echoes. + item["document"] = {"text": doc} + transformed_results.append(item) + + usage = response_json.get("usage") or {} + total_tokens = usage.get("total_tokens") + billed_units = RerankBilledUnits(total_tokens=total_tokens) + tokens = RerankTokens(input_tokens=total_tokens) + meta = RerankResponseMeta(billed_units=billed_units, tokens=tokens) + + return RerankResponse( + id=response_json.get("id") or str(uuid.uuid4()), + results=transformed_results, # type: ignore + meta=meta, + ) + + def get_error_class( + self, + error_message: str, + status_code: int, + headers: Union[dict, httpx.Headers], + ) -> BaseLLMException: + if isinstance(headers, dict): + headers = httpx.Headers(headers) + return DashScopeError( + status_code=status_code, + message=error_message, + headers=headers, + ) diff --git a/litellm/llms/datarobot/chat/transformation.py b/litellm/llms/datarobot/chat/transformation.py index 23ce63c25b2..f81e2420930 100644 --- a/litellm/llms/datarobot/chat/transformation.py +++ b/litellm/llms/datarobot/chat/transformation.py @@ -1,5 +1,5 @@ """ -Support for OpenAI's `/v1/chat/completions` endpoint. +Support for OpenAI's `/v1/chat/completions` endpoint. Calls done in OpenAI/openai.py as DataRobot is openai-compatible. """ diff --git a/litellm/llms/deepinfra/rerank/transformation.py b/litellm/llms/deepinfra/rerank/transformation.py index 276735f4758..e4bfbcb2513 100644 --- a/litellm/llms/deepinfra/rerank/transformation.py +++ b/litellm/llms/deepinfra/rerank/transformation.py @@ -1,5 +1,5 @@ """ -Translate between Cohere's `/rerank` format and Deepinfra's `/rerank` format. +Translate between Cohere's `/rerank` format and Deepinfra's `/rerank` format. """ from typing import Any, Dict, List, Optional, Union diff --git a/litellm/llms/deepseek/chat/transformation.py b/litellm/llms/deepseek/chat/transformation.py index 5cd8d119542..7ed3e484535 100644 --- a/litellm/llms/deepseek/chat/transformation.py +++ b/litellm/llms/deepseek/chat/transformation.py @@ -2,13 +2,15 @@ Translates from OpenAI's `/v1/chat/completions` to DeepSeek's `/v1/chat/completions` """ -from typing import Any, Coroutine, List, Literal, Optional, Tuple, Union, overload +from typing import Any, Coroutine, List, Literal, Optional, Tuple, Union, cast, overload +import litellm from litellm.litellm_core_utils.prompt_templates.common_utils import ( handle_messages_with_content_list_to_str_conversion, ) from litellm.secret_managers.main import get_secret_str from litellm.types.llms.openai import AllMessageValues +from litellm.utils import supports_reasoning from ...openai.chat.gpt_transformation import OpenAIGPTConfig @@ -62,6 +64,48 @@ class DeepSeekChatConfig(OpenAIGPTConfig): return optional_params + def _fill_reasoning_content( + self, messages: List[AllMessageValues] + ) -> List[AllMessageValues]: + """ + DeepSeek thinking mode requires `reasoning_content` to be passed back on + every assistant message in multi-turn conversations. If it is missing, + the API returns: + "The reasoning_content in the thinking mode must be passed back to the API." + + For each assistant message that is missing `reasoning_content`: + 1. Promote it from `provider_specific_fields["reasoning_content"]` if present + (LiteLLM stores provider-specific response fields there). + 2. Otherwise inject a single space — the minimum value the API accepts. + """ + result: List[AllMessageValues] = [] + for msg in messages: + if msg.get("role") == "assistant" and not msg.get("reasoning_content"): + patched = dict(cast(dict, msg)) + provider_fields = patched.get("provider_specific_fields") or {} + stored = provider_fields.get("reasoning_content") + if stored: + patched["reasoning_content"] = stored + cleaned = dict(provider_fields) + cleaned.pop("reasoning_content", None) + patched["provider_specific_fields"] = cleaned + else: + litellm.verbose_logger.warning( + "DeepSeek thinking mode: assistant message is missing " + "`reasoning_content` and none was saved in " + "`provider_specific_fields`. A single-space placeholder " + "is being injected to satisfy API validation, but the " + "model will receive a blank reasoning chain for this turn, " + "which may silently degrade multi-turn response quality. " + "Preserve `reasoning_content` from the original assistant " + "response when building multi-turn conversation history." + ) + patched["reasoning_content"] = " " + result.append(cast(AllMessageValues, patched)) + else: + result.append(msg) + return result + @overload def _transform_messages( self, messages: List[AllMessageValues], model: str, is_async: Literal[True] @@ -91,6 +135,66 @@ class DeepSeekChatConfig(OpenAIGPTConfig): messages=messages, model=model, is_async=False ) + def _thinking_mode_active(self, model: str, optional_params: dict) -> bool: + """ + Returns True only when thinking mode is actually active for this request: + - model supports reasoning (capability check) + - user explicitly passed thinking={"type": "enabled"} (opt-in check) + """ + return ( + supports_reasoning(model=model, custom_llm_provider="deepseek") + and (optional_params.get("thinking") or {}).get("type") == "enabled" + ) + + def transform_request( + self, + model: str, + messages: List[AllMessageValues], + optional_params: dict, + litellm_params: dict, + headers: dict, + ) -> dict: + """ + Ensures `reasoning_content` is forwarded on assistant messages for + multi-turn thinking-mode conversations (issue #28045). + + Only runs when thinking mode is actually active - guarded by both + supports_reasoning() (model capability) and optional_params["thinking"] + (user explicitly enabled it), preventing spurious injection on models + like deepseek-v3.2 that support thinking as opt-in but not always-on. + """ + if self._thinking_mode_active(model=model, optional_params=optional_params): + messages = self._fill_reasoning_content(messages) + return super().transform_request( + model=model, + messages=messages, + optional_params=optional_params, + litellm_params=litellm_params, + headers=headers, + ) + + async def async_transform_request( + self, + model: str, + messages: List[AllMessageValues], + optional_params: dict, + litellm_params: dict, + headers: dict, + ) -> dict: + """ + Async equivalent of transform_request — applies the same reasoning_content + fix for multi-turn thinking-mode conversations. + """ + if self._thinking_mode_active(model=model, optional_params=optional_params): + messages = self._fill_reasoning_content(messages) + return await super().async_transform_request( + model=model, + messages=messages, + optional_params=optional_params, + litellm_params=litellm_params, + headers=headers, + ) + def _get_openai_compatible_provider_info( self, api_base: Optional[str], api_key: Optional[str] ) -> Tuple[Optional[str], Optional[str]]: diff --git a/litellm/llms/deepseek/cost_calculator.py b/litellm/llms/deepseek/cost_calculator.py index 0f4490cb3df..e652ebeac54 100644 --- a/litellm/llms/deepseek/cost_calculator.py +++ b/litellm/llms/deepseek/cost_calculator.py @@ -1,5 +1,5 @@ """ -Cost calculator for DeepSeek Chat models. +Cost calculator for DeepSeek Chat models. Handles prompt caching scenario. """ diff --git a/litellm/llms/deepseek/messages/transformation.py b/litellm/llms/deepseek/messages/transformation.py new file mode 100644 index 00000000000..ad60478960e --- /dev/null +++ b/litellm/llms/deepseek/messages/transformation.py @@ -0,0 +1,133 @@ +""" +DeepSeek Anthropic-compatible messages transformation config. +""" + +from typing import Any, Dict, List, Optional, Tuple + +import litellm +from litellm.llms.anthropic.experimental_pass_through.messages.transformation import ( + AnthropicMessagesConfig, +) +from litellm.secret_managers.main import get_secret_str +from litellm.types.router import GenericLiteLLMParams + + +class DeepSeekAnthropicMessagesConfig(AnthropicMessagesConfig): + """ + DeepSeek exposes an Anthropic-compatible Messages API at + https://api.deepseek.com/anthropic. + + It accepts the native Anthropic Messages conversation shape, including + thinking blocks in assistant history, but rejects Anthropic's explicit + custom-tool discriminator (`{"type": "custom"}`). + """ + + @property + def custom_llm_provider(self) -> Optional[str]: + return "deepseek" + + @staticmethod + def get_api_key(api_key: Optional[str] = None) -> Optional[str]: + return api_key or get_secret_str("DEEPSEEK_API_KEY") or litellm.api_key + + @staticmethod + def get_api_base(api_base: Optional[str] = None) -> str: + return ( + api_base + or get_secret_str("DEEPSEEK_ANTHROPIC_API_BASE") + or get_secret_str("DEEPSEEK_API_BASE") + or "https://api.deepseek.com/anthropic" + ) + + def validate_anthropic_messages_environment( + self, + headers: dict, + model: str, + messages: List[Any], + optional_params: dict, + litellm_params: dict, + api_key: Optional[str] = None, + api_base: Optional[str] = None, + ) -> Tuple[dict, Optional[str]]: + dynamic_api_key = self.get_api_key(api_key=api_key) + + if ( + "x-api-key" not in headers + and "authorization" not in headers + and dynamic_api_key is not None + ): + headers["x-api-key"] = dynamic_api_key + + if "anthropic-version" not in headers: + headers["anthropic-version"] = "2023-06-01" + if "content-type" not in headers: + headers["content-type"] = "application/json" + + headers = self._update_headers_with_anthropic_beta( + headers=headers, + optional_params=optional_params, + custom_llm_provider=self.custom_llm_provider or "deepseek", + ) + + return headers, api_base + + def get_complete_url( + self, + api_base: Optional[str], + api_key: Optional[str], + model: str, + optional_params: dict, + litellm_params: dict, + stream: Optional[bool] = None, + ) -> str: + base_url = self.get_api_base(api_base=api_base).rstrip("/") + + if base_url.endswith("/v1/messages") and "/anthropic/" in base_url: + return base_url + if base_url.endswith("/v1/messages"): + base_url = base_url[: -len("/v1/messages")] + if base_url.endswith("/v1"): + base_url = base_url[: -len("/v1")] + if base_url.endswith("/beta"): + base_url = base_url[: -len("/beta")] + + if not base_url.endswith("/anthropic") and "/anthropic/" not in base_url: + base_url = f"{base_url}/anthropic" + + return f"{base_url}/v1/messages" + + @staticmethod + def _sanitize_tools_for_deepseek(tools: Any) -> Any: + if not isinstance(tools, list): + return tools + + sanitized_tools = [] + for tool in tools: + if isinstance(tool, dict) and tool.get("type") == "custom": + sanitized_tool = dict(tool) + sanitized_tool.pop("type", None) + sanitized_tools.append(sanitized_tool) + else: + sanitized_tools.append(tool) + return sanitized_tools + + def transform_anthropic_messages_request( + self, + model: str, + messages: List[Dict], + anthropic_messages_optional_request_params: Dict, + litellm_params: GenericLiteLLMParams, + headers: dict, + ) -> Dict: + anthropic_messages_request = super().transform_anthropic_messages_request( + model=model, + messages=messages, + anthropic_messages_optional_request_params=anthropic_messages_optional_request_params, + litellm_params=litellm_params, + headers=headers, + ) + if "tools" in anthropic_messages_request: + anthropic_messages_request["tools"] = self._sanitize_tools_for_deepseek( + anthropic_messages_request["tools"] + ) + return anthropic_messages_request diff --git a/litellm/llms/elevenlabs/text_to_speech/transformation.py b/litellm/llms/elevenlabs/text_to_speech/transformation.py index 6a59911701b..612fc687ef9 100644 --- a/litellm/llms/elevenlabs/text_to_speech/transformation.py +++ b/litellm/llms/elevenlabs/text_to_speech/transformation.py @@ -22,7 +22,6 @@ from litellm.types.utils import all_litellm_params from ..common_utils import ElevenLabsException - if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.types.llms.openai import HttpxBinaryResponseContent diff --git a/litellm/llms/fireworks_ai/chat/transformation.py b/litellm/llms/fireworks_ai/chat/transformation.py index ed6d167a118..d39adf0b6f4 100644 --- a/litellm/llms/fireworks_ai/chat/transformation.py +++ b/litellm/llms/fireworks_ai/chat/transformation.py @@ -4,6 +4,7 @@ from typing import Any, List, Literal, Optional, Tuple, Union, cast import httpx import litellm +from litellm._logging import verbose_logger from litellm._uuid import uuid from litellm.constants import RESPONSE_FORMAT_TOOL_NAME from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj @@ -26,6 +27,7 @@ from litellm.types.utils import ( ProviderSpecificModelInfo, ) from litellm.utils import ( + get_model_cost_mutation_generation, supports_function_calling, supports_reasoning, supports_tool_choice, @@ -112,6 +114,19 @@ class FireworksAIConfig(OpenAIGPTConfig): # Only add tools for models that support function calling if supports_function_calling(model=model, custom_llm_provider="fireworks_ai"): supported_params.append("tools") + supported_params.append("parallel_tool_calls") + else: + # Historically every Fireworks model advertised tool support, so a + # JSON entry that flips `supports_function_calling` to false will + # silently drop `tools` from requests. Surface this so users can + # tell why their tool calls suddenly stop working. + verbose_logger.debug( + "fireworks_ai model %r is marked as not supporting " + "function calling in model_prices_and_context_window.json; " + "`tools` and `parallel_tool_calls` will be dropped from the " + "request.", + model, + ) # Only add tool_choice for models that explicitly support it if supports_tool_choice(model=model, custom_llm_provider="fireworks_ai"): @@ -241,41 +256,110 @@ class FireworksAIConfig(OpenAIGPTConfig): disable_add_transform_inline_image_block=disable_add_transform_inline_image_block, ) filter_value_from_dict(cast(dict, message), "cache_control") - # Remove fields not permitted by FireworksAI that may cause: - # "Not permitted, field: 'messages[n].provider_specific_fields'" - if isinstance(message, dict) and "provider_specific_fields" in message: - cast(dict, message).pop("provider_specific_fields", None) + # Remove fields not permitted by FireworksAI (additionalProperties: false + # on their ChatMessage schema) that may cause: + # "Extra inputs are not permitted, field: 'messages[n].'" + if isinstance(message, dict): + m = cast(dict, message) + m.pop("provider_specific_fields", None) + m.pop("thinking_blocks", None) return messages - def get_provider_info(self, model: str) -> ProviderSpecificModelInfo: - # Models that support reasoning_effort - reasoning_supported_models = [ - "qwen3-8b", - "qwen3-32b", - "qwen3-coder-480b-a35b-instruct", - "deepseek-v3p1", - "deepseek-v3p2", - "glm-4p5", - "glm-4p5-air", - "glm-4p6", - "gpt-oss-120b", - "gpt-oss-20b", + # Cached index of fireworks_ai/* entries from litellm.model_cost. Building + # this index requires a full scan of model_cost (tens of thousands of + # entries), so we memoize it. The cache key is (id(model_cost), + # mutation_generation): the generation counter is bumped on every + # register_model / reload path, so add+remove or in-place value + # replacement (which can leave id and len unchanged) still invalidates. + _fireworks_index_cache: Optional[Tuple[int, int, List[Tuple[str, dict]]]] = None + + @classmethod + def _get_fireworks_index(cls) -> List[Tuple[str, dict]]: + model_cost = litellm.model_cost + signature = (id(model_cost), get_model_cost_mutation_generation()) + cached = cls._fireworks_index_cache + if ( + cached is not None + and cached[0] == signature[0] + and cached[1] == signature[1] + ): + return cached[2] + + index: List[Tuple[str, dict]] = [] + for key, model_info in model_cost.items(): + if not key.startswith("fireworks_ai/"): + continue + if not isinstance(model_info, dict): + continue + key_short = key[len("fireworks_ai/") :] + if key_short.startswith("accounts/fireworks/models/"): + key_short = key_short[len("accounts/fireworks/models/") :] + if not key_short: + continue + index.append((key_short, model_info)) + + cls._fireworks_index_cache = (signature[0], signature[1], index) + return index + + @staticmethod + def _matches_on_hyphen_boundary(short_name: str, key_short: str) -> bool: + """Return True if `key_short` appears in `short_name` aligned to + hyphen-separated word boundaries (or end-of-string). This avoids + spurious substring matches like `"some-model"` matching + `"awesome-model"`.""" + if short_name == key_short: + return True + if short_name.startswith(key_short + "-"): + return True + if short_name.endswith("-" + key_short): + return True + return ("-" + key_short + "-") in short_name + + def _get_model_cost_capability(self, model: str, capability: str) -> Optional[bool]: + short_name = model + if short_name.startswith("fireworks_ai/"): + short_name = short_name[len("fireworks_ai/") :] + if short_name.startswith("accounts/fireworks/models/"): + short_name = short_name[len("accounts/fireworks/models/") :] + + candidate_keys = [ + model, + f"fireworks_ai/{short_name}", + f"fireworks_ai/accounts/fireworks/models/{short_name}", ] - # Normalize model name - remove prefix if present - normalized_model = model - if model.startswith("fireworks_ai/"): - normalized_model = model.replace("fireworks_ai/", "") - if normalized_model.startswith("accounts/fireworks/models/"): - normalized_model = normalized_model.replace( - "accounts/fireworks/models/", "" - ) + for candidate_key in candidate_keys: + model_info = litellm.model_cost.get(candidate_key) + if model_info is not None and model_info.get(capability) is not None: + return cast(Optional[bool], model_info.get(capability)) - # Check if model supports reasoning - supports_reasoning_value = any( - reasoning_model in normalized_model - for reasoning_model in reasoning_supported_models + # Fallback: preserve historical substring matching for model name + # variants (e.g. fine-tuned or regionally-suffixed versions of a + # known model). Pick the *longest* matching entry so a more specific + # known model (e.g. "qwen3-8b-instruct") wins over a less specific + # one (e.g. "qwen3-8b") when the query model is more specific still. + # Use hyphen-aligned matching to avoid false positives where a short + # known model name is an unrelated substring of a longer one. + best_match_short: Optional[str] = None + best_match_value: Optional[bool] = None + for key_short, model_info in self._get_fireworks_index(): + if model_info.get(capability) is None: + continue + if not self._matches_on_hyphen_boundary(short_name, key_short): + continue + if best_match_short is None or len(key_short) > len(best_match_short): + best_match_short = key_short + best_match_value = cast(Optional[bool], model_info.get(capability)) + + return best_match_value + + def get_provider_info(self, model: str) -> ProviderSpecificModelInfo: + supports_function_calling_value = self._get_model_cost_capability( + model=model, capability="supports_function_calling" + ) + supports_reasoning_value = self._get_model_cost_capability( + model=model, capability="supports_reasoning" ) provider_specific_model_info: ProviderSpecificModelInfo = { @@ -285,9 +369,16 @@ class FireworksAIConfig(OpenAIGPTConfig): "supports_vision": True, # via document inlining } + if supports_function_calling_value is not None: + provider_specific_model_info["supports_function_calling"] = ( + supports_function_calling_value + ) + # Only include supports_reasoning if True if supports_reasoning_value: - provider_specific_model_info["supports_reasoning"] = True + provider_specific_model_info["supports_reasoning"] = ( + supports_reasoning_value + ) return provider_specific_model_info diff --git a/litellm/llms/gemini/agents/__init__.py b/litellm/llms/gemini/agents/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/litellm/llms/gemini/agents/transformation.py b/litellm/llms/gemini/agents/transformation.py new file mode 100644 index 00000000000..f6e0b95cf28 --- /dev/null +++ b/litellm/llms/gemini/agents/transformation.py @@ -0,0 +1,298 @@ +""" +Google AI Studio Agents API configuration. + +Proxies the Gemini v1beta Agents API: + POST /v1beta/agents create + GET /v1beta/agents list + GET /v1beta/agents/{name} get + DELETE /v1beta/agents/{name} delete + GET /v1beta/agents/{name}/versions list versions +""" + +from typing import Any, Dict, Optional, Tuple, Union + +import httpx + +from litellm._logging import verbose_logger +from litellm.llms.base_llm.agents.transformation import BaseAgentsAPIConfig +from litellm.llms.gemini.common_utils import GeminiError, GeminiModelInfo +from litellm.types.agents import ( + AgentCreateResponse, + AgentDeleteResult, + AgentListResponse, + AgentVersionsResponse, +) + +# Keys inside litellm_params that should be forwarded to the Gemini +# create-agent body verbatim. +_GEMINI_AGENT_BODY_KEYS = ("base_agent", "instructions", "base_environment") + +# LiteLLM-internal keys that must never be forwarded to Gemini. +_LITELLM_INTERNAL_KEYS = frozenset( + { + "custom_llm_provider", + "api_key", + "api_base", + "make_public", + "cost_per_query", + "input_cost_per_token", + "output_cost_per_token", + "require_trace_id_on_calls_to_agent", + "require_trace_id_on_calls_by_agent", + "max_iterations", + "max_budget_per_session", + "guardrails", + "is_public", + "agent_name", + "agent_id", + "agent_card_params", + "provider_agent_response", + } +) + + +class GeminiAgentsConfig(BaseAgentsAPIConfig): + """ + Configuration for the Google AI Studio (Gemini) native Agents API. + + Authentication uses x-goog-api-key, resolved from (in order): + 1. litellm_params["api_key"] + 2. GOOGLE_API_KEY env var + 3. GEMINI_API_KEY env var + """ + + @property + def api_version(self) -> str: + return "v1beta" + + def _base_url(self, api_base: Optional[str]) -> str: + return f"{GeminiModelInfo.get_api_base(api_base)}/{self.api_version}" + + # ------------------------------------------------------------------ # + # Shared helpers # + # ------------------------------------------------------------------ # + + def get_error_class( + self, + error_message: str, + status_code: int, + headers: Union[dict, httpx.Headers], + ) -> Exception: + return GeminiError( + message=error_message, + status_code=status_code, + headers=dict(headers), + ) + + def get_complete_url( + self, + api_base: Optional[str], + litellm_params: Dict[str, Any], + ) -> str: + return f"{self._base_url(api_base)}/agents" + + def validate_environment( + self, + headers: Dict[str, str], + litellm_params: Dict[str, Any], + ) -> Dict[str, str]: + headers = dict(headers) + headers["Content-Type"] = "application/json" + explicit_api_key = litellm_params.get("api_key") + # SECURITY: when the caller overrides ``api_base``, refuse to fall back + # to the process-wide GOOGLE_API_KEY / GEMINI_API_KEY env vars. Otherwise + # an authenticated proxy user could set ``api_base`` to an attacker- + # controlled host and have the proxy ship its shared Gemini key in the + # ``x-goog-api-key`` header. + if litellm_params.get("api_base") and not explicit_api_key: + raise ValueError( + "When overriding api_base for Gemini agents, you must also " + "supply an explicit api_key. Falling back to GOOGLE_API_KEY / " + "GEMINI_API_KEY env vars with a custom api_base is refused " + "to prevent leaking the shared provider key to arbitrary hosts." + ) + api_key = GeminiModelInfo.get_api_key(explicit_api_key) + if not api_key: + raise ValueError( + "Google API key is required. " + "Set GOOGLE_API_KEY or GEMINI_API_KEY, or pass api_key." + ) + headers["x-goog-api-key"] = api_key + return headers + + def _raise_for_status(self, raw_response: httpx.Response) -> None: + if not (200 <= raw_response.status_code < 300): + raise GeminiError( + message=raw_response.text, + status_code=raw_response.status_code, + headers=dict(raw_response.headers), + ) + + # ------------------------------------------------------------------ # + # CREATE # + # ------------------------------------------------------------------ # + + def transform_create_request( + self, + name: str, + litellm_params: Dict[str, Any], + ) -> Dict[str, Any]: + body: Dict[str, Any] = {"name": name} + for key in _GEMINI_AGENT_BODY_KEYS: + value = litellm_params.get(key) + if value is not None: + body[key] = value + verbose_logger.debug("GeminiAgentsConfig create body: %s", body) + return body + + def transform_create_response( + self, + raw_response: httpx.Response, + name: str, + ) -> AgentCreateResponse: + """ + Gemini returns: + {"id": "my-agent", "base_agent": "waverunner", + "system_instruction": "...", "base_environment": {...}} + """ + self._raise_for_status(raw_response) + try: + data: Dict[str, Any] = raw_response.json() + except Exception: + verbose_logger.warning( + "GeminiAgentsConfig: non-JSON create response (status=%d).", + raw_response.status_code, + ) + data = {"id": name} + # Gemini uses "id" as the identifier; normalise to both fields. + data.setdefault("id", name) + data.setdefault("name", data["id"]) + verbose_logger.debug("GeminiAgentsConfig create response: %s", data) + return AgentCreateResponse(**data) + + # ------------------------------------------------------------------ # + # LIST # + # ------------------------------------------------------------------ # + + def transform_list_request( + self, + api_base: Optional[str], + litellm_params: Dict[str, Any], + ) -> Tuple[str, Dict[str, Any]]: + url = f"{self._base_url(api_base)}/agents" + params: Dict[str, Any] = {} + if litellm_params.get("page_size"): + params["pageSize"] = litellm_params["page_size"] + if litellm_params.get("page_token"): + params["pageToken"] = litellm_params["page_token"] + return url, params + + def transform_list_response( + self, + raw_response: httpx.Response, + ) -> AgentListResponse: + self._raise_for_status(raw_response) + try: + data = raw_response.json() + except Exception: + data = {} + verbose_logger.debug("GeminiAgentsConfig list response: %s", data) + return AgentListResponse( + agents=data.get("agents", []), + next_page_token=data.get("nextPageToken"), + ) + + # ------------------------------------------------------------------ # + # GET # + # ------------------------------------------------------------------ # + + def transform_get_request( + self, + name: str, + api_base: Optional[str], + litellm_params: Dict[str, Any], + ) -> Tuple[str, Dict[str, Any]]: + url = f"{self._base_url(api_base)}/agents/{name}" + return url, {} + + def transform_get_response( + self, + raw_response: httpx.Response, + name: str, + ) -> AgentCreateResponse: + """Same shape as create response — Gemini returns "id" as identifier.""" + self._raise_for_status(raw_response) + try: + data = raw_response.json() + except Exception: + data = {"id": name} + data.setdefault("id", name) + data.setdefault("name", data["id"]) + verbose_logger.debug("GeminiAgentsConfig get response: %s", data) + return AgentCreateResponse(**data) + + # ------------------------------------------------------------------ # + # DELETE # + # ------------------------------------------------------------------ # + + def transform_delete_request( + self, + name: str, + api_base: Optional[str], + litellm_params: Dict[str, Any], + ) -> str: + return f"{self._base_url(api_base)}/agents/{name}" + + def transform_delete_response( + self, + raw_response: httpx.Response, + name: str, + ) -> AgentDeleteResult: + """Gemini returns an empty body ``{}`` with HTTP 200 on success.""" + self._raise_for_status(raw_response) + verbose_logger.debug( + "GeminiAgentsConfig delete (status=%d) agent '%s'", + raw_response.status_code, + name, + ) + return AgentDeleteResult(name=name, deleted=True) + + # ------------------------------------------------------------------ # + # LIST VERSIONS # + # ------------------------------------------------------------------ # + + def transform_list_versions_request( + self, + name: str, + api_base: Optional[str], + litellm_params: Dict[str, Any], + ) -> Tuple[str, Dict[str, Any]]: + url = f"{self._base_url(api_base)}/agents/{name}/versions" + params: Dict[str, Any] = {} + if litellm_params.get("page_size"): + params["pageSize"] = litellm_params["page_size"] + if litellm_params.get("page_token"): + params["pageToken"] = litellm_params["page_token"] + return url, params + + def transform_list_versions_response( + self, + raw_response: httpx.Response, + name: str, + ) -> AgentVersionsResponse: + """ + Gemini returns: + {"agentVersions": [{"agent": "waverunner", "name": "agents/.../versions/uuid", ...}]} + """ + self._raise_for_status(raw_response) + try: + data = raw_response.json() + except Exception: + data = {} + verbose_logger.debug( + "GeminiAgentsConfig list_versions response for '%s': %s", name, data + ) + return AgentVersionsResponse( + agent_versions=data.get("agentVersions", []), + next_page_token=data.get("nextPageToken"), + ) diff --git a/litellm/llms/gemini/chat/transformation.py b/litellm/llms/gemini/chat/transformation.py index 72569e5c6cd..b69b7e1913e 100644 --- a/litellm/llms/gemini/chat/transformation.py +++ b/litellm/llms/gemini/chat/transformation.py @@ -1,5 +1,7 @@ from typing import List, Optional, cast +import litellm + from litellm.litellm_core_utils.prompt_templates.factory import ( convert_generic_image_chunk_to_openai_image_obj, convert_to_anthropic_image_obj, @@ -101,7 +103,10 @@ class GoogleAIStudioGeminiConfig(VertexGeminiConfig): return supported_params def _transform_messages( - self, messages: List[AllMessageValues], model: Optional[str] = None + self, + messages: List[AllMessageValues], + model: Optional[str] = None, + litellm_params: Optional[dict] = None, ) -> List[ContentType]: """ Google AI Studio Gemini does not support HTTP/HTTPS URLs for files. @@ -141,14 +146,26 @@ class GoogleAIStudioGeminiConfig(VertexGeminiConfig): img_element["image_url"] = converted_image_url # type: ignore elif element.get("type") == "file": file_element = cast(ChatCompletionFileObject, element) - file_id = file_element["file"].get("file_id") + _file_field = file_element.get("file") + if _file_field is None: + raise litellm.BadRequestError( + message="Content block has type='file' but is missing the required 'file' field", + model=model, + llm_provider="gemini", + ) + file_id = _file_field.get("file_id") if file_id and ("http://" in file_id or "https://" in file_id): # Convert HTTP/HTTPS file URL to base64 data try: base64_data = convert_url_to_base64(file_id) - file_element["file"]["file_data"] = base64_data # type: ignore - file_element["file"].pop("file_id", None) # type: ignore + _file_field["file_data"] = base64_data # type: ignore + _file_field.pop("file_id", None) # type: ignore except Exception: # If conversion fails, leave as is and let the API handle it pass - return _gemini_convert_messages_with_history(messages=messages, model=model) + return _gemini_convert_messages_with_history( + messages=messages, + model=model, + litellm_params=litellm_params, + custom_llm_provider="gemini", + ) diff --git a/litellm/llms/gemini/google_genai/transformation.py b/litellm/llms/gemini/google_genai/transformation.py index 7c4c7dba626..ee201af7e1a 100644 --- a/litellm/llms/gemini/google_genai/transformation.py +++ b/litellm/llms/gemini/google_genai/transformation.py @@ -2,6 +2,7 @@ Transformation for Calling Google models in their native format. """ +from copy import deepcopy from typing import TYPE_CHECKING, Any, Dict, List, Literal, Optional, Tuple, Union, cast import httpx @@ -11,6 +12,10 @@ from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging from litellm.llms.base_llm.google_genai.transformation import ( BaseGoogleGenAIGenerateContentConfig, ) +from litellm.llms.vertex_ai.common_utils import ( + _build_vertex_schema, + supports_response_json_schema, +) from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import VertexLLM from litellm.types.router import GenericLiteLLMParams @@ -302,6 +307,52 @@ class GoogleGenAIConfig(BaseGoogleGenAIGenerateContentConfig, VertexLLM): litellm_params=litellm_params, ) + @staticmethod + def _normalize_response_schema( + generate_content_config_dict: Dict, model: str + ) -> None: + schema_key = next( + ( + k + for k in ("responseSchema", "response_schema") + if k in generate_content_config_dict + ), + None, + ) + json_schema_key = next( + ( + k + for k in ("responseJsonSchema", "response_json_schema") + if k in generate_content_config_dict + ), + None, + ) + + if schema_key is None: + return + + value = generate_content_config_dict[schema_key] + if not isinstance(value, dict): + return + + if supports_response_json_schema(model): + if json_schema_key is not None: + generate_content_config_dict.pop(schema_key) + return + generate_content_config_dict.pop(schema_key) + new_json_schema_key = ( + "response_json_schema" + if schema_key == "response_schema" + else "responseJsonSchema" + ) + generate_content_config_dict[new_json_schema_key] = value + else: + if json_schema_key is not None: + generate_content_config_dict.pop(json_schema_key) + generate_content_config_dict[schema_key] = _build_vertex_schema( + parameters=deepcopy(value), add_property_ordering=True + ) + def transform_generate_content_request( self, model: str, @@ -315,6 +366,8 @@ class GoogleGenAIConfig(BaseGoogleGenAIGenerateContentConfig, VertexLLM): GenerateContentRequestDict, ) + self._normalize_response_schema(generate_content_config_dict, model) + typed_generate_content_request = GenerateContentRequestDict( model=model, contents=contents, diff --git a/litellm/llms/gemini/interactions/transformation.py b/litellm/llms/gemini/interactions/transformation.py index 593cbf7c2cf..b18b6a28ce4 100644 --- a/litellm/llms/gemini/interactions/transformation.py +++ b/litellm/llms/gemini/interactions/transformation.py @@ -6,13 +6,18 @@ Per OpenAPI spec (https://ai.google.dev/static/api/interactions.openapi.json): - Get: GET https://generativelanguage.googleapis.com/{api_version}/interactions/{interaction_id} - Delete: DELETE https://generativelanguage.googleapis.com/{api_version}/interactions/{interaction_id} -This is a thin wrapper - no transformation needed since we follow the spec directly. +Schema versioning: +- Default (Api-Revision: 2026-05-20): new `steps` schema. +- Legacy (Api-Revision: 2026-05-07): old `outputs` schema, controlled via + litellm.use_legacy_interactions_schema = True. Remove flag after June 8, 2026. """ from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple import httpx +import litellm + from litellm._logging import verbose_logger from litellm.litellm_core_utils.core_helpers import process_response_headers from litellm.litellm_core_utils.url_utils import encode_url_path_segment @@ -64,6 +69,7 @@ class GoogleAIStudioInteractionsConfig(BaseInteractionsAPIConfig): "stream", "store", "background", + "environment", "response_modalities", "response_format", "response_mime_type", @@ -83,6 +89,15 @@ class GoogleAIStudioInteractionsConfig(BaseInteractionsAPIConfig): api_key = GeminiModelInfo.get_api_key(litellm_params.get("api_key")) if api_key: headers["x-goog-api-key"] = api_key + + # Inject the Api-Revision header to select the response schema. + # Default to the new `steps` schema unless the operator has opted out. + # Remove this conditional after June 8, 2026 and always use 2026-05-20. + if litellm.use_legacy_interactions_schema: + headers["Api-Revision"] = "2026-05-07" + else: + headers["Api-Revision"] = "2026-05-20" + return headers def get_complete_url( @@ -118,8 +133,19 @@ class GoogleAIStudioInteractionsConfig(BaseInteractionsAPIConfig): headers: dict, ) -> Dict: """ - Build request body per OpenAPI spec - minimal transformation. + Build request body per OpenAPI spec. + + When on the new schema (use_legacy_interactions_schema=False, the default): + - ``response_mime_type`` is folded into ``response_format`` and stripped from + the body (the field was removed in Api-Revision 2026-05-20). + - ``generation_config.image_config`` is moved to a ``response_format`` entry + with ``"type": "image"`` (also removed from generation_config in 2026-05-20). + + When on the legacy schema (use_legacy_interactions_schema=True): + - All fields are forwarded as-is. """ + use_legacy: bool = litellm.use_legacy_interactions_schema + request_body: Dict[str, Any] = {} # Model or Agent (one required) @@ -134,23 +160,81 @@ class GoogleAIStudioInteractionsConfig(BaseInteractionsAPIConfig): if input is not None: request_body["input"] = input - # Pass through optional params directly (they match the spec) + # Pass through optional params — legacy schema keeps all fields as-is. optional_keys = [ "tools", "system_instruction", - "generation_config", "stream", "store", "background", + "environment", "response_modalities", - "response_format", - "response_mime_type", "previous_interaction_id", ] for key in optional_keys: if optional_params.get(key) is not None: request_body[key] = optional_params[key] + if use_legacy: + # Legacy schema: forward response_mime_type and response_format as-is. + for key in ("response_format", "response_mime_type", "generation_config"): + if optional_params.get(key) is not None: + request_body[key] = optional_params[key] + else: + # New schema (Api-Revision: 2026-05-20): + # response_mime_type is removed — fold it into response_format. + response_format = optional_params.get("response_format") + response_mime_type = optional_params.get("response_mime_type") + + if ( + response_mime_type + and not isinstance(response_format, list) + and ( + not isinstance(response_format, dict) + or "mime_type" not in response_format + ) + ): + # Wrap the legacy schema into the new polymorphic format. + new_rf: Dict[str, Any] = { + "type": "text", + "mime_type": response_mime_type, + } + if response_format is not None: + new_rf["schema"] = response_format + response_format = new_rf + + if response_format is not None: + request_body["response_format"] = response_format + + # image_config moves out of generation_config into response_format. + generation_config: Optional[Dict[str, Any]] = optional_params.get( + "generation_config" + ) + if generation_config is not None: + image_config = None + if isinstance(generation_config, dict): + generation_config = dict( + generation_config + ) # avoid mutating the caller's dict + image_config = generation_config.pop("image_config", None) + if not generation_config: + generation_config = None + + if generation_config is not None: + request_body["generation_config"] = generation_config + + if image_config is not None: + # Move image_config to response_format with type=image. + image_rf: Dict[str, Any] = {"type": "image", **image_config} + existing_rf = request_body.get("response_format") + if existing_rf is None: + request_body["response_format"] = image_rf + elif isinstance(existing_rf, list): + request_body["response_format"] = [*existing_rf, image_rf] + else: + # Convert single entry to array for multimodal output. + request_body["response_format"] = [existing_rf, image_rf] + return request_body def transform_response( diff --git a/litellm/llms/gemini/realtime/transformation.py b/litellm/llms/gemini/realtime/transformation.py index 4378db06358..cf1fc75ef10 100644 --- a/litellm/llms/gemini/realtime/transformation.py +++ b/litellm/llms/gemini/realtime/transformation.py @@ -3,8 +3,10 @@ This file contains the transformation logic for the Gemini realtime API. """ import json +from collections import OrderedDict from typing import Any, Dict, List, Optional, Union, cast +import litellm from litellm import verbose_logger from litellm._uuid import uuid from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj @@ -29,6 +31,7 @@ from litellm.types.llms.openai import ( OpenAIRealtimeDoneEvent, OpenAIRealtimeEvents, OpenAIRealtimeEventTypes, + OpenAIRealtimeFunctionCallArgumentsDone, OpenAIRealtimeOutputItemDone, OpenAIRealtimeResponseAudioDone, OpenAIRealtimeResponseContentPartAdded, @@ -36,10 +39,12 @@ from litellm.types.llms.openai import ( OpenAIRealtimeResponseDoneObject, OpenAIRealtimeResponseTextDone, OpenAIRealtimeStreamResponseBaseObject, + OpenAIRealtimeStreamResponseOutputItem, OpenAIRealtimeStreamResponseOutputItemAdded, OpenAIRealtimeStreamSession, OpenAIRealtimeStreamSessionEvents, OpenAIRealtimeTurnDetection, + ResponsesAPIStreamEvents, ) from litellm.types.llms.vertex_ai import ( GeminiResponseModalities, @@ -56,15 +61,43 @@ from litellm.utils import get_empty_usage from ..common_utils import encode_unserializable_types, get_api_key_from_env -MAP_GEMINI_FIELD_TO_OPENAI_EVENT: Dict[str, OpenAIRealtimeEventTypes] = { +MAP_GEMINI_FIELD_TO_OPENAI_EVENT: Dict[ + str, Union[OpenAIRealtimeEventTypes, ResponsesAPIStreamEvents] +] = { "setupComplete": OpenAIRealtimeEventTypes.SESSION_CREATED, "serverContent.generationComplete": OpenAIRealtimeEventTypes.RESPONSE_TEXT_DONE, "serverContent.turnComplete": OpenAIRealtimeEventTypes.RESPONSE_DONE, "serverContent.interrupted": OpenAIRealtimeEventTypes.RESPONSE_DONE, + "toolCall": ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DONE, +} + +# Top-level keys in a Gemini realtime message that map_openai_event knows how +# to handle. Other keys (e.g. ``usageMetadata``) can appear alongside these as +# siblings and must be skipped by the main transform loop — otherwise +# map_openai_event raises ``ValueError`` and the WebSocket session terminates. +_KNOWN_GEMINI_TOP_LEVEL_KEYS: set = { + map_key.split(".", 1)[0] for map_key in MAP_GEMINI_FIELD_TO_OPENAI_EVENT } class GeminiRealtimeConfig(BaseRealtimeConfig): + # Cap the LRU of in-flight tool calls so long sessions with many tool + # calls don't grow the dict without bound. Sized large enough to cover + # bursts of pending tool responses; the oldest entry is evicted when a + # new call beyond the cap arrives. + _TOOL_CALL_ID_TO_NAME_MAX = 256 + + def __init__(self): + super().__init__() + # Store call_id → function_name mapping for tool call round-trip + self._tool_call_id_to_name: "OrderedDict[str, str]" = OrderedDict() + # Buffer ``usageMetadata`` that Gemini Live emits as a standalone + # frame (between turns) so the next ``response.done`` attributes the + # tokens consumed. Without this an authenticated client can drive + # tool-call or normal turns whose token usage is recorded as zero, + # bypassing spend and budget accounting. + self._pending_usage_metadata: Optional[dict] = None + def validate_environment( self, headers: dict, model: str, api_key: Optional[str] = None ) -> dict: @@ -190,10 +223,9 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): ) vertex_gemini_config = VertexGeminiConfig() - optional_params["generationConfig"]["tools"] = ( - vertex_gemini_config._map_function( - value=value, optional_params=optional_params - ) + # Tools should be at the top level of setup, not inside generationConfig + optional_params["tools"] = vertex_gemini_config._map_function( + value=value, optional_params=optional_params ) elif key == "input_audio_transcription" and value is not None: optional_params["inputAudioTranscription"] = {} @@ -214,6 +246,272 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): optional_params.pop("generationConfig") return optional_params + @staticmethod + def _extract_turn_detection(session: dict) -> Optional[dict]: + """Extract turn_detection from a session.update payload. + + Handles both the flat beta shape (``session.turn_detection``) and the + GA shape (``session.audio.input.turn_detection``). + """ + if not isinstance(session, dict): + return None + td = session.get("turn_detection") + if isinstance(td, dict): + return td + audio = session.get("audio") + if isinstance(audio, dict): + input_cfg = audio.get("input") + if isinstance(input_cfg, dict): + td = input_cfg.get("turn_detection") + if isinstance(td, dict): + return td + return None + + @staticmethod + def _normalize_session_payload_for_mapping(session: dict) -> dict: + """Normalize GA-remapped session fields back to their beta keys. + + ``map_openai_params`` only recognises the flat OpenAI-beta key names + (``modalities``, ``input_audio_transcription``, ``turn_detection``). + For GA clients the upstream shim renames these into the nested GA + schema (``output_modalities``, ``audio.input.transcription``, + ``audio.input.turn_detection``), which would otherwise be silently + dropped here. Surface them back at the top level so the existing + mapping logic picks them up without duplicating provider-specific + knowledge of the GA schema in ``map_openai_params``. + """ + if not isinstance(session, dict): + return session + + normalized = dict(session) + + if "modalities" not in normalized and "output_modalities" in normalized: + normalized["modalities"] = normalized["output_modalities"] + + audio = normalized.get("audio") + if isinstance(audio, dict): + input_cfg = audio.get("input") + if isinstance(input_cfg, dict): + if ( + "input_audio_transcription" not in normalized + and "transcription" in input_cfg + ): + normalized["input_audio_transcription"] = input_cfg["transcription"] + + extracted_turn_detection = GeminiRealtimeConfig._extract_turn_detection( + normalized + ) + if extracted_turn_detection is not None and not isinstance( + normalized.get("turn_detection"), dict + ): + normalized["turn_detection"] = extracted_turn_detection + + return normalized + + def _handle_session_update( + self, + json_message: dict, + model: str, + session_configuration_request: Optional[str], + ) -> List[str]: + """ + Handle session.update by sending setup to Gemini. + + On the FIRST session.update (when session_configuration_request is None), + the full setup with all configuration is sent. + + Subsequent session.update messages are forwarded as a follow-up setup + with the new fields merged into the original setup. Gemini Live treats + a follow-up BidiGenerateContentSetup as a full session replacement + rather than a partial merge, so we carry forward the previous setup + (tools, generationConfig, inputAudioTranscription, systemInstruction, + ...) and overlay the new fields on top. This preserves the old + behavior where clients could refine the session via session.update + (e.g. add tools after the auto-setup on connect), and also keeps the + guardrail-driven turn_detection update working. + """ + session_payload = json_message.get("session") or {} + # Normalize GA-remapped fields (``output_modalities``, + # nested ``audio.input.transcription``, + # ``audio.input.turn_detection``) back to their flat beta keys so + # ``map_openai_params`` picks them up. Without this, GA clients' + # explicit modality / transcription / turn-detection settings + # would be silently dropped because ``map_openai_params`` only + # recognises the flat OpenAI-beta key names. + session_payload = self._normalize_session_payload_for_mapping(session_payload) + new_overrides = self.map_openai_params( + optional_params={}, non_default_params=session_payload + ) + + if session_configuration_request is None: + generation_config = new_overrides.setdefault("generationConfig", {}) + generation_config.setdefault("responseModalities", ["AUDIO"]) + new_overrides.setdefault("inputAudioTranscription", {}) + new_overrides["model"] = f"models/{model}" + verbose_logger.debug( + "Gemini Realtime: Sending initial setup with tools to backend" + ) + return [json.dumps({"setup": new_overrides})] + + if not new_overrides: + verbose_logger.debug( + "Gemini Realtime: Ignoring session.update (no mappable fields)" + ) + return [] + + try: + original_setup = cast( + BidiGenerateContentSetup, + json.loads(session_configuration_request).get("setup", {}), + ) + except (json.JSONDecodeError, AttributeError): + original_setup = {} + + # Deep-merge ``generationConfig`` and ``realtimeInputConfig`` so a + # partial session.update (e.g. only ``temperature`` or only + # ``modalities``) does not silently drop unrelated sub-keys + # (``responseModalities``, ``maxOutputTokens``, ...) from the original + # setup. + follow_up_setup: BidiGenerateContentSetup = { + **original_setup, + **new_overrides, + "model": f"models/{model}", + } + original_generation_config = original_setup.get("generationConfig") + new_generation_config = new_overrides.get("generationConfig") + if isinstance(original_generation_config, dict) and isinstance( + new_generation_config, dict + ): + follow_up_setup["generationConfig"] = { + **original_generation_config, + **new_generation_config, + } + original_realtime_input_config = original_setup.get("realtimeInputConfig") + new_realtime_input_config = new_overrides.get("realtimeInputConfig") + if isinstance(original_realtime_input_config, dict) and isinstance( + new_realtime_input_config, dict + ): + merged_realtime_input_config = { + **original_realtime_input_config, + **new_realtime_input_config, + } + # Deep-merge ``automaticActivityDetection`` so a partial VAD + # update (e.g. the guardrail-injected ``disabled: True`` from + # ``create_response: False``) does not silently drop unrelated + # knobs like ``silenceDurationMs`` / ``prefixPaddingMs`` from + # the original setup. + original_automatic_activity_detection = original_realtime_input_config.get( + "automaticActivityDetection" + ) + new_automatic_activity_detection = new_realtime_input_config.get( + "automaticActivityDetection" + ) + if isinstance(original_automatic_activity_detection, dict) and isinstance( + new_automatic_activity_detection, dict + ): + merged_realtime_input_config["automaticActivityDetection"] = { + **original_automatic_activity_detection, + **new_automatic_activity_detection, + } + follow_up_setup["realtimeInputConfig"] = cast( + BidiGenerateContentRealtimeInputConfig, + merged_realtime_input_config, + ) + verbose_logger.debug( + "Gemini Realtime: Forwarding session.update as follow-up setup" + ) + return [json.dumps({"setup": follow_up_setup})] + + def _handle_conversation_item(self, json_message: dict) -> List[str]: + """ + Handle conversation.item.create for user text or function call output. + + Converts OpenAI format to Gemini's clientContent (for user text) or + toolResponse (for function outputs). + """ + item = json_message.get("item", {}) + item_type = item.get("type") + + # Handle function call output (tool response) + if item_type == "function_call_output": + return self._handle_function_call_output(item) + + # Handle regular text content + return self._handle_user_text_content(item) + + def _handle_function_call_output(self, item: dict) -> List[str]: + """Transform function_call_output to Gemini toolResponse format.""" + call_id = item.get("call_id", "") + output = item.get("output", "{}") + + verbose_logger.debug( + f"Gemini Realtime: Transforming function_call_output for call_id={call_id}" + ) + + # Parse the output to get the result. Gemini's + # functionResponses[].response field is a Struct, so it must be a + # dict; wrap any non-dict (primitives, lists, invalid JSON) under a + # `result` key. + try: + parsed_output = json.loads(output) if isinstance(output, str) else output + except json.JSONDecodeError: + parsed_output = output + output_dict = ( + parsed_output + if isinstance(parsed_output, dict) + else {"result": parsed_output} + ) + + # Look up the function name from stored mapping. Keep the entry so a + # client SDK that retries function_call_output (or sends it twice for + # the same tool call) still produces a Gemini toolResponse with the + # required ``name`` field; refresh the LRU position so an active + # call_id stays warm across long sessions. + function_name = self._tool_call_id_to_name.get(call_id) + if function_name: + self._tool_call_id_to_name.move_to_end(call_id) + else: + verbose_logger.warning( + f"Gemini Realtime: Function name not found for call_id={call_id}. " + "This may cause Gemini to reject the response." + ) + + # Build Gemini toolResponse format + function_response = { + "id": call_id, + "response": output_dict, + } + if function_name: + function_response["name"] = function_name + + tool_response_message = { + "toolResponse": {"functionResponses": [function_response]} + } + + return [json.dumps(tool_response_message)] + + def _handle_user_text_content(self, item: dict) -> List[str]: + """Transform user text content to Gemini clientContent format.""" + content_list = item.get("content", []) + text_parts = [ + c.get("text", "") + for c in content_list + if isinstance(c, dict) and c.get("type") == "input_text" + ] + text = " ".join(filter(None, text_parts)) + if not text: + return [] + + # Build clientContent message with turns (proper Gemini Live API format) + client_content_message = { + "clientContent": { + "turns": [{"role": "user", "parts": [{"text": text}]}], + "turnComplete": True, + } + } + + return [json.dumps(client_content_message)] + def transform_realtime_request( self, message: str, @@ -233,55 +531,42 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): messages: List[str] = [] msg_type = json_message.get("type") - ## HANDLE SESSION UPDATE — translate to Gemini setup; no realtime_input needed ## + ## HANDLE SESSION UPDATE — translate to Gemini setup ## if msg_type == "session.update": - client_session_configuration_request = self.map_openai_params( - optional_params={}, non_default_params=json_message["session"] + return self._handle_session_update( + json_message, model, session_configuration_request ) - client_session_configuration_request["model"] = f"models/{model}" - messages.append(json.dumps({"setup": client_session_configuration_request})) - return messages ## HANDLE response.create — Gemini responds automatically; nothing to forward ## if msg_type == "response.create": return [] - ## HANDLE INPUT AUDIO BUFFER ## + ## HANDLE conversation.item.create — extract user text or function call output ## + if msg_type == "conversation.item.create": + return self._handle_conversation_item(json_message) + + ## HANDLE INPUT AUDIO BUFFER - use realtimeInput for audio streaming ## if msg_type == "input_audio_buffer.append": realtime_input_dict["audio"] = HttpxBlobType( mimeType=self.get_audio_mime_type(), data=json_message["audio"] ) - ## HANDLE conversation.item.create — extract actual user text ## - elif msg_type == "conversation.item.create": - item = json_message.get("item", {}) - content_list = item.get("content", []) - text_parts = [ - c.get("text", "") - for c in content_list - if isinstance(c, dict) and c.get("type") == "input_text" - ] - text = " ".join(filter(None, text_parts)) - if not text: - return [] - realtime_input_dict["text"] = text - else: - # Unknown/unsupported OpenAI event type — drop silently rather than - # forwarding raw JSON as text input to the model. - return [] - if len(realtime_input_dict) != 1: - raise ValueError( - f"Only one argument can be set, got {len(realtime_input_dict)}:" - f" {list(realtime_input_dict.keys())}" + realtime_input_dict = cast( + BidiGenerateContentRealtimeInput, + encode_unserializable_types( + cast(Dict[str, object], realtime_input_dict) + ), ) - realtime_input_dict = cast( - BidiGenerateContentRealtimeInput, - encode_unserializable_types(cast(Dict[str, object], realtime_input_dict)), - ) - - messages.append(json.dumps({"realtime_input": realtime_input_dict})) - return messages + gemini_msg = json.dumps({"realtimeInput": realtime_input_dict}) + verbose_logger.debug( + "Gemini Realtime: Sending audio realtimeInput to backend" + ) + messages.append(gemini_msg) + return messages + # Unknown/unsupported OpenAI event type — drop silently rather than + # forwarding raw JSON as text input to the model. + return [] def transform_session_created_event( self, @@ -300,7 +585,7 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): generation_config = ( session_configuration_request_dict.get("generationConfig", {}) or {} ) - gemini_modalities = generation_config.get("responseModalities", ["TEXT"]) + gemini_modalities = generation_config.get("responseModalities", ["AUDIO"]) _modalities = [ modality.lower() for modality in cast(List[str], gemini_modalities) ] @@ -352,18 +637,18 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): delta_type: ALL_DELTA_TYPES, session_configuration_request: Optional[str] = None, ) -> List[OpenAIRealtimeEvents]: - if session_configuration_request is None: - raise ValueError( - "session_configuration_request is required for Gemini API calls" - ) - - session_configuration_request_dict: BidiGenerateContentSetup = json.loads( - session_configuration_request - ).get("setup", {}) + session_configuration_request_dict: BidiGenerateContentSetup = {} + if session_configuration_request is not None: + try: + session_configuration_request_dict = json.loads( + session_configuration_request + ).get("setup", {}) + except json.JSONDecodeError: + session_configuration_request_dict = {} generation_config = session_configuration_request_dict.get( "generationConfig", {} ) - gemini_modalities = generation_config.get("responseModalities", ["TEXT"]) + gemini_modalities = generation_config.get("responseModalities", ["AUDIO"]) _modalities = [ modality.lower() for modality in cast(List[str], gemini_modalities) ] @@ -576,6 +861,86 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): returned_items.append(response_output_item_done) return returned_items + def _consume_usage_metadata_for_response_done(self, frame: dict) -> Optional[dict]: + """Return the ``usageMetadata`` to attribute to a ``response.done``. + + Gemini Live emits ``usageMetadata`` either alongside the closing + frame (``serverContent.turnComplete`` / ``toolCall``) or as a + standalone frame between turns. The standalone form would otherwise + be discarded by the no-op branch in ``transform_realtime_response`` + and the consumed tokens silently dropped from spend/budget + accounting. ``_pending_usage_metadata`` buffers any such standalone + frames so the next emitted ``response.done`` carries the deferred + token counts. + + Returns the in-frame ``usageMetadata`` if present (and clears the + buffer since the in-frame counts are the authoritative attribution + for this turn), otherwise returns the buffered counts. ``None`` is + returned when neither is available so the caller can fall back to + ``get_empty_usage()``. + """ + # ``pop`` (rather than ``get``) so a single Gemini frame containing + # multiple closing keys (e.g. both ``toolCall`` and + # ``serverContent.turnComplete``) cannot attribute the same + # ``usageMetadata`` to two ``response.done`` events and double-count + # tokens in spend/budget accounting. + in_frame = frame.pop("usageMetadata", None) if isinstance(frame, dict) else None + if isinstance(in_frame, dict): + self._pending_usage_metadata = None + return in_frame + buffered = self._pending_usage_metadata + self._pending_usage_metadata = None + return buffered + + def transform_tool_call_events( + self, + tool_call_message: dict, + response_id: Optional[str] = None, + output_item_id: Optional[str] = None, + ) -> List[OpenAIRealtimeFunctionCallArgumentsDone]: + """ + Transform Gemini toolCall message to OpenAI function call events. + + Converts Gemini's functionCalls format to OpenAI's response.function_call_arguments.done events. + Also stores call_id → name mapping for later use in function_call_output responses. + """ + function_calls = tool_call_message.get("functionCalls", []) + resolved_response_id = response_id or f"resp_{uuid.uuid4()}" + resolved_output_item_id = output_item_id or f"item_{uuid.uuid4()}" + + verbose_logger.debug( + f"Gemini Realtime: Transforming {len(function_calls)} tool call(s) to OpenAI format" + ) + + events: List[OpenAIRealtimeFunctionCallArgumentsDone] = [] + for idx, fc in enumerate(function_calls): + call_id = fc.get("id", "") + name = fc.get("name", "") + + # Store call_id → name mapping for round-trip. Use an LRU so + # repeated function_call_output lookups (retries) still hit, while + # sessions with many tool calls don't grow the dict unboundedly. + if call_id and name: + self._tool_call_id_to_name[call_id] = name + self._tool_call_id_to_name.move_to_end(call_id) + while len(self._tool_call_id_to_name) > self._TOOL_CALL_ID_TO_NAME_MAX: + self._tool_call_id_to_name.popitem(last=False) + + events.append( + OpenAIRealtimeFunctionCallArgumentsDone( + type="response.function_call_arguments.done", + event_id=f"event_{uuid.uuid4()}", + response_id=resolved_response_id, + item_id=f"{resolved_output_item_id}_tool_{idx}", + output_index=idx, + call_id=call_id, + name=name, + arguments=json.dumps(fc.get("args", {})), + ) + ) + + return events + @staticmethod def get_nested_value(obj: dict, path: str) -> Any: keys = path.split(".") @@ -681,14 +1046,20 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): "generationConfig", {} ) temperature = generation_config.get("temperature") - max_output_tokens = generation_config.get("max_output_tokens") - gemini_modalities = generation_config.get("responseModalities", ["TEXT"]) + max_output_tokens = generation_config.get("maxOutputTokens") + gemini_modalities = generation_config.get("responseModalities", ["AUDIO"]) _modalities = [ modality.lower() for modality in cast(List[str], gemini_modalities) ] - if "usageMetadata" in message: + resolved_usage_metadata = self._consume_usage_metadata_for_response_done( + cast(dict, message) + ) + if resolved_usage_metadata is not None: _chat_completion_usage = VertexGeminiConfig._calculate_usage( - completion_response=message, + completion_response=cast( + BidiGenerateContentServerMessage, + {**cast(dict, message), "usageMetadata": resolved_usage_metadata}, + ), ) else: _chat_completion_usage = get_empty_usage() @@ -716,7 +1087,9 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): if temperature is not None: response_done_event["response"]["temperature"] = temperature if max_output_tokens is not None: - response_done_event["response"]["max_output_tokens"] = max_output_tokens + response_done_event["response"]["max_output_tokens"] = cast( + int, max_output_tokens + ) return response_done_event @@ -808,13 +1181,18 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): def map_openai_event( self, key: str, - value: dict, + value: Any, current_delta_type: Optional[ALL_DELTA_TYPES], - json_message: dict, - ) -> OpenAIRealtimeEventTypes: - model_turn_event = value.get("modelTurn") - generation_complete_event = value.get("generationComplete") - openai_event: Optional[OpenAIRealtimeEventTypes] = None + ) -> Union[OpenAIRealtimeEventTypes, ResponsesAPIStreamEvents]: + if isinstance(value, dict): + model_turn_event = value.get("modelTurn") + generation_complete_event = value.get("generationComplete") + else: + model_turn_event = None + generation_complete_event = None + openai_event: Optional[ + Union[OpenAIRealtimeEventTypes, ResponsesAPIStreamEvents] + ] = None if model_turn_event: # check if model turn event openai_event = self.map_model_turn_event(model_turn_event) elif generation_complete_event: @@ -822,15 +1200,27 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): delta_type=current_delta_type ) else: - # Check if this key or any nested key matches our mapping - for map_key, openai_event in MAP_GEMINI_FIELD_TO_OPENAI_EVENT.items(): - if map_key == key or ( - "." in map_key - and GeminiRealtimeConfig.get_nested_value(json_message, map_key) - is not None - ): - openai_event = openai_event + # Check if this key or any nested key matches our mapping. Use a + # distinct loop variable so we don't shadow ``openai_event`` and + # leak the last dict value when no entry matches. Scope dotted-key + # lookups to the current ``key``/``value`` pair — checking the + # whole ``json_message`` would let a sibling key (e.g. + # ``serverContent.turnComplete``) misclassify the event currently + # being processed (e.g. ``toolCall``). + for map_key, candidate_event in MAP_GEMINI_FIELD_TO_OPENAI_EVENT.items(): + if map_key == key: + openai_event = candidate_event break + if "." in map_key: + prefix, _, nested_path = map_key.partition(".") + if ( + prefix == key + and isinstance(value, dict) + and GeminiRealtimeConfig.get_nested_value(value, nested_path) + is not None + ): + openai_event = candidate_event + break if openai_event is None: raise ValueError(f"Unknown openai event: {key}, value: {value}") return openai_event @@ -854,6 +1244,15 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): message_str = str(message) raise ValueError(f"Invalid JSON message: {message_str}") + verbose_logger.debug( + "Realtime Response Transform: Gemini frame keys=%s", + ( + sorted(json_message.keys()) + if isinstance(json_message, dict) + else type(json_message).__name__ + ), + ) + logging_session_id = logging_obj.litellm_trace_id current_output_item_id = realtime_response_transform_input[ @@ -913,32 +1312,44 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): ) # If serverContent only contained transcription(s) and no model - # content, return early — the main loop would fail on unknown keys. + # content, mark it as already handled so the main loop skips it + # (map_openai_event would raise on an unknown serverContent + # subkey). Fall through so sibling top-level keys such as + # ``toolCall`` are still processed in the main loop. _model_content_keys = { "modelTurn", "turnComplete", "interrupted", "generationComplete", } - if not any(k in server_content for k in _model_content_keys): - return { - "response": returned_message, - "current_output_item_id": current_output_item_id, - "current_response_id": current_response_id, - "current_delta_chunks": current_delta_chunks, - "current_conversation_id": current_conversation_id, - "current_item_chunks": current_item_chunks, - "current_delta_type": current_delta_type, - "session_configuration_request": session_configuration_request, - } + server_content_handled = not any( + k in server_content for k in _model_content_keys + ) + else: + server_content_handled = False - for key, value in json_message.items(): + tool_call_handled = False + # Snapshot the items so handlers below can safely mutate + # ``json_message`` (e.g. ``_consume_usage_metadata_for_response_done`` + # pops ``usageMetadata`` to prevent a single frame from attributing + # the same token counts to two ``response.done`` events). + for key, value in list(json_message.items()): + # Skip sibling metadata keys (e.g. ``usageMetadata``) that can + # accompany a primary payload like ``toolCall`` or ``serverContent``. + # ``map_openai_event`` raises ValueError on unknown keys, which + # would otherwise terminate the WebSocket session. + if key not in _KNOWN_GEMINI_TOP_LEVEL_KEYS: + continue + # serverContent was a transcription-only payload already emitted + # above; skip it here so map_openai_event doesn't raise on the + # missing model-content subkeys. + if key == "serverContent" and server_content_handled: + continue # Check if this key or any nested key matches our mapping openai_event = self.map_openai_event( key=key, value=value, current_delta_type=current_delta_type, - json_message=json_message, ) if openai_event == OpenAIRealtimeEventTypes.SESSION_CREATED: @@ -947,8 +1358,226 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): logging_session_id, realtime_response_transform_input["session_configuration_request"], ) - session_configuration_request = json.dumps(transformed_message) returned_message.append(transformed_message) + elif openai_event == ResponsesAPIStreamEvents.FUNCTION_CALL_ARGUMENTS_DONE: + # Handle toolCall from Gemini. If the payload has no function + # calls, emit nothing — an orphaned response.created/done pair + # with no output items would confuse OpenAI-compatible clients. + # Mark the key as intentionally consumed (mirroring + # ``server_content_handled``) so any sibling keys in the same + # frame are still processed by the rest of the loop and the + # post-loop guard doesn't treat the no-op as fatal. + if not value.get("functionCalls"): + tool_call_handled = True + continue + + if current_conversation_id is None: + current_conversation_id = f"conv_{uuid.uuid4()}" + + # Extract session-level response metadata once so both + # response.created and response.done can include matching + # modalities/temperature/max_output_tokens fields. + session_setup: BidiGenerateContentSetup = {} + if session_configuration_request is not None: + try: + session_setup = json.loads(session_configuration_request).get( + "setup", {} + ) + except (json.JSONDecodeError, TypeError): + session_setup = {} + tool_call_generation_config = ( + session_setup.get("generationConfig", {}) or {} + ) + tool_call_modalities = [ + modality.lower() + for modality in cast( + List[str], + tool_call_generation_config.get( + "responseModalities", ["AUDIO"] + ), + ) + ] + + # Emit response.created preamble if this is the first event in the response + if current_response_id is None: + current_response_id = f"resp_{uuid.uuid4()}" + current_output_item_id = f"item_{uuid.uuid4()}" + + # Mirror the audio/text path: include modalities, + # temperature, and max_output_tokens on response.created so + # spec-compliant clients see consistent response metadata + # regardless of whether the response starts with content or + # a tool call. + returned_message.append( + { + "type": "response.created", + "event_id": f"event_{uuid.uuid4()}", + "response": { + "object": "realtime.response", + "id": current_response_id, + "status": "in_progress", + "output": [], + "conversation_id": current_conversation_id, + "modalities": tool_call_modalities, + "temperature": tool_call_generation_config.get( + "temperature" + ), + "max_output_tokens": tool_call_generation_config.get( + "maxOutputTokens" + ), + }, + } + ) + + tool_call_events = self.transform_tool_call_events( + value, + response_id=current_response_id, + output_item_id=current_output_item_id, + ) + # Emit output_item.added and conversation.item.created for each function call + for idx, tool_call in enumerate(tool_call_events): + item_id = tool_call["item_id"] + function_call_item: OpenAIRealtimeStreamResponseOutputItem = { + "id": item_id, + "object": "realtime.item", + "type": "function_call", + "status": "completed", + "call_id": tool_call["call_id"], + "name": tool_call["name"], + "arguments": tool_call["arguments"], + } + # response.output_item.added + returned_message.append( + OpenAIRealtimeStreamResponseOutputItemAdded( + type="response.output_item.added", + event_id=f"event_{uuid.uuid4()}", + response_id=current_response_id, + output_index=idx, + item={ + **function_call_item, + "status": "in_progress", + "arguments": "", + }, + ) + ) + # response.function_call_arguments.delta — Gemini delivers + # the full arguments string in a single toolCall frame + # rather than streaming partial chunks, so emit one delta + # carrying the complete payload before the matching + # ``.done`` event. Spec-compliant OpenAI Realtime SDK + # clients accumulate ``delta.delta`` and rely on at least + # one delta before ``.done``. + returned_message.append( + cast( + OpenAIRealtimeEvents, + { + "type": "response.function_call_arguments.delta", + "event_id": f"event_{uuid.uuid4()}", + "response_id": current_response_id, + "item_id": item_id, + "output_index": idx, + "call_id": tool_call["call_id"], + "delta": tool_call["arguments"], + }, + ) + ) + # response.function_call_arguments.done + returned_message.append(tool_call) + # response.output_item.done — pass a fresh copy so + # downstream handlers that mutate the item dict (e.g. the + # beta-protocol translator) don't corrupt the references + # used by sibling events sharing the same function_call_item. + returned_message.append( + OpenAIRealtimeOutputItemDone( + type="response.output_item.done", + event_id=f"event_{uuid.uuid4()}", + response_id=current_response_id, + output_index=idx, + item={**function_call_item}, + ) + ) + # conversation.item.created + returned_message.append( + OpenAIRealtimeConversationItemCreated( + type="conversation.item.created", + event_id=f"event_{uuid.uuid4()}", + item={**function_call_item}, + ) + ) + + # response.done - close the response so clients can submit tool + # results. Mirror the non-tool-call RESPONSE_DONE path: if Gemini + # delivered ``usageMetadata`` alongside this ``toolCall`` frame, + # propagate the real token counts so spend/budget accounting + # records the tokens consumed by the tool-call turn. Standalone + # ``usageMetadata`` frames emitted in a separate WebSocket frame + # are buffered on the instance so the next ``response.done`` + # picks them up (otherwise an authenticated client could drive + # tool-call turns whose token usage is recorded as zero, + # bypassing budgets). Falls back to an empty usage block when + # neither is available (OpenAI-compatible clients expect + # ``usage`` to always be present on response.done). + resolved_tool_call_usage_metadata = ( + self._consume_usage_metadata_for_response_done(json_message) + ) + if resolved_tool_call_usage_metadata is not None: + _tool_call_chat_completion_usage = ( + VertexGeminiConfig._calculate_usage( + completion_response=cast( + BidiGenerateContentServerMessage, + { + **json_message, + "usageMetadata": resolved_tool_call_usage_metadata, + }, + ), + ) + ) + else: + _tool_call_chat_completion_usage = get_empty_usage() + tool_call_responses_api_usage = LiteLLMCompletionResponsesConfig._transform_chat_completion_usage_to_responses_usage( + _tool_call_chat_completion_usage, + ) + tool_call_done_event = OpenAIRealtimeDoneEvent( + type="response.done", + event_id=f"event_{uuid.uuid4()}", + response=OpenAIRealtimeResponseDoneObject( + id=current_response_id, + object="realtime.response", + status="completed", + output=[ + { + "id": te["item_id"], + "object": "realtime.item", + "type": "function_call", + "status": "completed", + "call_id": te["call_id"], + "name": te["name"], + "arguments": te["arguments"], + } + for te in tool_call_events + ], + conversation_id=current_conversation_id, + modalities=tool_call_modalities, + usage=tool_call_responses_api_usage.model_dump(), + ), + ) + tool_call_temperature = tool_call_generation_config.get("temperature") + if tool_call_temperature is not None: + tool_call_done_event["response"][ + "temperature" + ] = tool_call_temperature + tool_call_max_output_tokens = tool_call_generation_config.get( + "maxOutputTokens" + ) + if tool_call_max_output_tokens is not None: + tool_call_done_event["response"]["max_output_tokens"] = cast( + int, tool_call_max_output_tokens + ) + returned_message.append(tool_call_done_event) + # Reset IDs so the next model turn (after tool results) starts a + # fresh response with its own response.created preamble. + current_output_item_id = None + current_response_id = None elif openai_event == OpenAIRealtimeEventTypes.RESPONSE_DONE: transformed_response_done_event = self.transform_response_done_event( message=BidiGenerateContentServerMessage(**json_message), # type: ignore @@ -958,16 +1587,37 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): output_items=None, ) returned_message.append(transformed_response_done_event) + # Reset IDs so a subsequent turn (e.g. a `toolCall` arriving in + # a later WebSocket frame after `turnComplete`) starts a fresh + # response with its own `response.created` preamble instead of + # reusing the just-completed response ID. + current_output_item_id = None + current_response_id = None elif ( openai_event == OpenAIRealtimeEventTypes.RESPONSE_TEXT_DELTA or openai_event == OpenAIRealtimeEventTypes.RESPONSE_TEXT_DONE or openai_event == OpenAIRealtimeEventTypes.RESPONSE_AUDIO_DELTA or openai_event == OpenAIRealtimeEventTypes.RESPONSE_AUDIO_DONE ): + # Pass the locally-updated state (rather than the original + # input snapshot) so that prior iterations of this loop — + # e.g. a tool-call or response.done that just reset + # current_response_id/current_output_item_id to None — are + # honoured by the modality handler. + _modality_input: RealtimeResponseTransformInput = { + **realtime_response_transform_input, + "current_output_item_id": current_output_item_id, + "current_response_id": current_response_id, + "current_conversation_id": current_conversation_id, + "current_delta_chunks": current_delta_chunks, + "current_item_chunks": current_item_chunks, + "current_delta_type": current_delta_type, + "session_configuration_request": session_configuration_request, + } _returned_message = self.handle_openai_modality_event( openai_event, json_message, - realtime_response_transform_input, + _modality_input, delta_type="text" if "text" in openai_event.value else "audio", ) returned_message.extend(_returned_message["returned_message"]) @@ -979,6 +1629,41 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): else: raise ValueError(f"Unknown openai event: {openai_event}") if len(returned_message) == 0: + # A frame whose only top-level keys are sibling metadata (e.g. + # a standalone ``{"usageMetadata": {...}}`` emitted by Gemini + # Live between turns) is not an error — there is just nothing + # to forward to the OpenAI-shaped client. Returning the + # unchanged state keeps the WebSocket alive; raising would + # terminate the session for a benign no-op frame. + # serverContent already consumed by the transcription handler is + # a benign no-op for downstream — treat it like a metadata-only + # key when deciding whether to raise. + unhandled_known_keys = [ + key + for key in json_message + if key in _KNOWN_GEMINI_TOP_LEVEL_KEYS + and not (key == "serverContent" and server_content_handled) + and not (key == "toolCall" and tool_call_handled) + ] + # Buffer standalone usage metadata so the next response.done can + # attribute the token counts. Without this, an authenticated + # client driving turns whose usageMetadata is emitted in a + # separate frame would have those tokens recorded as zero spend, + # bypassing budget enforcement. + standalone_usage_metadata = json_message.get("usageMetadata") + if isinstance(standalone_usage_metadata, dict): + self._pending_usage_metadata = standalone_usage_metadata + if not unhandled_known_keys: + return { + "response": returned_message, + "current_output_item_id": current_output_item_id, + "current_response_id": current_response_id, + "current_delta_chunks": current_delta_chunks, + "current_conversation_id": current_conversation_id, + "current_item_chunks": current_item_chunks, + "current_delta_type": current_delta_type, + "session_configuration_request": session_configuration_request, + } if isinstance(message, bytes): message_str = message.decode("utf-8", errors="replace") else: @@ -993,6 +1678,13 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): transformed_message=returned_message, current_item_chunks=current_item_chunks, ) + + for msg in returned_message: + event_type = msg.get("type") if isinstance(msg, dict) else "unknown" + verbose_logger.debug( + "Realtime Response Transform: OpenAI event=%s", event_type + ) + return { "response": returned_message, "current_output_item_id": current_output_item_id, @@ -1005,7 +1697,10 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): } def requires_session_configuration(self) -> bool: - return True + # Default behavior is backwards-compatible: send setup on connect. + # Opt-in to deferred setup for tool-injection flow via: + # litellm.gemini_live_defer_setup = True + return not litellm.gemini_live_defer_setup def session_configuration_request(self, model: str) -> str: """ diff --git a/litellm/llms/gemini/videos/transformation.py b/litellm/llms/gemini/videos/transformation.py index c7116940b22..77a95bfa5ab 100644 --- a/litellm/llms/gemini/videos/transformation.py +++ b/litellm/llms/gemini/videos/transformation.py @@ -55,7 +55,7 @@ def _convert_image_to_gemini_format(image_file) -> Dict[str, str]: def _usage_video_resolution_from_parameters( - parameters: Dict[str, Any] + parameters: Dict[str, Any], ) -> Optional[str]: """Normalize Veo ``parameters.resolution`` for usage and cost tracking.""" res = parameters.get("resolution") @@ -581,12 +581,23 @@ class GeminiVideoConfig(BaseVideoConfig): raise NotImplementedError("video get character is not supported for Gemini") def transform_video_edit_request( - self, prompt, video_id, api_base, litellm_params, headers, extra_body=None + self, + prompt, + video_id, + api_base, + litellm_params, + headers, + extra_body=None, + prefetched_source_data=None, ): raise NotImplementedError("video edit is not supported for Gemini") def transform_video_edit_response( - self, raw_response, logging_obj, custom_llm_provider=None + self, + raw_response, + logging_obj, + custom_llm_provider=None, + request_data=None, ): raise NotImplementedError("video edit is not supported for Gemini") diff --git a/litellm/llms/infinity/rerank/transformation.py b/litellm/llms/infinity/rerank/transformation.py index 314bf2f8a36..b9804605454 100644 --- a/litellm/llms/infinity/rerank/transformation.py +++ b/litellm/llms/infinity/rerank/transformation.py @@ -1,5 +1,5 @@ """ -Transformation logic from Cohere's /v1/rerank format to Infinity's `/v1/rerank` format. +Transformation logic from Cohere's /v1/rerank format to Infinity's `/v1/rerank` format. Why separate file? Make it easy to see how transformation works """ diff --git a/litellm/llms/jina_ai/rerank/transformation.py b/litellm/llms/jina_ai/rerank/transformation.py index ad4416925a6..56be754fc34 100644 --- a/litellm/llms/jina_ai/rerank/transformation.py +++ b/litellm/llms/jina_ai/rerank/transformation.py @@ -1,5 +1,5 @@ """ -Transformation logic from Cohere's /v1/rerank format to Jina AI's `/v1/rerank` format. +Transformation logic from Cohere's /v1/rerank format to Jina AI's `/v1/rerank` format. Why separate file? Make it easy to see how transformation works diff --git a/litellm/llms/langgraph/chat/transformation.py b/litellm/llms/langgraph/chat/transformation.py index 00cc3a8f516..9808b665b54 100644 --- a/litellm/llms/langgraph/chat/transformation.py +++ b/litellm/llms/langgraph/chat/transformation.py @@ -139,14 +139,16 @@ class LangGraphConfig(BaseConfig): def _convert_messages_to_langgraph_format( self, messages: List[AllMessageValues] - ) -> List[Dict[str, str]]: + ) -> List[Dict[str, Any]]: """ Convert OpenAI-format messages to LangGraph format. OpenAI format: {"role": "user", "content": "..."} LangGraph format: {"role": "human", "content": "..."} + + Preserves per-message ``metadata`` when present (e.g. A2A ``skillId``). """ - langgraph_messages: List[Dict[str, str]] = [] + langgraph_messages: List[Dict[str, Any]] = [] for msg in messages: role = msg.get("role", "user") content = msg.get("content", "") @@ -169,7 +171,15 @@ class LangGraphConfig(BaseConfig): if not isinstance(content, str): content = str(content) - langgraph_messages.append({"role": langgraph_role, "content": content}) + langgraph_message: Dict[str, Any] = { + "role": langgraph_role, + "content": content, + } + message_metadata = msg.get("metadata") + if isinstance(message_metadata, dict) and message_metadata: + langgraph_message["metadata"] = message_metadata + + langgraph_messages.append(langgraph_message) return langgraph_messages diff --git a/litellm/llms/lm_studio/embed/transformation.py b/litellm/llms/lm_studio/embed/transformation.py index 1285550c30f..87f4f6e73d5 100644 --- a/litellm/llms/lm_studio/embed/transformation.py +++ b/litellm/llms/lm_studio/embed/transformation.py @@ -1,5 +1,5 @@ """ -Transformation logic from OpenAI /v1/embeddings format to LM Studio's `/v1/embeddings` format. +Transformation logic from OpenAI /v1/embeddings format to LM Studio's `/v1/embeddings` format. Why separate file? Make it easy to see how transformation works diff --git a/litellm/llms/novita/chat/transformation.py b/litellm/llms/novita/chat/transformation.py index c05d2d7b2c5..5a64a124ade 100644 --- a/litellm/llms/novita/chat/transformation.py +++ b/litellm/llms/novita/chat/transformation.py @@ -1,5 +1,5 @@ """ -Support for OpenAI's `/v1/chat/completions` endpoint. +Support for OpenAI's `/v1/chat/completions` endpoint. Calls done in OpenAI/openai.py as Novita AI is openai-compatible. diff --git a/litellm/llms/nvidia_nim/chat/transformation.py b/litellm/llms/nvidia_nim/chat/transformation.py index b8f8b04eb53..2ef92a90626 100644 --- a/litellm/llms/nvidia_nim/chat/transformation.py +++ b/litellm/llms/nvidia_nim/chat/transformation.py @@ -1,7 +1,7 @@ """ -Nvidia NIM endpoint: https://docs.api.nvidia.com/nim/reference/databricks-dbrx-instruct-infer +Nvidia NIM endpoint: https://docs.api.nvidia.com/nim/reference/databricks-dbrx-instruct-infer -This is OpenAI compatible +This is OpenAI compatible This file only contains param mapping logic diff --git a/litellm/llms/nvidia_nim/embed.py b/litellm/llms/nvidia_nim/embed.py index 24c6cc34e4d..61c8e8244e4 100644 --- a/litellm/llms/nvidia_nim/embed.py +++ b/litellm/llms/nvidia_nim/embed.py @@ -1,7 +1,7 @@ """ Nvidia NIM embeddings endpoint: https://docs.api.nvidia.com/nim/reference/nvidia-nv-embedqa-e5-v5-infer -This is OpenAI compatible +This is OpenAI compatible This file only contains param mapping logic diff --git a/litellm/llms/oci/chat/cohere.py b/litellm/llms/oci/chat/cohere.py new file mode 100644 index 00000000000..ac92fd22aa8 --- /dev/null +++ b/litellm/llms/oci/chat/cohere.py @@ -0,0 +1,386 @@ +""" +OCI Generative AI — Cohere-specific chat transformation helpers. + +Handles message history building, tool definition adaptation, non-streaming +response parsing, and streaming chunk parsing for models served with +``apiFormat="COHERE"`` (e.g. ``cohere.command-*``). +""" + +import datetime +import json +from typing import Any, Dict, List, Optional + +import httpx +from pydantic import ValidationError + +from litellm.llms.oci.chat.generic import ( + _normalize_oci_finish_reason, + _synthesize_oci_tool_call_id, +) +from litellm.llms.oci.common_utils import ( + OCI_JSON_TO_PYTHON_TYPES, + OCIError, + enrich_cohere_param_description, + resolve_oci_schema_anyof, + resolve_oci_schema_refs, + sanitize_oci_schema, +) +from litellm.types.llms.oci import ( + CohereChatResult, + CohereMessage, + CohereParameterDefinition, + CohereStreamChunk, + CohereTool, + CohereToolCall, + CohereToolMessage, + CohereToolResult, +) +from litellm.types.llms.openai import AllMessageValues +from litellm.types.utils import ( + Choices, + Delta, + ModelResponse, + ModelResponseStream, + StreamingChoices, +) +from litellm.types.utils import Usage + + +def _extract_text_content(content: Any) -> str: + """Return the plain-text representation of a message content value.""" + if content is None: + return "" + if isinstance(content, str): + return content + if isinstance(content, list): + return "".join( + item.get("text", "") + for item in content + if isinstance(item, dict) and item.get("type") == "text" + ) + return str(content) + + +def adapt_messages_to_cohere_standard( + messages: List[AllMessageValues], +) -> List[CohereMessage]: + """Build a Cohere ``chatHistory`` list from an OpenAI-format message array. + + - All messages except the *last user message* are included. The caller pulls + the last user message into the request's top-level ``message`` field, so + trailing tool results (the standard agentic continuation pattern) still + appear in ``chatHistory`` and reach the model. + - If no user message exists, every message is included (no slice). + - System messages must be filtered out by the caller (they are routed into + ``preambleOverride`` separately) — they are not represented in + ``chatHistory``. + - Tool results are expressed as OCI ``CohereToolMessage.toolResults`` entries, + with the originating call's name and parameters resolved from the preceding + assistant message via a ``tool_call_id`` lookup. + """ + # First pass: build tool_call_id → CohereToolCall so tool-result messages can + # reference the originating call by name and parameters. + tool_call_lookup: Dict[str, CohereToolCall] = {} + for msg in messages: + if msg.get("role") == "assistant": + tool_calls_raw: Any = msg.get("tool_calls") or [] + for tc in tool_calls_raw: + tc_id = tc.get("id", "") + raw_args: Any = tc.get("function", {}).get("arguments", "{}") + try: + params: Dict[str, Any] = ( + json.loads(raw_args) if isinstance(raw_args, str) else raw_args + ) + except json.JSONDecodeError: + params = {} + tool_call_lookup[tc_id] = CohereToolCall( + name=str(tc.get("function", {}).get("name", "")), + parameters=params, + ) + + last_user_index = next( + ( + i + for i in range(len(messages) - 1, -1, -1) + if messages[i].get("role") == "user" + ), + None, + ) + history_source = ( + messages + if last_user_index is None + else [m for i, m in enumerate(messages) if i != last_user_index] + ) + + chat_history: List[CohereMessage] = [] + for msg in history_source: + role = msg.get("role") + content = _extract_text_content(msg.get("content")) + + tool_calls: Optional[List[CohereToolCall]] = None + if role == "assistant" and msg.get("tool_calls"): # type: ignore[union-attr,typeddict-item] + tool_calls = [] + for tc in msg["tool_calls"]: # type: ignore[union-attr,typeddict-item] + raw_arguments: Any = tc.get("function", {}).get("arguments", {}) + if isinstance(raw_arguments, str): + try: + arguments: Dict[str, Any] = json.loads(raw_arguments) + except json.JSONDecodeError: + arguments = {} + else: + arguments = raw_arguments + tool_calls.append( + CohereToolCall( + name=str(tc.get("function", {}).get("name", "")), + parameters=arguments, + ) + ) + + if role == "user": + chat_history.append(CohereMessage(role="USER", message=content)) + elif role == "assistant": + chat_history.append( + CohereMessage(role="CHATBOT", message=content, toolCalls=tool_calls) + ) + elif role == "tool": + tool_call_id = str(msg.get("tool_call_id", "") or "") + cohere_call = tool_call_lookup.get( + tool_call_id, CohereToolCall(name="", parameters={}) + ) + tool_result = CohereToolResult( + call=cohere_call, + outputs=[{"output": content}], + ) + # OpenAI emits one tool-role message per parallel tool call, but + # the OCI Cohere API expects all results from a single assistant + # turn to share one TOOL history entry with multiple toolResults. + # Merge consecutive tool messages so the model sees the parallel + # call/result pairing correctly during agentic loops. + if chat_history and isinstance(chat_history[-1], CohereToolMessage): + chat_history[-1].toolResults.append(tool_result) + else: + chat_history.append(CohereToolMessage(toolResults=[tool_result])) + + return chat_history + + +def adapt_tool_definitions_to_cohere_standard( + tools: List[Dict[str, Any]], +) -> List[CohereTool]: + """Adapt OpenAI-format tool definitions to the OCI Cohere format. + + - Resolves ``$ref``/``$defs`` and ``anyOf`` patterns that OCI rejects. + - Maps JSON Schema type names to Python type names (``"string"`` → ``"str"``). + - Embeds unsupported constraints (enum, format, range, pattern) into the + parameter description so the model can still see them. + """ + cohere_tools = [] + for tool in tools: + function_def = tool.get("function", {}) + raw_params = function_def.get("parameters", {}) + + resolved = sanitize_oci_schema( + resolve_oci_schema_anyof(resolve_oci_schema_refs(raw_params)) + ) + properties = resolved.get("properties", {}) + required = resolved.get("required", []) + + parameter_definitions = {} + for param_name, param_schema in properties.items(): + json_type = param_schema.get("type", "string") + python_type = OCI_JSON_TO_PYTHON_TYPES.get(json_type, json_type) + parameter_definitions[param_name] = CohereParameterDefinition( + description=enrich_cohere_param_description( + param_schema.get("description", ""), param_schema + ), + type=python_type, + isRequired=param_name in required, + ) + + cohere_tools.append( + CohereTool( + name=function_def.get("name", ""), + description=function_def.get("description", ""), + parameterDefinitions=parameter_definitions, + ) + ) + + return cohere_tools + + +def handle_cohere_response( + json_response: dict, + model: str, + model_response: ModelResponse, + raw_response: httpx.Response, +) -> ModelResponse: + """Parse a non-streaming Cohere OCI response into a LiteLLM ModelResponse.""" + try: + cohere_response = CohereChatResult(**json_response) + except (TypeError, ValidationError) as e: + raise OCIError( + message=f"Response cannot be casted to CohereChatResult: {str(e)}", + status_code=raw_response.status_code, + ) + + model_response.model = model + model_response.created = int(datetime.datetime.now().timestamp()) + + response_text = cohere_response.chatResponse.text + finish_reason = _normalize_oci_finish_reason( + cohere_response.chatResponse.finishReason + ) + + tool_calls: Optional[List[Dict[str, Any]]] = None + if cohere_response.chatResponse.toolCalls: + tool_calls = [ + { + "id": _synthesize_oci_tool_call_id( + i, tc.name, json.dumps(tc.parameters, sort_keys=True) + ), + "type": "function", + "function": { + "name": tc.name, + "arguments": json.dumps(tc.parameters), + }, + } + for i, tc in enumerate(cohere_response.chatResponse.toolCalls) + ] + + content: Optional[str] = response_text if response_text else None + + # Only include ``tool_calls`` in the message dict when actually present. + # Passing an explicit ``None`` would let downstream consumers that key off + # ``"tool_calls" in message`` (rather than truthiness) incorrectly conclude + # that tool calls were attempted. Matches the generic handler's behaviour, + # which only sets ``message.tool_calls`` when tool calls are present. + message: Dict[str, Any] = {"role": "assistant", "content": content} + if tool_calls is not None: + message["tool_calls"] = tool_calls + + model_response.choices = [ + Choices( + index=0, + message=message, + finish_reason=finish_reason, + ) + ] + + usage_info = cohere_response.chatResponse.usage + if usage_info is not None: + model_response.usage = Usage( # type: ignore[attr-defined] + prompt_tokens=usage_info.promptTokens, + completion_tokens=usage_info.completionTokens, + total_tokens=usage_info.totalTokens, + ) + else: + model_response.usage = Usage(prompt_tokens=0, completion_tokens=0, total_tokens=0) # type: ignore[attr-defined] + + return model_response + + +def handle_cohere_stream_chunk( + dict_chunk: dict, + prior_tool_calls_emitted: bool = False, + prior_text_emitted: bool = False, +) -> ModelResponseStream: + """Parse a single Cohere SSE chunk into a LiteLLM ModelResponseStream. + + ``prior_tool_calls_emitted`` lets the caller signal whether tool calls + were already emitted in earlier chunks of the same stream. When set, the + terminal consolidation chunk's tool calls are suppressed (they would + duplicate prior deltas); otherwise they are passed through so a stream + that delivers tool calls only on the terminal chunk doesn't silently + drop them. + + ``prior_text_emitted`` plays the analogous role for the ``text`` field: + when set, the terminal consolidation chunk's ``text`` is suppressed + (it would re-emit the full assembled response on top of prior deltas); + when unset (e.g. a degenerate stream that delivers the entire response + in a single SSE event carrying both ``chatHistory`` and ``finishReason``), + the text is passed through so the response content isn't silently lost. + """ + try: + typed_chunk = CohereStreamChunk(**dict_chunk) + except (TypeError, ValidationError) as e: + raise OCIError( + status_code=500, + message=f"Chunk cannot be parsed as CohereStreamChunk: {str(e)}", + ) + + if typed_chunk.index is None: + typed_chunk.index = 0 + + # OCI Cohere's terminal SSE event re-sends the full assembled response in + # `text` alongside a populated `chatHistory` and a non-null `finishReason`. + # Emitting that text would concatenate the whole response onto the + # already-streamed deltas. We require both signals to be present so that a + # future API change which adds `chatHistory` to intermediate chunks (or a + # rare early-populated case) doesn't silently drop legitimate token deltas. + is_terminal_consolidation = ( + typed_chunk.chatHistory is not None and typed_chunk.finishReason is not None + ) + # On non-terminal text-free chunks (e.g. tool-call-only or keep-alive + # chunks) emit ``content=None`` rather than ``content=""`` so downstream + # stream-mergers that distinguish "no text in this delta" from "an + # explicitly empty text delta" behave correctly. + # + # We only suppress the terminal chunk's ``text`` when the caller has + # confirmed that text deltas were already emitted earlier — otherwise + # (e.g. a degenerate stream that delivers the whole response in a + # single SSE event), passing it through is the only chance to surface it. + text: Optional[str] = ( + None if (is_terminal_consolidation and prior_text_emitted) else typed_chunk.text + ) + + # Tool calls on the terminal consolidation chunk (whether from + # `typed_chunk.toolCalls` or from `chatHistory`) typically restate what + # was already streamed in intermediate chunks. Re-emitting them would + # mint fresh `uuid4` IDs and cause downstream consumers to execute each + # tool call twice. We only suppress when the caller has confirmed that + # tool calls were already emitted earlier — otherwise (e.g. a short + # response that delivers tool calls exclusively on the terminal chunk), + # passing them through is the only chance to surface them. + cohere_tool_calls = ( + None + if (is_terminal_consolidation and prior_tool_calls_emitted) + else typed_chunk.toolCalls + ) + + tool_calls: Optional[List[Dict[str, Any]]] = None + if cohere_tool_calls: + tool_calls = [ + { + # Cohere protocol has no tool-call id, so we synthesize one + # deterministically from the call's content/position. A random + # uuid4 per chunk would cause downstream stream-mergers to + # treat each chunk as a distinct tool call. + "id": _synthesize_oci_tool_call_id( + i, tc.name, json.dumps(tc.parameters, sort_keys=True) + ), + "type": "function", + "function": { + "name": tc.name, + "arguments": json.dumps(tc.parameters), + }, + } + for i, tc in enumerate(cohere_tool_calls) + ] + + finish_reason = _normalize_oci_finish_reason(typed_chunk.finishReason) + + return ModelResponseStream( + choices=[ + StreamingChoices( + index=typed_chunk.index, + delta=Delta( + content=text, + tool_calls=tool_calls, + provider_specific_fields=None, + thinking_blocks=None, + reasoning_content=None, + ), + finish_reason=finish_reason, + ) + ] + ) diff --git a/litellm/llms/oci/chat/generic.py b/litellm/llms/oci/chat/generic.py new file mode 100644 index 00000000000..2cc1ac77a40 --- /dev/null +++ b/litellm/llms/oci/chat/generic.py @@ -0,0 +1,477 @@ +""" +OCI Generative AI — Generic-format chat transformation helpers. + +Handles message building, tool definition adaptation, non-streaming response +parsing, and streaming chunk parsing for models served with +``apiFormat="GENERIC"`` (e.g. Meta Llama, xAI Grok, Google Gemini). +""" + +import datetime +import hashlib +from typing import Any, Dict, List, Optional, Union + +import httpx +from pydantic import ValidationError + +from litellm.llms.oci.common_utils import ( + OCIError, + resolve_oci_schema_anyof, + resolve_oci_schema_refs, + sanitize_oci_schema, +) +from litellm.types.llms.oci import ( + OCICompletionResponse, + OCIContentPartUnion, + OCIImageContentPart, + OCIImageUrl, + OCIMessage, + OCIRoles, + OCIStreamChunk, + OCITextContentPart, + OCIToolCall, + OCIToolDefinition, + OCIVendors, +) +from litellm.types.llms.openai import AllMessageValues +from litellm.types.utils import ( + Delta, + ModelResponse, + ModelResponseStream, + StreamingChoices, +) +from litellm.types.utils import ChatCompletionMessageToolCall, Usage + +# Maps OpenAI role names to OCI GENERIC role names. +open_ai_to_generic_oci_role_map: Dict[str, OCIRoles] = { + "system": "SYSTEM", + "user": "USER", + "assistant": "ASSISTANT", + "tool": "TOOL", +} + + +# --------------------------------------------------------------------------- +# Message building +# --------------------------------------------------------------------------- + + +def adapt_messages_to_generic_oci_standard_content_message( + role: str, content: Union[str, list] +) -> OCIMessage: + """Convert a plain-text or multipart content message to OCI format.""" + new_content: List[OCIContentPartUnion] = [] + if isinstance(content, str): + return OCIMessage( + role=open_ai_to_generic_oci_role_map[role], + content=[OCITextContentPart(text=content)], + toolCalls=None, + toolCallId=None, + ) + + for content_item in content: + if not isinstance(content_item, dict): + raise OCIError( + status_code=400, message="Each content item must be a dictionary" + ) + + item_type = content_item.get("type") + if not isinstance(item_type, str): + raise OCIError( + status_code=400, + message="Each content item must have a string `type` field", + ) + if item_type not in ["text", "image_url"]: + raise OCIError( + status_code=400, + message=f"Content type `{item_type}` is not supported by OCI", + ) + + if item_type == "text": + text = content_item.get("text") + if not isinstance(text, str): + raise OCIError( + status_code=400, + message="Content item of type `text` must have a string `text` field", + ) + new_content.append(OCITextContentPart(text=text)) + + elif item_type == "image_url": + image_url = content_item.get("image_url") + if isinstance(image_url, dict): + image_url = image_url.get("url") + if not isinstance(image_url, str): + raise OCIError( + status_code=400, + message="Prop `image_url` must be a string or an object with a `url` property", + ) + new_content.append(OCIImageContentPart(imageUrl=OCIImageUrl(url=image_url))) + + return OCIMessage( + role=open_ai_to_generic_oci_role_map[role], + content=new_content, + toolCalls=None, + toolCallId=None, + ) + + +def adapt_messages_to_generic_oci_standard_tool_call( + role: str, tool_calls: list +) -> OCIMessage: + """Convert an assistant tool-call message to OCI format.""" + tool_calls_formatted = [] + for tool_call in tool_calls: + if not isinstance(tool_call, dict): + raise OCIError( + status_code=400, message="Each tool call must be a dictionary" + ) + if tool_call.get("type") != "function": + raise OCIError( + status_code=400, message="OCI only supports function tool calls" + ) + + tool_call_id = tool_call.get("id") + if not isinstance(tool_call_id, str): + raise OCIError(status_code=400, message="Tool call `id` must be a string") + + tool_function = tool_call.get("function") + if not isinstance(tool_function, dict): + raise OCIError( + status_code=400, message="Tool call `function` must be a dictionary" + ) + + function_name = tool_function.get("name") + if not isinstance(function_name, str): + raise OCIError( + status_code=400, message="Tool call `function.name` must be a string" + ) + + arguments = tool_call["function"].get("arguments", "{}") + if not isinstance(arguments, str): + raise OCIError( + status_code=400, + message="Tool call `function.arguments` must be a JSON string", + ) + + tool_calls_formatted.append( + OCIToolCall( + id=tool_call_id, + type="FUNCTION", + name=function_name, + arguments=arguments, + ) + ) + + return OCIMessage( + role=open_ai_to_generic_oci_role_map[role], + content=None, + toolCalls=tool_calls_formatted, + toolCallId=None, + ) + + +def adapt_messages_to_generic_oci_standard_tool_response( + role: str, tool_call_id: str, content: str +) -> OCIMessage: + """Convert a tool-result message to OCI format.""" + return OCIMessage( + role=open_ai_to_generic_oci_role_map[role], + content=[OCITextContentPart(text=content)], + toolCalls=None, + toolCallId=tool_call_id, + ) + + +def adapt_messages_to_generic_oci_standard( + messages: List[AllMessageValues], +) -> List[OCIMessage]: + """Convert an OpenAI-format message array to OCI GENERIC format.""" + new_messages = [] + for message in messages: + role = message["role"] + content = message.get("content") + tool_calls = message.get("tool_calls") + tool_call_id = message.get("tool_call_id") + + if role == "assistant" and tool_calls is not None: + if not isinstance(tool_calls, list): + raise OCIError( + status_code=400, message="Message `tool_calls` must be a list" + ) + new_messages.append( + adapt_messages_to_generic_oci_standard_tool_call(role, tool_calls) + ) + + elif role in ["system", "user", "assistant"] and content is not None: + if not isinstance(content, (str, list)): + raise OCIError( + status_code=400, + message="Message `content` must be a string or list of content parts", + ) + new_messages.append( + adapt_messages_to_generic_oci_standard_content_message(role, content) + ) + + elif role == "tool": + if not isinstance(tool_call_id, str): + raise OCIError( + status_code=400, + message="Tool result message must have a string `tool_call_id`", + ) + if not isinstance(content, str): + raise OCIError( + status_code=400, + message="Tool result message `content` must be a string", + ) + new_messages.append( + adapt_messages_to_generic_oci_standard_tool_response( + role, tool_call_id, content + ) + ) + + return new_messages + + +# --------------------------------------------------------------------------- +# Tool definition adaptation +# --------------------------------------------------------------------------- + + +def adapt_tool_definition_to_oci_standard( + tools: List[Dict], vendor: OCIVendors +) -> List[OCIToolDefinition]: + """Convert OpenAI-format tool definitions to OCI GENERIC format. + + Resolves ``$ref``/``$defs`` and ``anyOf`` that the OCI endpoint rejects. + """ + new_tools = [] + for tool in tools: + if tool["type"] != "function": + raise OCIError(status_code=400, message="OCI only supports function tools") + + tool_function = tool.get("function") + if not isinstance(tool_function, dict): + raise OCIError( + status_code=400, message="Tool `function` must be a dictionary" + ) + + raw_params = tool_function.get("parameters", {}) + resolved_params = sanitize_oci_schema( + resolve_oci_schema_anyof(resolve_oci_schema_refs(raw_params)) + ) + + new_tools.append( + OCIToolDefinition( + type="FUNCTION", + name=tool_function.get("name"), + description=tool_function.get("description", ""), + parameters=resolved_params, + ) + ) + + return new_tools + + +def _normalize_oci_finish_reason(raw: Optional[str]) -> Optional[str]: + """Map an OCI-specific finish reason to its OpenAI-standard equivalent. + + OCI emits ``COMPLETE`` / ``MAX_TOKENS`` / ``TOOL_CALL(S)`` plus a long tail + of error/cancel reasons (``ERROR``, ``ERROR_TOXIC``, ``ERROR_LIMIT``, + ``USER_CANCEL``, ``CONTENT_FILTERED``, ``CANCELLED``, ...). The OpenAI + spec only defines ``stop`` / ``length`` / ``tool_calls`` / ... — anything + else is collapsed to ``"stop"`` so downstream consumers switching on + ``finish_reason`` keep working. A ``None`` input passes through unchanged. + """ + if raw is None: + return None + if raw == "COMPLETE": + return "stop" + if raw == "MAX_TOKENS": + return "length" + if raw in ("TOOL_CALL", "TOOL_CALLS"): + return "tool_calls" + return "stop" + + +def _synthesize_oci_tool_call_id(position: int, name: str, arguments: str) -> str: + """Deterministic synthetic tool-call id derived from chunk content. + + Used as a fallback when OCI omits ``id`` (always the case for the OCI + Cohere protocol, occasionally the case for OCI GENERIC streaming chunks). + A random ``uuid4`` per chunk would cause downstream stream-merging + consumers — which key off the tool-call ``id`` — to treat re-emissions of + the same logical call (e.g. terminal consolidation chunks, retries) as + distinct calls. A content-derived digest stays stable across identical + re-emissions while differing across truly distinct calls. + """ + digest = hashlib.sha256( + f"{position}|{name}|{arguments}".encode("utf-8"), + usedforsecurity=False, + ).hexdigest()[:24] + return f"call_{digest}" + + +def adapt_tools_to_openai_standard( + tools: List[OCIToolCall], +) -> List[ChatCompletionMessageToolCall]: + """Convert OCI tool-call objects in a response to the OpenAI format.""" + return [ + ChatCompletionMessageToolCall( + id=tool.id or _synthesize_oci_tool_call_id(i, tool.name, tool.arguments), + type="function", + function={"name": tool.name, "arguments": tool.arguments}, + ) + for i, tool in enumerate(tools) + ] + + +# --------------------------------------------------------------------------- +# Response parsing +# --------------------------------------------------------------------------- + + +def handle_generic_response( + json_data: dict, + model: str, + model_response: ModelResponse, + raw_response: httpx.Response, +) -> ModelResponse: + """Parse a non-streaming GENERIC OCI response into a LiteLLM ModelResponse.""" + try: + completion_response = OCICompletionResponse(**json_data) + except (TypeError, ValidationError) as e: + raise OCIError( + message=f"Response cannot be casted to OCICompletionResponse: {str(e)}", + status_code=raw_response.status_code, + ) + + iso_str = completion_response.chatResponse.timeCreated + dt = datetime.datetime.fromisoformat(iso_str.replace("Z", "+00:00")) + model_response.created = int(dt.timestamp()) + model_response.model = completion_response.modelId + + if not completion_response.chatResponse.choices: + raise OCIError( + message="OCI response contained no choices", + status_code=raw_response.status_code, + ) + + response_choice = completion_response.chatResponse.choices[0] + message = model_response.choices[0].message # type: ignore + response_message = response_choice.message + if response_message is not None: + if response_message.content: + # Concatenate all text parts — matches the streaming handler, which + # iterates the full content array. Skips non-text parts (e.g. image + # parts) so a leading non-text part doesn't suppress trailing text. + text: Optional[str] = None + for item in response_message.content: + if isinstance(item, OCITextContentPart): + text = (text or "") + item.text + if text is not None: + message.content = text + if response_message.toolCalls: + message.tool_calls = adapt_tools_to_openai_standard( + response_message.toolCalls + ) + + model_response.choices[0].finish_reason = _normalize_oci_finish_reason( # type: ignore[union-attr,assignment] + response_choice.finishReason + ) + + oci_usage = completion_response.chatResponse.usage + reasoning_tokens: Optional[int] = None + if ( + oci_usage.completionTokensDetails + and oci_usage.completionTokensDetails.reasoningTokens is not None + ): + reasoning_tokens = oci_usage.completionTokensDetails.reasoningTokens + model_response.usage = Usage( # type: ignore[attr-defined] + prompt_tokens=oci_usage.promptTokens, + completion_tokens=oci_usage.completionTokens or 0, + total_tokens=oci_usage.totalTokens, + reasoning_tokens=reasoning_tokens, + ) + + return model_response + + +def handle_generic_stream_chunk(dict_chunk: dict) -> ModelResponseStream: + """Parse a single GENERIC SSE chunk into a LiteLLM ModelResponseStream.""" + # OCI streams tool calls progressively — early chunks may omit required fields. + if dict_chunk.get("message") and dict_chunk["message"].get("toolCalls"): + for tool_call in dict_chunk["message"]["toolCalls"]: + tool_call.setdefault("arguments", "") + tool_call.setdefault("id", "") + tool_call.setdefault("name", "") + + try: + typed_chunk = OCIStreamChunk(**dict_chunk) + except (TypeError, ValidationError) as e: + raise OCIError( + status_code=500, + message=f"Chunk cannot be parsed as OCIStreamChunk: {str(e)}", + ) + + if typed_chunk.index is None: + typed_chunk.index = 0 + + # Emit ``content=None`` rather than ``content=""`` on chunks with no text + # parts (e.g. tool-call-only or keep-alive chunks) so downstream + # stream-mergers that distinguish "no text in this delta" from "an + # explicitly empty text delta" behave correctly. + text: Optional[str] = None + if typed_chunk.message and typed_chunk.message.content: + for item in typed_chunk.message.content: + if isinstance(item, OCITextContentPart): + text = (text or "") + item.text + elif isinstance(item, OCIImageContentPart): + raise OCIError( + status_code=500, + message="OCI returned image content in a streaming response — not supported", + ) + else: + raise OCIError( + status_code=500, + message=f"Unsupported content type in OCI streaming response: {item.type}", + ) + + # Build plain tool-call dicts inline (matching the shape produced by + # ``handle_cohere_stream_chunk``) rather than calling + # ``adapt_tools_to_openai_standard`` and ``model_dump``-ing the typed + # objects. Both code paths feed ``Delta.tool_calls``, so emitting the + # same minimal ``{"id", "type", "function": {"name", "arguments"}}`` + # shape keeps downstream stream-mergers behaving identically across + # GENERIC and Cohere chunks. + tool_calls: Optional[List[Dict[str, Any]]] = None + if typed_chunk.message and typed_chunk.message.toolCalls: + tool_calls = [ + { + "id": tc.id or _synthesize_oci_tool_call_id(i, tc.name, tc.arguments), + "type": "function", + "function": { + "name": tc.name, + "arguments": tc.arguments, + }, + } + for i, tc in enumerate(typed_chunk.message.toolCalls) + ] + + finish_reason: Optional[str] = _normalize_oci_finish_reason( + typed_chunk.finishReason + ) + + return ModelResponseStream( + choices=[ + StreamingChoices( + index=typed_chunk.index, + delta=Delta( + content=text, + tool_calls=tool_calls, + provider_specific_fields=None, + thinking_blocks=None, + reasoning_content=None, + ), + finish_reason=finish_reason, + ) + ] + ) diff --git a/litellm/llms/oci/chat/transformation.py b/litellm/llms/oci/chat/transformation.py index 62104e921a4..f050f9eea36 100644 --- a/litellm/llms/oci/chat/transformation.py +++ b/litellm/llms/oci/chat/transformation.py @@ -1,20 +1,26 @@ -import base64 -import datetime -import hashlib +""" +OCI Generative AI — chat transformation orchestrator. + +This module wires together the Cohere-specific and Generic-model helpers to +implement the LiteLLM BaseConfig interface. Heavy-lifting lives in: + + - :mod:`litellm.llms.oci.chat.cohere` — Cohere message/tool/response logic + - :mod:`litellm.llms.oci.chat.generic` — Generic message/tool/response logic + - :mod:`litellm.llms.oci.common_utils` — auth, signing, schema utilities +""" + import json -from dataclasses import dataclass from typing import ( TYPE_CHECKING, Any, AsyncIterator, Dict, + Iterator, List, Optional, - Protocol, Tuple, Union, ) -from urllib.parse import urlparse import httpx @@ -28,43 +34,43 @@ from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, version, ) -from litellm.llms.oci.common_utils import OCIError +from litellm.llms.oci.chat.cohere import ( + _extract_text_content, + adapt_messages_to_cohere_standard, + adapt_tool_definitions_to_cohere_standard, + handle_cohere_response, + handle_cohere_stream_chunk, +) +from litellm.llms.oci.chat.generic import ( + adapt_messages_to_generic_oci_standard, + adapt_tool_definition_to_oci_standard, + handle_generic_response, + handle_generic_stream_chunk, +) +from litellm.llms.oci.common_utils import ( + OCI_API_VERSION, + OCIError, + OCIRequestWrapper, # re-exported for backwards compatibility + get_oci_base_url, + resolve_oci_credentials, + sign_oci_request, + validate_oci_environment, +) from litellm.types.llms.oci import ( CohereChatRequest, - CohereMessage, - CohereChatResult, - CohereParameterDefinition, - CohereStreamChunk, - CohereTool, - CohereToolCall, OCIChatRequestPayload, OCICompletionPayload, - OCICompletionResponse, - OCIContentPartUnion, - OCIImageContentPart, - OCIImageUrl, - OCIMessage, - OCIRoles, OCIServingMode, - OCIStreamChunk, - OCITextContentPart, - OCIToolCall, - OCIToolDefinition, OCIVendors, ) from litellm.types.llms.openai import AllMessageValues from litellm.types.utils import ( - Delta, LlmProviders, ModelResponse, ModelResponseStream, - StreamingChoices, -) -from litellm.utils import ( - ChatCompletionMessageToolCall, - CustomStreamWrapper, - Usage, ) +from litellm.utils import supports_reasoning +from litellm.litellm_core_utils.streaming_handler import CustomStreamWrapper if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj @@ -74,142 +80,157 @@ else: LiteLLMLoggingObj = Any -class OCISignerProtocol(Protocol): - """ - Protocol for OCI request signers (e.g., oci.signer.Signer). - - This protocol defines the interface expected for OCI SDK signer objects. - Compatible with the OCI Python SDK's Signer class. - - See: https://docs.oracle.com/en-us/iaas/tools/python/latest/api/signing.html - """ - - def do_request_sign( - self, request: Any, *, enforce_content_headers: bool = False - ) -> None: - """ - Sign an HTTP request by adding authentication headers. - - Args: - request: Request object with method, url, headers, body, and path_url attributes - enforce_content_headers: Whether to enforce content-type and content-length headers - """ - ... - - -@dataclass -class OCIRequestWrapper: - """ - Wrapper for HTTP requests compatible with OCI signer interface. - - This class wraps request data in a format compatible with OCI SDK signers, - which expect objects with method, url, headers, body, and path_url attributes. - """ - - method: str - url: str - headers: dict - body: bytes - - @property - def path_url(self) -> str: - """Returns the path + query string for OCI signing.""" - parsed_url = urlparse(self.url) - return parsed_url.path + ("?" + parsed_url.query if parsed_url.query else "") - - -def sha256_base64(data: bytes) -> str: - digest = hashlib.sha256(data).digest() - return base64.b64encode(digest).decode() - - -def build_signature_string(method, path, headers, signed_headers): - lines = [] - for header in signed_headers: - if header == "(request-target)": - value = f"{method.lower()} {path}" - else: - value = headers[header] - lines.append(f"{header}: {value}") - return "\n".join(lines) - - -def load_private_key_from_str(key_str: str): - try: - from cryptography.hazmat.primitives import serialization - from cryptography.hazmat.primitives.asymmetric import rsa - except ImportError as e: - raise ImportError( - "cryptography package is required for OCI authentication. " - "Please install it with: pip install cryptography" - ) from e - - key = serialization.load_pem_private_key( - key_str.encode("utf-8"), - password=None, - ) - if not isinstance(key, rsa.RSAPrivateKey): - raise TypeError( - "The provided private key is not an RSA key, which is required for OCI signing." - ) - return key - - -def load_private_key_from_file(file_path: str): - """Loads a private key from a file path""" - try: - with open(file_path, "r", encoding="utf-8") as f: - key_str = f.read().strip() - except FileNotFoundError: - raise FileNotFoundError(f"Private key file not found: {file_path}") - except OSError as e: - raise OSError(f"Failed to read private key file '{file_path}': {e}") from e - - if not key_str: - raise ValueError(f"Private key file is empty: {file_path}") - - return load_private_key_from_str(key_str) - - -def get_vendor_from_model(model: str) -> OCIVendors: - """ - Extracts the vendor from the model name. - - OCI GenAI API uses two apiFormat values: - - "COHERE" for Cohere models (command-r, command-a, etc.) - - "GENERIC" for all other models (Meta Llama, xAI Grok, Google Gemini, etc.) - - Args: - model (str): The model name (e.g., "cohere.command-a-03-2025", "meta.llama-3.3-70b-instruct"). - Returns: - OCIVendors: The vendor enum value. - """ - vendor = model.split(".")[0].lower() - if vendor == "cohere": - return OCIVendors.COHERE - else: - return OCIVendors.GENERIC - - -# 5 minute timeout (models may need to load) +# Streaming timeout — generous because OCI models may need to warm up on first request STREAMING_TIMEOUT = 60 * 5 +def _model_uses_max_completion_tokens(model: str) -> bool: + """Return True for OCI-hosted models that require ``maxCompletionTokens``. + + Reasoning models on OCI (e.g. the OpenAI GPT-5 family) reject ``maxTokens`` + with HTTP 400 and require ``maxCompletionTokens`` per OpenAI's reasoning-API + convention. Driven by ``supports_reasoning`` in + ``model_prices_and_context_window.json`` so new model families are picked + up via a catalog update rather than a code change. + """ + if not model: + return False + name = model[4:] if model.lower().startswith("oci/") else model + return supports_reasoning(model=name, custom_llm_provider="oci") + + +def _iter_sse_events(stream: Iterator[str]) -> Iterator[str]: + """Yield one ``data:`` SSE line at a time from a sync text stream. + + The OCI streaming endpoint does not align SSE event boundaries with HTTP + read boundaries. A single read may carry multiple events, a single event + may straddle two reads, and some events arrive separated by only ``\\n`` + instead of ``\\n\\n``. This helper buffers across reads and yields each + complete ``data:`` line so JSON parsing downstream never sees a partial + payload. + """ + buffer = "" + for item in stream: + buffer += item + while "\n" in buffer: + line, buffer = buffer.split("\n", 1) + stripped = line.strip() + if stripped.startswith("data:"): + yield stripped + stripped = buffer.strip() + if stripped.startswith("data:"): + yield stripped + + +async def _aiter_sse_events(stream: AsyncIterator[str]) -> AsyncIterator[str]: + """Async twin of :func:`_iter_sse_events`.""" + buffer = "" + async for item in stream: + buffer += item + while "\n" in buffer: + line, buffer = buffer.split("\n", 1) + stripped = line.strip() + if stripped.startswith("data:"): + yield stripped + stripped = buffer.strip() + if stripped.startswith("data:"): + yield stripped + + +def _normalize_tool_choice(selected_params: Dict) -> None: + tc = selected_params.get("toolChoice") + if tc is None: + return + if isinstance(tc, str): + tc_map = { + "auto": {"type": "AUTO"}, + "none": {"type": "NONE"}, + "required": {"type": "REQUIRED"}, + "any": {"type": "REQUIRED"}, + } + selected_params["toolChoice"] = tc_map.get( + tc.lower(), {"type": "FUNCTION", "name": tc} + ) + return + if isinstance(tc, dict): + raw_type = tc.get("type") + if not isinstance(raw_type, str): + raise OCIError( + status_code=400, + message=f"Invalid tool_choice for OCI: missing or non-string 'type' in {tc!r}", + ) + upper = raw_type.upper() + if upper == "FUNCTION": + fn = tc.get("function") + name = fn.get("name") if isinstance(fn, dict) else tc.get("name") + if not (isinstance(name, str) and name): + raise OCIError( + status_code=400, + message="Invalid tool_choice for OCI: 'FUNCTION' type requires a non-empty function name", + ) + selected_params["toolChoice"] = {"type": "FUNCTION", "name": name} + elif upper in {"AUTO", "NONE", "REQUIRED"}: + selected_params["toolChoice"] = {"type": upper} + else: + raise OCIError( + status_code=400, + message=( + f"Invalid tool_choice for OCI: unsupported type {raw_type!r}; " + "expected one of 'FUNCTION', 'AUTO', 'NONE', 'REQUIRED'" + ), + ) + return + raise OCIError( + status_code=400, + message=( + f"Invalid tool_choice for OCI: expected str or dict, got " + f"{type(tc).__name__}" + ), + ) + + +def _normalize_response_format(selected_params: Dict, vendor: OCIVendors) -> None: + rf = selected_params.get("responseFormat") + if not isinstance(rf, dict) or "type" not in rf: + return + rf_payload = dict(rf) + selected_params["responseFormat"] = rf_payload + response_type = rf_payload["type"] + if "json_schema" in rf_payload: + raw_schema = rf_payload.pop("json_schema") + rf_payload["jsonSchema"] = ( + dict(raw_schema) if isinstance(raw_schema, dict) else raw_schema + ) + if vendor == OCIVendors.COHERE: + rf_payload["type"] = response_type + else: + fmt = response_type.upper() + rf_payload["type"] = "JSON_OBJECT" if fmt == "JSON" else fmt + + +def get_vendor_from_model(model: str) -> OCIVendors: + """Return the OCI vendor enum for a model name. + + OCI GenAI uses two ``apiFormat`` values: + + - ``"COHERE"`` for Cohere models (``cohere.*``) + - ``"GENERIC"`` for all others (Meta Llama, xAI Grok, Google Gemini, …) + """ + name = model[4:] if model.lower().startswith("oci/") else model + vendor = name.split(".")[0].lower() + if vendor == "cohere": + return OCIVendors.COHERE + return OCIVendors.GENERIC + + class OCIChatConfig(BaseConfig): - """ - Configuration class for OCI's API interface. - """ + """LiteLLM BaseConfig implementation for OCI Generative AI chat.""" - def __init__( - self, - ) -> None: - locals_ = locals().copy() - for key, value in locals_.items(): - if key != "self" and value is not None: - setattr(self.__class__, key, value) - # mark the class as using a custom stream wrapper because the default only iterates on lines - setattr(self.__class__, "has_custom_stream_wrapper", True) + @property + def has_custom_stream_wrapper(self) -> bool: + return True + def __init__(self) -> None: self.openai_to_oci_generic_param_map = { "stream": "isStream", "max_tokens": "maxTokens", @@ -221,6 +242,7 @@ class OCIChatConfig(BaseConfig): "logit_bias": "logitBias", "n": "numGenerations", "presence_penalty": "presencePenalty", + "reasoning_effort": "reasoningEffort", "seed": "seed", "stop": "stop", "tool_choice": "toolChoice", @@ -239,25 +261,43 @@ class OCIChatConfig(BaseConfig): "response_format": "responseFormat", } - # Cohere and Gemini use the same parameter mapping as GENERIC - self.openai_to_oci_cohere_param_map = ( - self.openai_to_oci_generic_param_map.copy() - ) + # Cohere param map differs from GENERIC in three ways: + # - tool_choice is unsupported + # - stop sequences key is "stopSequences" not "stop" + # - n (numGenerations) is GENERIC-only + # The unsupported keys are kept in the map with value ``False`` so + # ``map_openai_params`` either drops them (under drop_params) or raises + # a clear error, rather than silently passing them through. + self.openai_to_oci_cohere_param_map = { + k: ("stopSequences" if k == "stop" else v) + for k, v in self.openai_to_oci_generic_param_map.items() + } + self.openai_to_oci_cohere_param_map["tool_choice"] = False + self.openai_to_oci_cohere_param_map["n"] = False + # ``top_k`` is not a standard OpenAI param, but Cohere's chat request + # accepts ``topK`` and LiteLLM commonly forwards ``top_k`` as a + # passthrough param. Cohere-only — ``OCIChatRequestPayload`` (GENERIC) + # has no ``topK`` field. + self.openai_to_oci_cohere_param_map["top_k"] = "topK" + # OCI Cohere models are not reasoning models; mark reasoning_effort + # explicitly unsupported so callers either get a clear error or have + # the param dropped under drop_params, rather than silently passing + # through and tripping Pydantic validation on CohereChatRequest. + self.openai_to_oci_cohere_param_map["reasoning_effort"] = False + # CohereChatRequest has no logProbs/logitBias fields, so passing these + # through would be silently dropped by Pydantic. Mark them unsupported + # so get_supported_openai_params doesn't advertise them and callers + # get a clear error (or drop_params behaviour) instead. + self.openai_to_oci_cohere_param_map["logprobs"] = False + self.openai_to_oci_cohere_param_map["logit_bias"] = False def get_supported_openai_params(self, model: str) -> List[str]: - supported_params = [] - vendor = get_vendor_from_model(model) - if vendor == OCIVendors.COHERE: - open_ai_to_oci_param_map = self.openai_to_oci_cohere_param_map - open_ai_to_oci_param_map.pop("tool_choice") - open_ai_to_oci_param_map.pop("max_retries") - else: - open_ai_to_oci_param_map = self.openai_to_oci_generic_param_map - for key, value in open_ai_to_oci_param_map.items(): - if value: - supported_params.append(key) - - return supported_params + param_map = ( + self.openai_to_oci_cohere_param_map + if get_vendor_from_model(model) == OCIVendors.COHERE + else self.openai_to_oci_generic_param_map + ) + return [key for key, value in param_map.items() if value] def map_openai_params( self, @@ -268,238 +308,34 @@ class OCIChatConfig(BaseConfig): ) -> dict: adapted_params = {} vendor = get_vendor_from_model(model) - if vendor == OCIVendors.COHERE: - open_ai_to_oci_param_map = self.openai_to_oci_cohere_param_map - else: - open_ai_to_oci_param_map = self.openai_to_oci_generic_param_map - - all_params = {**non_default_params, **optional_params} - - for key, value in all_params.items(): - alias = open_ai_to_oci_param_map.get(key) + param_map = ( + self.openai_to_oci_cohere_param_map + if vendor == OCIVendors.COHERE + else self.openai_to_oci_generic_param_map + ) + for key, value in {**non_default_params, **optional_params}.items(): + alias = param_map.get(key) if alias is False: - # Workaround for mypy issue if drop_params or litellm.drop_params: continue - raise Exception(f"param `{key}` is not supported on OCI") - + raise OCIError( + status_code=400, + message=f"param `{key}` is not supported on OCI", + ) if alias is None: adapted_params[key] = value continue - adapted_params[alias] = value - + # Preserve the original OpenAI ``response_format`` key alongside the + # OCI-mapped ``responseFormat`` so downstream litellm framework code + # (e.g. ``json_mode`` detection, logging) that inspects + # ``optional_params["response_format"]`` continues to work. if alias == "responseFormat": adapted_params["response_format"] = value return adapted_params - def _sign_with_oci_signer( - self, - headers: dict, - optional_params: dict, - request_data: dict, - api_base: str, - ) -> Tuple[dict, bytes]: - """ - Sign request using OCI SDK Signer object. - - Args: - headers: Request headers to be signed - optional_params: Optional parameters including oci_signer - request_data: The request body dict to be sent in HTTP request - api_base: The complete URL for the HTTP request - - Returns: - Tuple of (signed_headers, encoded_body) - - Raises: - OCIError: If signing fails - ValueError: If HTTP method is unsupported - """ - oci_signer = optional_params.get("oci_signer") - body = json.dumps(request_data).encode("utf-8") - method = str(optional_params.get("method", "POST")).upper() - - if method not in ["POST", "GET", "PUT", "DELETE", "PATCH"]: - raise ValueError(f"Unsupported HTTP method: {method}") - - prepared_headers = headers.copy() - prepared_headers.setdefault("content-type", "application/json") - prepared_headers.setdefault("content-length", str(len(body))) - - request_wrapper = OCIRequestWrapper( - method=method, url=api_base, headers=prepared_headers, body=body - ) - - if oci_signer is None: - raise ValueError( - "oci_signer cannot be None when calling _sign_with_oci_signer" - ) - - try: - oci_signer.do_request_sign(request_wrapper, enforce_content_headers=True) - except Exception as e: - raise OCIError( - status_code=500, - message=( - f"Failed to sign request with provided oci_signer: {str(e)}. " - "The signer must implement the OCI SDK Signer interface with a " - "do_request_sign(request, enforce_content_headers=True) method. " - "See: https://docs.oracle.com/en-us/iaas/tools/python/latest/api/signing.html" - ), - ) from e - - headers.update(request_wrapper.headers) - return headers, body - - def _sign_with_manual_credentials( - self, - headers: dict, - optional_params: dict, - request_data: dict, - api_base: str, - ) -> Tuple[dict, None]: - """ - Sign request using manual OCI credentials. - - Args: - headers: Request headers to be signed - optional_params: Optional parameters including OCI credentials - request_data: The request body dict to be sent in HTTP request - api_base: The complete URL for the HTTP request - - Returns: - Tuple of (signed_headers, None) - - Raises: - Exception: If required credentials are missing - ImportError: If cryptography package is not installed - """ - oci_region = optional_params.get("oci_region", "us-ashburn-1") - api_base = ( - api_base - or litellm.api_base - or f"https://inference.generativeai.{oci_region}.oci.oraclecloud.com" - ) - oci_user = optional_params.get("oci_user") - oci_fingerprint = optional_params.get("oci_fingerprint") - oci_tenancy = optional_params.get("oci_tenancy") - oci_key = optional_params.get("oci_key") - oci_key_file = optional_params.get("oci_key_file") - - if ( - not oci_user - or not oci_fingerprint - or not oci_tenancy - or not (oci_key or oci_key_file) - ): - raise Exception( - "Missing required parameters: oci_user, oci_fingerprint, oci_tenancy, " - "and at least one of oci_key or oci_key_file." - ) - - method = str(optional_params.get("method", "POST")).upper() - body = json.dumps(request_data).encode("utf-8") - parsed = urlparse(api_base) - path = parsed.path or "/" - host = parsed.netloc - - date = datetime.datetime.utcnow().strftime("%a, %d %b %Y %H:%M:%S GMT") - content_type = headers.get("content-type", "application/json") - content_length = str(len(body)) - x_content_sha256 = sha256_base64(body) - - headers_to_sign = { - "date": date, - "host": host, - "content-type": content_type, - "content-length": content_length, - "x-content-sha256": x_content_sha256, - } - - signed_headers = [ - "date", - "(request-target)", - "host", - "content-length", - "content-type", - "x-content-sha256", - ] - signing_string = build_signature_string( - method, path, headers_to_sign, signed_headers - ) - - try: - from cryptography.hazmat.primitives import hashes - from cryptography.hazmat.primitives.asymmetric import padding - except ImportError as e: - raise ImportError( - "cryptography package is required for OCI authentication. " - "Please install it with: pip install cryptography" - ) from e - - # Handle oci_key - it should be a string (PEM content) - oci_key_content = None - if oci_key: - if isinstance(oci_key, str): - oci_key_content = oci_key - # Fix common issues with PEM content - # Replace escaped newlines with actual newlines - oci_key_content = oci_key_content.replace("\\n", "\n") - # Ensure proper line endings - if "\r\n" in oci_key_content: - oci_key_content = oci_key_content.replace("\r\n", "\n") - else: - raise OCIError( - status_code=400, - message=f"oci_key must be a string containing the PEM private key content. " - f"Got type: {type(oci_key).__name__}", - ) - - private_key = ( - load_private_key_from_str(oci_key_content) - if oci_key_content - else load_private_key_from_file(oci_key_file) if oci_key_file else None - ) - - if private_key is None: - raise OCIError( - status_code=400, - message="Private key is required for OCI authentication. Please provide either oci_key or oci_key_file.", - ) - - signature = private_key.sign( - signing_string.encode("utf-8"), - padding.PKCS1v15(), - hashes.SHA256(), - ) - signature_b64 = base64.b64encode(signature).decode() - - key_id = f"{oci_tenancy}/{oci_user}/{oci_fingerprint}" - - authorization = ( - 'Signature version="1",' - f'keyId="{key_id}",' - 'algorithm="rsa-sha256",' - f'headers="{" ".join(signed_headers)}",' - f'signature="{signature_b64}"' - ) - - headers.update( - { - "authorization": authorization, - "date": date, - "host": host, - "content-type": content_type, - "content-length": content_length, - "x-content-sha256": x_content_sha256, - } - ) - - return headers, None - def sign_request( self, headers: dict, @@ -510,61 +346,16 @@ class OCIChatConfig(BaseConfig): model: Optional[str] = None, stream: Optional[bool] = None, fake_stream: Optional[bool] = None, - ) -> Tuple[dict, Optional[bytes]]: - """ - Sign the OCI request by adding authentication headers. - - Supports two signing modes: - 1. OCI SDK Signer: Use an oci_signer object to sign the request - 2. Manual Signing: Use OCI credentials to manually sign the request - - Args: - headers: Request headers to be signed - optional_params: Optional parameters including auth credentials or oci_signer - request_data: The request body dict to be sent in HTTP request - api_base: The complete URL for the HTTP request - api_key: Optional API key (not used for OCI) - model: Optional model name - stream: Optional streaming flag - fake_stream: Optional fake streaming flag - - Returns: - Tuple of (signed_headers, encoded_body): - - If oci_signer is provided: Returns (headers, body) where body is the encoded JSON - - If manual credentials are provided: Returns (headers, None) as body is not returned - for the manual signing path - - Raises: - OCIError: If signing fails with oci_signer - Exception: If required credentials are missing - ImportError: If cryptography package is not installed (manual signing only) - - Example: - >>> from oci.signer import Signer - >>> signer = Signer( - ... tenancy="ocid1.tenancy.oc1..", - ... user="ocid1.user.oc1..", - ... fingerprint="xx:xx:xx", - ... private_key_file_location="~/.oci/key.pem" - ... ) - >>> headers, body = config.sign_request( - ... headers={}, - ... optional_params={"oci_signer": signer}, - ... request_data={"message": "Hello"}, - ... api_base="https://inference.generativeai.us-ashburn-1.oci.oraclecloud.com/..." - ... ) - """ - oci_signer = optional_params.get("oci_signer") - - # If a signer is provided, use it for request signing - if oci_signer is not None: - return self._sign_with_oci_signer( - headers, optional_params, request_data, api_base - ) - - # Standard manual credential signing - return self._sign_with_manual_credentials( - headers, optional_params, request_data, api_base + ) -> Tuple[dict, bytes]: + return sign_oci_request( + headers=headers, + optional_params=optional_params, + request_data=request_data, + api_base=api_base, + api_key=api_key, + model=model, + stream=stream, + fake_stream=fake_stream, ) def validate_environment( @@ -577,80 +368,35 @@ class OCIChatConfig(BaseConfig): api_key: Optional[str] = None, api_base: Optional[str] = None, ) -> dict: - """ - Validate the OCI environment and credentials. - - Supports two authentication modes: - 1. OCI SDK Signer: Pass an oci_signer object (e.g., oci.signer.Signer) - 2. Manual Credentials: Pass oci_user, oci_fingerprint, oci_tenancy, and oci_key/oci_key_file - - Args: - headers: Request headers to populate - model: Model name - messages: List of chat messages - optional_params: Optional parameters including authentication credentials - litellm_params: LiteLLM parameters - api_key: Optional API key (not used for OCI) - api_base: Optional API base URL - - Returns: - Updated headers dict - - Raises: - Exception: If required parameters are missing or invalid - """ - oci_signer = optional_params.get("oci_signer") - oci_region = optional_params.get("oci_region", "us-ashburn-1") - - # Determine api_base - api_base = ( - api_base - or litellm.api_base - or f"https://inference.generativeai.{oci_region}.oci.oraclecloud.com" - ) - - if not api_base: - raise Exception( - "Either `api_base` must be provided or `litellm.api_base` must be set. " - "Alternatively, you can set the `oci_region` optional parameter to use the default OCI region." - ) - - # Validate credentials only if signer is not provided - if oci_signer is None: - oci_user = optional_params.get("oci_user") - oci_fingerprint = optional_params.get("oci_fingerprint") - oci_tenancy = optional_params.get("oci_tenancy") - oci_key = optional_params.get("oci_key") - oci_key_file = optional_params.get("oci_key_file") - oci_compartment_id = optional_params.get("oci_compartment_id") - - if ( - not oci_user - or not oci_fingerprint - or not oci_tenancy - or not (oci_key or oci_key_file) - or not oci_compartment_id - ): - raise Exception( - "Missing required parameters: oci_user, oci_fingerprint, oci_tenancy, oci_compartment_id " - "and at least one of oci_key or oci_key_file. " - "Alternatively, provide an oci_signer object from the OCI SDK." - ) - - # Common header setup - headers.update( - { - "content-type": "application/json", - "user-agent": f"litellm/{version}", - } - ) - if not messages: - raise Exception( - "kwarg `messages` must be an array of messages that follow the openai chat standard" + raise OCIError( + status_code=400, + message="kwarg `messages` must be an array of messages that follow the openai chat standard", ) - - return headers + if optional_params.get("oci_signer") is None: + creds = resolve_oci_credentials(optional_params) + missing = [ + k + for k in ( + "oci_user", + "oci_fingerprint", + "oci_tenancy", + "oci_compartment_id", + ) + if not creds.get(k) + ] + if missing or not (creds.get("oci_key") or creds.get("oci_key_file")): + raise OCIError( + status_code=401, + message=( + "Missing required parameters: oci_user, oci_fingerprint, oci_tenancy, " + "oci_compartment_id and at least one of oci_key or oci_key_file. " + "These can be supplied via optional_params or via OCI_USER, OCI_FINGERPRINT, " + "OCI_TENANCY, OCI_COMPARTMENT_ID, OCI_KEY_FILE environment variables. " + "Alternatively, provide an oci_signer object from the OCI SDK." + ), + ) + return validate_oci_environment(headers, optional_params, api_key) def get_complete_url( self, @@ -661,43 +407,63 @@ class OCIChatConfig(BaseConfig): litellm_params: dict, stream: Optional[bool] = None, ) -> str: - oci_region = optional_params.get("oci_region", "us-ashburn-1") - return f"https://inference.generativeai.{oci_region}.oci.oraclecloud.com/20231130/actions/chat" + base = get_oci_base_url(optional_params, api_base or litellm.api_base) + return f"{base}/{OCI_API_VERSION}/actions/chat" - def _get_optional_params(self, vendor: OCIVendors, optional_params: dict) -> Dict: - selected_params = {} - if vendor == OCIVendors.COHERE: - open_ai_to_oci_param_map = self.openai_to_oci_cohere_param_map - # remove tool_choice from the map - open_ai_to_oci_param_map.pop("tool_choice") - # Add default values for Cohere API - selected_params = { - "maxTokens": 600, - "temperature": 1, - "topK": 0, - "topP": 0.75, - "frequencyPenalty": 0, - } - else: - open_ai_to_oci_param_map = self.openai_to_oci_generic_param_map + def _get_optional_params( + self, vendor: OCIVendors, optional_params: dict, model: str = "" + ) -> Dict: + param_map = ( + self.openai_to_oci_cohere_param_map + if vendor == OCIVendors.COHERE + else self.openai_to_oci_generic_param_map + ) + selected_params: Dict = {} - # Map OpenAI params to OCI params - for openai_key, oci_key in open_ai_to_oci_param_map.items(): - if oci_key and openai_key in optional_params: - selected_params[oci_key] = optional_params[openai_key] # type: ignore[index] + # OpenAI reasoning models on OCI (e.g. GPT-5 family) reject "maxTokens" + # and require "maxCompletionTokens" per OCI's /20231130/Chat schema. + # Driven by the supports_reasoning flag in the model catalog. Cohere's + # endpoint uses "maxTokens" regardless, so the override is GENERIC-only. + max_tokens_key = ( + "maxCompletionTokens" + if vendor != OCIVendors.COHERE + and model + and _model_uses_max_completion_tokens(model) + else "maxTokens" + ) - # Also check for already-mapped OCI params (for backward compatibility) - for oci_value in open_ai_to_oci_param_map.values(): - if ( - oci_value - and oci_value in optional_params - and oci_value not in selected_params - ): - selected_params[oci_value] = optional_params[oci_value] # type: ignore[index] + # ``map_openai_params`` runs before ``transform_request`` (and thus + # before this helper), so by the time we see ``optional_params`` the + # OpenAI keys have already been translated to their OCI aliases. + # We still accept the original OpenAI key as a fallback for callers + # that build ``optional_params`` directly, with OpenAI keys winning + # over OCI aliases when both happen to be present. The first OpenAI + # key reaching a given OCI target wins, so ``max_tokens`` / + # ``max_completion_tokens`` (both → ``maxTokens``) don't double-write. + for openai_key, oci_alias in param_map.items(): + if not oci_alias: + continue + target = max_tokens_key if oci_alias == "maxTokens" else oci_alias + if target in selected_params: + continue + if openai_key in optional_params: + selected_params[target] = optional_params[openai_key] # type: ignore[index] + elif oci_alias in optional_params: + selected_params[target] = optional_params[oci_alias] # type: ignore[index] + + # OCI expects uppercase reasoning levels (LOW/MEDIUM/HIGH/NONE); OpenAI + # clients send lowercase. OpenAI's "disable" maps to OCI's "NONE". + if "reasoningEffort" in selected_params: + effort = selected_params["reasoningEffort"] + if isinstance(effort, str): + normalized = effort.upper() + if normalized == "DISABLE": + normalized = "NONE" + selected_params["reasoningEffort"] = normalized if "tools" in selected_params: if vendor == OCIVendors.COHERE: - selected_params["tools"] = self.adapt_tool_definitions_to_cohere_standard( # type: ignore[assignment] + selected_params["tools"] = adapt_tool_definitions_to_cohere_standard( # type: ignore[assignment] selected_params["tools"] # type: ignore[arg-type] ) else: @@ -705,146 +471,15 @@ class OCIChatConfig(BaseConfig): selected_params["tools"], vendor # type: ignore[arg-type] ) - # Transform response_format type to OCI uppercase format - if "responseFormat" in selected_params: - rf = selected_params["responseFormat"] - if isinstance(rf, dict) and "type" in rf: - rf_payload = dict(rf) - selected_params["responseFormat"] = rf_payload + # Normalise tool_choice to OCI's flat uppercase dict form + # ({"type": "AUTO"|"NONE"|"REQUIRED"} or {"type": "FUNCTION", "name": ""}). + # OCI rejects both the OpenAI string and the nested OpenAI dict shape. + _normalize_tool_choice(selected_params) - response_type = rf_payload["type"] - schema_payload: Optional[Any] = None - - if "json_schema" in rf_payload: - raw_schema_payload = rf_payload.pop("json_schema") - if isinstance(raw_schema_payload, dict): - schema_payload = dict(raw_schema_payload) - else: - schema_payload = raw_schema_payload - - if schema_payload is not None: - rf_payload["jsonSchema"] = schema_payload - - if vendor == OCIVendors.COHERE: - # Cohere expects lower-case type values - rf_payload["type"] = response_type - else: - format_type = response_type.upper() - if format_type == "JSON": - format_type = "JSON_OBJECT" - rf_payload["type"] = format_type + _normalize_response_format(selected_params, vendor) return selected_params - def adapt_messages_to_cohere_standard( - self, messages: List[AllMessageValues] - ) -> List[CohereMessage]: - """Build chat history for Cohere models.""" - chat_history = [] - for msg in messages[:-1]: # All messages except the last one - role = msg.get("role") - content = msg.get("content") - - if isinstance(content, list): - # Extract text from content array - text_content = "" - for content_item in content: - if ( - isinstance(content_item, dict) - and content_item.get("type") == "text" - ): - text_content += content_item.get("text", "") - content = text_content - - # Ensure content is a string - if not isinstance(content, str): - content = str(content) if content is not None else "" - - # Handle tool calls - tool_calls: Optional[List[CohereToolCall]] = None - if role == "assistant" and "tool_calls" in msg and msg.get("tool_calls"): # type: ignore[union-attr,typeddict-item] - tool_calls = [] - for tool_call in msg["tool_calls"]: # type: ignore[union-attr,typeddict-item] - # Parse arguments if they're a JSON string - raw_arguments: Any = tool_call.get("function", {}).get( - "arguments", {} - ) - if isinstance(raw_arguments, str): - try: - arguments: Dict[str, Any] = json.loads(raw_arguments) - except json.JSONDecodeError: - arguments = {} - else: - arguments = raw_arguments - - tool_calls.append( - CohereToolCall( - name=str(tool_call.get("function", {}).get("name", "")), - parameters=arguments, - ) - ) - - if role == "user": - chat_history.append(CohereMessage(role="USER", message=content)) - elif role == "assistant": - chat_history.append( - CohereMessage(role="CHATBOT", message=content, toolCalls=tool_calls) - ) - elif role == "tool": - # Tool messages need special handling - chat_history.append( - CohereMessage( - role="TOOL", - message=content, - toolCalls=None, # Tool messages don't have tool calls - ) - ) - - return chat_history - - def adapt_tool_definitions_to_cohere_standard( - self, tools: List[Dict[str, Any]] - ) -> List[CohereTool]: - """Adapt tool definitions to Cohere format.""" - cohere_tools = [] - for tool in tools: - function_def = tool.get("function", {}) - parameters = function_def.get("parameters", {}).get("properties", {}) - required = function_def.get("parameters", {}).get("required", []) - - parameter_definitions = {} - for param_name, param_schema in parameters.items(): - parameter_definitions[param_name] = CohereParameterDefinition( - description=param_schema.get("description", ""), - type=param_schema.get("type", "string"), - isRequired=param_name in required, - ) - - cohere_tools.append( - CohereTool( - name=function_def.get("name", ""), - description=function_def.get("description", ""), - parameterDefinitions=parameter_definitions, - ) - ) - - return cohere_tools - - def _extract_text_content(self, content: Any) -> str: - """Extract text content from message content.""" - if isinstance(content, str): - return content - elif isinstance(content, list): - text_content = "" - for content_item in content: - if ( - isinstance(content_item, dict) - and content_item.get("type") == "text" - ): - text_content += content_item.get("text", "") - return text_content - return str(content) - def transform_request( self, model: str, @@ -853,186 +488,78 @@ class OCIChatConfig(BaseConfig): litellm_params: dict, headers: dict, ) -> dict: - oci_compartment_id = optional_params.get("oci_compartment_id", None) + creds = resolve_oci_credentials(optional_params) + oci_compartment_id = creds["oci_compartment_id"] if not oci_compartment_id: - raise Exception("kwarg `oci_compartment_id` is required for OCI requests") + raise OCIError( + status_code=400, + message=( + "oci_compartment_id is required for OCI chat requests. " + "Pass it as optional_params or set the OCI_COMPARTMENT_ID env var." + ), + ) vendor = get_vendor_from_model(model) oci_serving_mode = optional_params.get("oci_serving_mode", "ON_DEMAND") if oci_serving_mode not in ["ON_DEMAND", "DEDICATED"]: - raise Exception( - "kwarg `oci_serving_mode` must be either 'ON_DEMAND' or 'DEDICATED'" + raise OCIError( + status_code=400, + message="kwarg `oci_serving_mode` must be either 'ON_DEMAND' or 'DEDICATED'", ) if oci_serving_mode == "DEDICATED": - oci_endpoint_id = optional_params.get("oci_endpoint_id", model) - servingMode = OCIServingMode( + serving_mode = OCIServingMode( servingType="DEDICATED", - endpointId=oci_endpoint_id, + endpointId=optional_params.get("oci_endpoint_id", model), ) else: - servingMode = OCIServingMode( - servingType="ON_DEMAND", - modelId=model, - ) + serving_mode = OCIServingMode(servingType="ON_DEMAND", modelId=model) - # Build request based on vendor type if vendor == OCIVendors.COHERE: - # For Cohere, we need to use the specific Cohere format - # Extract the last user message as the main message - user_messages = [msg for msg in messages if msg.get("role") == "user"] + user_messages = [m for m in messages if m.get("role") == "user"] if not user_messages: - raise Exception("No user message found for Cohere model") + raise OCIError( + status_code=400, + message="No user message found — Cohere models require at least one user message", + ) - # Extract system messages into preambleOverride - system_messages = [msg for msg in messages if msg.get("role") == "system"] + system_messages = [m for m in messages if m.get("role") == "system"] preamble_override = None if system_messages: preamble = "\n".join( - self._extract_text_content(msg["content"]) - for msg in system_messages + _extract_text_content(m["content"]) for m in system_messages ) if preamble: preamble_override = preamble - # Create Cohere-specific chat request - optional_cohere_params = self._get_optional_params( - OCIVendors.COHERE, optional_params - ) chat_request = CohereChatRequest( apiFormat="COHERE", - message=self._extract_text_content(user_messages[-1]["content"]), - chatHistory=self.adapt_messages_to_cohere_standard(messages), + message=_extract_text_content(user_messages[-1]["content"]), + chatHistory=adapt_messages_to_cohere_standard( + [m for m in messages if m.get("role") != "system"] + ), preambleOverride=preamble_override, - **optional_cohere_params, + **self._get_optional_params(OCIVendors.COHERE, optional_params, model), ) - data = OCICompletionPayload( compartmentId=oci_compartment_id, - servingMode=servingMode, + servingMode=serving_mode, chatRequest=chat_request, ) else: - # Use generic format for other vendors data = OCICompletionPayload( compartmentId=oci_compartment_id, - servingMode=servingMode, + servingMode=serving_mode, chatRequest=OCIChatRequestPayload( apiFormat=vendor.value, messages=adapt_messages_to_generic_oci_standard(messages), - **self._get_optional_params(vendor, optional_params), + **self._get_optional_params(vendor, optional_params, model), ), ) return data.model_dump(exclude_none=True) - def _handle_cohere_response( - self, json_response: dict, model: str, model_response: ModelResponse - ) -> ModelResponse: - """Handle Cohere-specific response format.""" - cohere_response = CohereChatResult(**json_response) - # Cohere response format (uses camelCase) - model_id = model - - # Set basic response info - model_response.model = model_id - model_response.created = int(datetime.datetime.now().timestamp()) - - # Extract the response text - response_text = cohere_response.chatResponse.text - oci_finish_reason = cohere_response.chatResponse.finishReason - - # Map finish reason - if oci_finish_reason == "COMPLETE": - finish_reason = "stop" - elif oci_finish_reason == "MAX_TOKENS": - finish_reason = "length" - else: - finish_reason = "stop" - - # Handle tool calls - tool_calls: Optional[List[Dict[str, Any]]] = None - if cohere_response.chatResponse.toolCalls: - tool_calls = [] - for tool_call in cohere_response.chatResponse.toolCalls: - tool_calls.append( - { - "id": f"call_{len(tool_calls)}", # Generate a simple ID - "type": "function", - "function": { - "name": tool_call.name, - "arguments": json.dumps(tool_call.parameters), - }, - } - ) - - # Create choice - from litellm.types.utils import Choices - - choice = Choices( - index=0, - message={ - "role": "assistant", - "content": response_text, - "tool_calls": tool_calls, - }, - finish_reason=finish_reason, - ) - model_response.choices = [choice] - - # Extract usage info - usage_info = cohere_response.chatResponse.usage - from litellm.types.utils import Usage - - model_response.usage = Usage( # type: ignore[attr-defined] - prompt_tokens=usage_info.promptTokens, # type: ignore[union-attr] - completion_tokens=usage_info.completionTokens, # type: ignore[union-attr] - total_tokens=usage_info.totalTokens, # type: ignore[union-attr] - ) - - return model_response - - def _handle_generic_response( - self, - json: dict, - model: str, - model_response: ModelResponse, - raw_response: httpx.Response, - ) -> ModelResponse: - """Handle generic OCI response format.""" - try: - completion_response = OCICompletionResponse(**json) - except TypeError as e: - raise OCIError( - message=f"Response cannot be casted to OCICompletionResponse: {str(e)}", - status_code=raw_response.status_code, - ) - - iso_str = completion_response.chatResponse.timeCreated - dt = datetime.datetime.fromisoformat(iso_str.replace("Z", "+00:00")) - model_response.created = int(dt.timestamp()) - - model_response.model = completion_response.modelId - - message = model_response.choices[0].message # type: ignore - response_message = completion_response.chatResponse.choices[0].message - if response_message.content and response_message.content[0].type == "TEXT": - message.content = response_message.content[0].text - if response_message.toolCalls: - message.tool_calls = adapt_tools_to_openai_standard( - response_message.toolCalls - ) - - usage = Usage( - prompt_tokens=completion_response.chatResponse.usage.promptTokens, - completion_tokens=completion_response.chatResponse.usage.completionTokens, - total_tokens=completion_response.chatResponse.usage.totalTokens, - ) - model_response.usage = usage # type: ignore - - return model_response - def transform_response( self, model: str, @@ -1047,34 +574,31 @@ class OCIChatConfig(BaseConfig): api_key: Optional[str] = None, json_mode: Optional[bool] = None, ) -> ModelResponse: - json = raw_response.json() # noqa: F811 + response_json = raw_response.json() - error = json.get("error") - - if error is not None: - raise OCIError( - message=str(json["error"]), - status_code=raw_response.status_code, - ) - - if not isinstance(json, dict): + if not isinstance(response_json, dict): raise OCIError( message="Invalid response format from OCI", status_code=raw_response.status_code, ) - vendor = get_vendor_from_model(model) + if response_json.get("error") is not None: + raise OCIError( + message=str(response_json["error"]), + status_code=raw_response.status_code, + ) - # Handle response based on vendor type + vendor = get_vendor_from_model(model) if vendor == OCIVendors.COHERE: - model_response = self._handle_cohere_response(json, model, model_response) + model_response = handle_cohere_response( + response_json, model, model_response, raw_response + ) else: - model_response = self._handle_generic_response( - json, model, model_response, raw_response + model_response = handle_generic_response( + response_json, model, model_response, raw_response ) model_response._hidden_params["additional_headers"] = raw_response.headers - return model_response @track_llm_api_timing() @@ -1091,8 +615,6 @@ class OCIChatConfig(BaseConfig): json_mode: Optional[bool] = None, signed_json_body: Optional[bytes] = None, ) -> "OCIStreamWrapper": - if "stream" in data: - del data["stream"] if client is None or isinstance(client, AsyncHTTPHandler): client = _get_httpx_client(params={}) @@ -1100,7 +622,11 @@ class OCIChatConfig(BaseConfig): response = client.post( api_base, headers=headers, - data=json.dumps(data), + data=( + signed_json_body + if signed_json_body is not None + else json.dumps(data) + ), stream=True, logging_obj=logging_obj, timeout=STREAMING_TIMEOUT, @@ -1111,15 +637,12 @@ class OCIChatConfig(BaseConfig): if response.status_code != 200: raise OCIError(status_code=response.status_code, message=response.text) - completion_stream = response.iter_text() - - streaming_response = OCIStreamWrapper( - completion_stream=completion_stream, + return OCIStreamWrapper( + completion_stream=_iter_sse_events(response.iter_text()), model=model, custom_llm_provider=custom_llm_provider, logging_obj=logging_obj, ) - return streaming_response @track_llm_api_timing() async def get_async_custom_stream_wrapper( @@ -1135,17 +658,18 @@ class OCIChatConfig(BaseConfig): json_mode: Optional[bool] = None, signed_json_body: Optional[bytes] = None, ) -> "OCIStreamWrapper": - if "stream" in data: - del data["stream"] - if client is None or isinstance(client, HTTPHandler): - client = get_async_httpx_client(llm_provider=LlmProviders.BYTEZ, params={}) + client = get_async_httpx_client(llm_provider=LlmProviders.OCI, params={}) try: response = await client.post( api_base, headers=headers, - data=json.dumps(data), + data=( + signed_json_body + if signed_json_body is not None + else json.dumps(data) + ), stream=True, logging_obj=logging_obj, timeout=STREAMING_TIMEOUT, @@ -1156,22 +680,12 @@ class OCIChatConfig(BaseConfig): if response.status_code != 200: raise OCIError(status_code=response.status_code, message=response.text) - completion_stream = response.aiter_text() - - async def split_chunks(completion_stream: AsyncIterator[str]): - async for item in completion_stream: - for chunk in item.split("\n\n"): - if not chunk: - continue - yield chunk.strip() - - streaming_response = OCIStreamWrapper( - completion_stream=split_chunks(completion_stream), + return OCIStreamWrapper( + completion_stream=_aiter_sse_events(response.aiter_text()), model=model, custom_llm_provider=custom_llm_provider, logging_obj=logging_obj, ) - return streaming_response def get_error_class( self, error_message: str, status_code: int, headers: Union[dict, httpx.Headers] @@ -1179,332 +693,61 @@ class OCIChatConfig(BaseConfig): return OCIError(status_code=status_code, message=error_message) -open_ai_to_generic_oci_role_map: Dict[str, OCIRoles] = { - "system": "SYSTEM", - "user": "USER", - "assistant": "ASSISTANT", - "tool": "TOOL", -} - - -def adapt_messages_to_generic_oci_standard_content_message( - role: str, content: Union[str, list] -) -> OCIMessage: - new_content: List[OCIContentPartUnion] = [] - if isinstance(content, str): - return OCIMessage( - role=open_ai_to_generic_oci_role_map[role], - content=[OCITextContentPart(text=content)], - toolCalls=None, - toolCallId=None, - ) - - # content is a list of content items: - # [ - # {"type": "text", "text": "Hello"}, - # {"type": "image_url", "image_url": "https://example.com/image.png"} - # ] - for content_item in content: - if not isinstance(content_item, dict): - raise Exception("Each content item must be a dictionary") - - type = content_item.get("type") - if not isinstance(type, str): - raise Exception("Prop `type` is not a string") - - if type not in ["text", "image_url"]: - raise Exception(f"Prop `{type}` is not supported") - - if type == "text": - text = content_item.get("text") - if not isinstance(text, str): - raise Exception("Prop `text` is not a string") - new_content.append(OCITextContentPart(text=text)) - - elif type == "image_url": - image_url = content_item.get("image_url") - # Handle both OpenAI format (object with url) and string format - if isinstance(image_url, dict): - image_url = image_url.get("url") - if not isinstance(image_url, str): - raise Exception( - "Prop `image_url` must be a string or an object with a `url` property" - ) - new_content.append(OCIImageContentPart(imageUrl=OCIImageUrl(url=image_url))) - - return OCIMessage( - role=open_ai_to_generic_oci_role_map[role], - content=new_content, - toolCalls=None, - toolCallId=None, - ) - - -def adapt_messages_to_generic_oci_standard_tool_call( - role: str, tool_calls: list -) -> OCIMessage: - tool_calls_formated = [] - for tool_call in tool_calls: - if not isinstance(tool_call, dict): - raise Exception("Each tool call must be a dictionary") - - if tool_call.get("type") != "function": - raise Exception("OCI only supports function tools") - - tool_call_id = tool_call.get("id") - if not isinstance(tool_call_id, str): - raise Exception("Prop `id` is not a string") - - tool_function = tool_call.get("function") - if not isinstance(tool_function, dict): - raise Exception("Prop `function` is not a dictionary") - - function_name = tool_function.get("name") - if not isinstance(function_name, str): - raise Exception("Prop `name` is not a string") - - arguments = tool_call["function"].get("arguments", "{}") - if not isinstance(arguments, str): - raise Exception("Prop `arguments` is not a string") - - # tool_calls_formated.append(OCIToolCall( - # id=tool_call_id, - # type="FUNCTION", - # function=OCIFunction( - # name=function_name, - # arguments=arguments - # ) - # )) - - tool_calls_formated.append( - OCIToolCall( - id=tool_call_id, - type="FUNCTION", - name=function_name, - arguments=arguments, - ) - ) - - return OCIMessage( - role=open_ai_to_generic_oci_role_map[role], - content=None, - toolCalls=tool_calls_formated, - toolCallId=None, - ) - - -def adapt_messages_to_generic_oci_standard_tool_response( - role: str, tool_call_id: str, content: str -) -> OCIMessage: - return OCIMessage( - role=open_ai_to_generic_oci_role_map[role], - content=[OCITextContentPart(text=content)], - toolCalls=None, - toolCallId=tool_call_id, - ) - - -def adapt_messages_to_generic_oci_standard( - messages: List[AllMessageValues], -) -> List[OCIMessage]: - new_messages = [] - for message in messages: - role = message["role"] - content = message.get("content") - tool_calls = message.get("tool_calls") - tool_call_id = message.get("tool_call_id") - - if role == "assistant" and tool_calls is not None: - if not isinstance(tool_calls, list): - raise Exception("Prop `tool_calls` must be a list of tool calls") - new_messages.append( - adapt_messages_to_generic_oci_standard_tool_call(role, tool_calls) - ) - - elif role in ["system", "user", "assistant"] and content is not None: - if not isinstance(content, (str, list)): - raise Exception( - "Prop `content` must be a string or a list of content items" - ) - new_messages.append( - adapt_messages_to_generic_oci_standard_content_message(role, content) - ) - - elif role == "tool": - if not isinstance(tool_call_id, str): - raise Exception("Prop `tool_call_id` is required and must be a string") - if not isinstance(content, str): - raise Exception("Prop `content` is not a string") - new_messages.append( - adapt_messages_to_generic_oci_standard_tool_response( - role, tool_call_id, content - ) - ) - - return new_messages - - -def adapt_tool_definition_to_oci_standard(tools: List[Dict], vendor: OCIVendors): - new_tools = [] - for tool in tools: - if tool["type"] != "function": - raise Exception("OCI only supports function tools") - - tool_function = tool.get("function") - if not isinstance(tool_function, dict): - raise Exception("Prop `function` is not a dictionary") - - new_tool = OCIToolDefinition( - type="FUNCTION", - name=tool_function.get("name"), - description=tool_function.get("description", ""), - parameters=tool_function.get("parameters", {}), - ) - new_tools.append(new_tool) - - return new_tools - - -def adapt_tools_to_openai_standard( - tools: List[OCIToolCall], -) -> List[ChatCompletionMessageToolCall]: - new_tools = [] - for tool in tools: - new_tool = ChatCompletionMessageToolCall( - id=tool.id, - type="function", - function={ - "name": tool.name, - "arguments": tool.arguments, - }, - ) - new_tools.append(new_tool) - return new_tools - - class OCIStreamWrapper(CustomStreamWrapper): - """ - Custom stream wrapper for OCI responses. - This class is used to handle streaming responses from OCI's API. - """ + """Custom stream wrapper that dispatches OCI SSE chunks to the correct handler.""" - def __init__( - self, - **kwargs: Any, - ): + def __init__(self, **kwargs: Any): super().__init__(**kwargs) + # Tracks whether any prior Cohere chunk in this stream has emitted + # tool calls. The Cohere handler uses this to decide whether the + # terminal consolidation chunk's tool calls are duplicates (suppress) + # or the only copy of the tool calls (pass through). + self._cohere_tool_calls_emitted = False + # Analogous flag for text content. Lets the Cohere handler distinguish + # the common case (prior deltas already streamed the text, so the + # terminal chunk's text is a duplicate to suppress) from the degenerate + # single-event case (terminal chunk carries the only copy of the text). + self._cohere_text_emitted = False - def chunk_creator(self, chunk: Any): + def chunk_creator(self, chunk: Any) -> ModelResponseStream: if not isinstance(chunk, str): raise ValueError(f"Chunk is not a string: {chunk}") if not chunk.startswith("data:"): raise ValueError(f"Chunk does not start with 'data:': {chunk}") - dict_chunk = json.loads(chunk[5:]) # Remove 'data: ' prefix and parse JSON - - # Check if this is a Cohere stream chunk - if "apiFormat" in dict_chunk and dict_chunk.get("apiFormat") == "COHERE": - return self._handle_cohere_stream_chunk(dict_chunk) - else: - return self._handle_generic_stream_chunk(dict_chunk) - - def _handle_cohere_stream_chunk(self, dict_chunk: dict): - """Handle Cohere-specific streaming chunks.""" try: - typed_chunk = CohereStreamChunk(**dict_chunk) - except TypeError as e: - raise ValueError(f"Chunk cannot be casted to CohereStreamChunk: {str(e)}") + dict_chunk = json.loads(chunk[5:]) + except json.JSONDecodeError as e: + raise OCIError( + status_code=500, + message=f"Chunk cannot be parsed as JSON: {str(e)}", + ) - if typed_chunk.index is None: - typed_chunk.index = 0 + if dict_chunk.get("apiFormat") == "COHERE": + result = handle_cohere_stream_chunk( + dict_chunk, + prior_tool_calls_emitted=self._cohere_tool_calls_emitted, + prior_text_emitted=self._cohere_text_emitted, + ) + if not self._cohere_tool_calls_emitted: + for choice in result.choices: + if getattr(choice.delta, "tool_calls", None) is not None: + self._cohere_tool_calls_emitted = True + break + if not self._cohere_text_emitted: + for choice in result.choices: + if getattr(choice.delta, "content", None): + self._cohere_text_emitted = True + break + return result + return handle_generic_stream_chunk(dict_chunk) - # Extract text content - text = typed_chunk.text or "" - # Map finish reason to standard format - finish_reason = typed_chunk.finishReason - if finish_reason == "COMPLETE": - finish_reason = "stop" - elif finish_reason == "MAX_TOKENS": - finish_reason = "length" - elif finish_reason is None: - finish_reason = None - else: - finish_reason = "stop" - - # For Cohere, we don't have tool calls in the streaming format - tool_calls = None - - return ModelResponseStream( - choices=[ - StreamingChoices( - index=typed_chunk.index if typed_chunk.index else 0, - delta=Delta( - content=text, - tool_calls=tool_calls, - provider_specific_fields=None, - thinking_blocks=None, - reasoning_content=None, - ), - finish_reason=finish_reason, - ) - ] - ) - - def _handle_generic_stream_chunk(self, dict_chunk: dict): - """Handle generic OCI streaming chunks.""" - # Fix missing required fields in tool calls before Pydantic validation - # OCI streams tool calls progressively, so early chunks may be missing required fields - if dict_chunk.get("message") and dict_chunk["message"].get("toolCalls"): - for tool_call in dict_chunk["message"]["toolCalls"]: - if "arguments" not in tool_call: - tool_call["arguments"] = "" - if "id" not in tool_call: - tool_call["id"] = "" - if "name" not in tool_call: - tool_call["name"] = "" - - try: - typed_chunk = OCIStreamChunk(**dict_chunk) - except TypeError as e: - raise ValueError(f"Chunk cannot be casted to OCIStreamChunk: {str(e)}") - - if typed_chunk.index is None: - typed_chunk.index = 0 - - text = "" - if typed_chunk.message and typed_chunk.message.content: - for item in typed_chunk.message.content: - if isinstance(item, OCITextContentPart): - text += item.text - elif isinstance(item, OCIImageContentPart): - raise ValueError( - "OCI does not support image content in streaming responses" - ) - else: - raise ValueError( - f"Unsupported content type in OCI response: {item.type}" - ) - - tool_calls = None - if typed_chunk.message and typed_chunk.message.toolCalls: - tool_calls = adapt_tools_to_openai_standard(typed_chunk.message.toolCalls) - - return ModelResponseStream( - choices=[ - StreamingChoices( - index=typed_chunk.index if typed_chunk.index else 0, - delta=Delta( - content=text, - tool_calls=( - [tool.model_dump() for tool in tool_calls] - if tool_calls - else None - ), - provider_specific_fields=None, # OCI does not have provider specific fields in the response - thinking_blocks=None, # OCI does not have thinking blocks in the response - reasoning_content=None, # OCI does not have reasoning content in the response - ), - finish_reason=typed_chunk.finishReason, - ) - ] - ) +__all__ = [ + "OCIChatConfig", + "OCIStreamWrapper", + "OCIRequestWrapper", + "OCI_API_VERSION", + "STREAMING_TIMEOUT", + "get_vendor_from_model", + "version", +] diff --git a/litellm/llms/oci/common_utils.py b/litellm/llms/oci/common_utils.py index 661a6c89e4b..8785b1548a5 100644 --- a/litellm/llms/oci/common_utils.py +++ b/litellm/llms/oci/common_utils.py @@ -1,9 +1,42 @@ -from typing import Optional +import base64 +import hashlib +import json +import os +import re +from dataclasses import dataclass +from email.utils import formatdate +from typing import Any, Dict, Optional, Protocol, Tuple +from urllib.parse import urlparse import httpx from litellm.llms.base_llm.chat.transformation import BaseLLMException +try: + from cryptography.hazmat.primitives import hashes, serialization + from cryptography.hazmat.primitives.asymmetric import padding, rsa + + _CRYPTOGRAPHY_AVAILABLE = True +except ImportError: + _CRYPTOGRAPHY_AVAILABLE = False + +try: + from litellm._version import version as _litellm_version +except ImportError: + _litellm_version = "0.0.0" + + +# OCI GenAI REST API version — stable since service launch, unlikely to change +OCI_API_VERSION = "20231130" + + +def _require_cryptography() -> None: + if not _CRYPTOGRAPHY_AVAILABLE: + raise ImportError( + "cryptography package is required for OCI authentication. " + "Please install it with: pip install cryptography" + ) + class OCIError(BaseLLMException): def __init__( @@ -17,3 +50,520 @@ class OCIError(BaseLLMException): message=message, headers=headers, ) + + +# --------------------------------------------------------------------------- +# OCI signing protocol and helpers +# --------------------------------------------------------------------------- + + +class OCISignerProtocol(Protocol): + """ + Protocol for OCI request signers (e.g., oci.signer.Signer). + + Compatible with the OCI Python SDK's Signer class. + See: https://docs.oracle.com/en-us/iaas/tools/python/latest/api/signing.html + """ + + def do_request_sign( + self, request: Any, *, enforce_content_headers: bool = False + ) -> None: + pass + + +@dataclass +class OCIRequestWrapper: + """ + Wrapper for HTTP requests compatible with OCI signer interface. + + Wraps request data in the format expected by OCI SDK signers, which require + objects with method, url, headers, body, and path_url attributes. + """ + + method: str + url: str + headers: dict + body: bytes + + @property + def path_url(self) -> str: + """Returns the path + query string for OCI signing.""" + parsed = urlparse(self.url) + return parsed.path + ("?" + parsed.query if parsed.query else "") + + +def sha256_base64(data: bytes) -> str: + # SHA-256 is used here to compute the x-content-sha256 header required by the + # OCI HTTP signing specification (RSA-SHA256 request signing), not for password + # or secret hashing. This is the correct and mandated algorithm for this purpose. + # See: https://docs.oracle.com/en-us/iaas/Content/API/Concepts/signingrequests.htm + # + # ``usedforsecurity=False`` declares non-security intent to static analyzers + # (CodeQL ``py/weak-sensitive-data-hashing``) — without it the request body + # gets flagged as "password-like data" via taint tracking. + digest = hashlib.sha256(data, usedforsecurity=False).digest() # noqa: S324 + return base64.b64encode(digest).decode() + + +def build_signature_string( + method: str, path: str, headers: dict, signed_headers: list +) -> str: + lines = [] + for header in signed_headers: + if header == "(request-target)": + value = f"{method.lower()} {path}" + else: + value = headers[header] + lines.append(f"{header}: {value}") + return "\n".join(lines) + + +def load_private_key_from_str(key_str: str) -> Any: + _require_cryptography() + key = serialization.load_pem_private_key( # type: ignore[union-attr] + key_str.encode("utf-8"), + password=None, + ) + if not isinstance(key, rsa.RSAPrivateKey): # type: ignore[union-attr] + raise TypeError( + "The provided private key is not an RSA key, which is required for OCI signing." + ) + return key + + +def load_private_key_from_file(file_path: str) -> Any: + """Loads a private key from a file path.""" + try: + with open(file_path, "r", encoding="utf-8") as f: + key_str = f.read().strip() + except FileNotFoundError: + raise FileNotFoundError(f"Private key file not found: {file_path}") + except OSError as e: + raise OSError(f"Failed to read private key file '{file_path}': {e}") from e + + if not key_str: + raise ValueError(f"Private key file is empty: {file_path}") + + return load_private_key_from_str(key_str) + + +# --------------------------------------------------------------------------- +# Env-var credential resolution +# --------------------------------------------------------------------------- + +_OCI_REGION_ENV = "OCI_REGION" +_OCI_USER_ENV = "OCI_USER" +_OCI_FINGERPRINT_ENV = "OCI_FINGERPRINT" +_OCI_TENANCY_ENV = "OCI_TENANCY" +_OCI_KEY_FILE_ENV = "OCI_KEY_FILE" +_OCI_KEY_ENV = "OCI_KEY" +_OCI_COMPARTMENT_ID_ENV = "OCI_COMPARTMENT_ID" + + +def resolve_oci_credentials(optional_params: dict) -> dict: + """ + Merge OCI credentials from optional_params (explicit, always wins) and + environment variables (fallback). + + Returns a dict with resolved values for: + oci_region, oci_user, oci_fingerprint, oci_tenancy, + oci_key, oci_key_file, oci_compartment_id + """ + return { + "oci_region": optional_params.get("oci_region") + or os.environ.get(_OCI_REGION_ENV) + or "us-ashburn-1", + "oci_user": optional_params.get("oci_user") or os.environ.get(_OCI_USER_ENV), + "oci_fingerprint": optional_params.get("oci_fingerprint") + or os.environ.get(_OCI_FINGERPRINT_ENV), + "oci_tenancy": optional_params.get("oci_tenancy") + or os.environ.get(_OCI_TENANCY_ENV), + "oci_key": optional_params.get("oci_key") or os.environ.get(_OCI_KEY_ENV), + "oci_key_file": optional_params.get("oci_key_file") + or os.environ.get(_OCI_KEY_FILE_ENV), + "oci_compartment_id": optional_params.get("oci_compartment_id") + or os.environ.get(_OCI_COMPARTMENT_ID_ENV), + } + + +_OCI_REGION_RE = re.compile(r"^[a-z][a-z0-9-]{0,30}[a-z0-9]$") +_OCI_ACTION_PATH_RE = re.compile(rf"/{OCI_API_VERSION}/actions/[^/?#]+/?$") + + +def get_oci_base_url(optional_params: dict, api_base: Optional[str] = None) -> str: + """Return the OCI inference base URL, respecting any explicit api_base override. + + If ``api_base`` already ends with a fully-formed OCI action path + (``/{OCI_API_VERSION}/actions/``), that suffix is stripped so callers + can append their own action path without producing a doubled URL. + """ + if api_base: + return _OCI_ACTION_PATH_RE.sub("", api_base).rstrip("/") + creds = resolve_oci_credentials(optional_params) + region = creds["oci_region"] + if not isinstance(region, str) or not _OCI_REGION_RE.match(region): + raise OCIError( + status_code=400, + message=( + f"Invalid OCI region {region!r}: must match " + "^[a-z][a-z0-9-]{0,30}[a-z0-9]$ (e.g. 'us-ashburn-1')." + ), + ) + return f"https://inference.generativeai.{region}.oci.oraclecloud.com" + + +# --------------------------------------------------------------------------- +# Signing implementations (shared by chat, embed, and rerank configs) +# --------------------------------------------------------------------------- + + +def sign_with_oci_signer( + headers: dict, + optional_params: dict, + request_data: dict, + api_base: str, +) -> Tuple[dict, bytes]: + """Sign a request using an OCI SDK Signer object passed in optional_params.""" + oci_signer = optional_params.get("oci_signer") + body = json.dumps(request_data).encode("utf-8") + method = str(optional_params.get("method", "POST")).upper() + + if method not in {"POST", "GET", "PUT", "DELETE", "PATCH"}: + raise ValueError(f"Unsupported HTTP method: {method}") + + prepared_headers = {**headers} + prepared_headers.setdefault("content-type", "application/json") + prepared_headers.setdefault("content-length", str(len(body))) + + request_wrapper = OCIRequestWrapper( + method=method, url=api_base, headers=prepared_headers, body=body + ) + + if oci_signer is None: + raise ValueError("oci_signer cannot be None when calling sign_with_oci_signer") + + try: + oci_signer.do_request_sign(request_wrapper, enforce_content_headers=True) + except Exception as e: + raise OCIError( + status_code=500, + message=( + f"Failed to sign request with provided oci_signer: {str(e)}. " + "The signer must implement the OCI SDK Signer interface with a " + "do_request_sign(request, enforce_content_headers=True) method. " + "See: https://docs.oracle.com/en-us/iaas/tools/python/latest/api/signing.html" + ), + ) from e + + headers.update(request_wrapper.headers) + return headers, body + + +def sign_with_manual_credentials( + headers: dict, + optional_params: dict, + request_data: dict, + api_base: str, +) -> Tuple[dict, bytes]: + """Sign a request using manually provided OCI credentials (user/fingerprint/tenancy/key).""" + creds = resolve_oci_credentials(optional_params) + oci_user = creds["oci_user"] + oci_fingerprint = creds["oci_fingerprint"] + oci_tenancy = creds["oci_tenancy"] + oci_key = creds["oci_key"] + oci_key_file = creds["oci_key_file"] + + if ( + not oci_user + or not oci_fingerprint + or not oci_tenancy + or not (oci_key or oci_key_file) + ): + raise OCIError( + status_code=401, + message=( + "Missing required OCI credentials: oci_user, oci_fingerprint, oci_tenancy, " + "and at least one of oci_key or oci_key_file. " + "These can also be supplied via environment variables: " + f"{_OCI_USER_ENV}, {_OCI_FINGERPRINT_ENV}, {_OCI_TENANCY_ENV}, {_OCI_KEY_ENV} (or {_OCI_KEY_FILE_ENV}). " + "Alternatively, provide an oci_signer object from the OCI SDK." + ), + ) + + method = str(optional_params.get("method", "POST")).upper() + body = json.dumps(request_data).encode("utf-8") + parsed = urlparse(api_base) + path = parsed.path or "/" + host = parsed.netloc + + date = formatdate(usegmt=True) + content_type = headers.get("content-type", "application/json") + content_length = str(len(body)) + x_content_sha256 = sha256_base64(body) + + headers_to_sign: Dict[str, str] = { + "date": date, + "host": host, + "content-type": content_type, + "content-length": content_length, + "x-content-sha256": x_content_sha256, + } + + signed_header_names = [ + "date", + "(request-target)", + "host", + "content-length", + "content-type", + "x-content-sha256", + ] + signing_string = build_signature_string( + method, path, headers_to_sign, signed_header_names + ) + + _require_cryptography() + + # Resolve the private key — prefer inline PEM content over file path + oci_key_content: Optional[str] = None + if oci_key: + if not isinstance(oci_key, str): + raise OCIError( + status_code=400, + message=( + f"oci_key must be a string containing the PEM private key content. " + f"Got type: {type(oci_key).__name__}" + ), + ) + oci_key_content = oci_key.replace("\\n", "\n").replace("\r\n", "\n") + + private_key = ( + load_private_key_from_str(oci_key_content) + if oci_key_content + else load_private_key_from_file(oci_key_file) if oci_key_file else None + ) + + if private_key is None: + raise OCIError( + status_code=400, + message="Private key is required for OCI authentication. Provide either oci_key or oci_key_file.", + ) + + signature = private_key.sign( + signing_string.encode("utf-8"), + padding.PKCS1v15(), # type: ignore[union-attr] + hashes.SHA256(), # type: ignore[union-attr] + ) + signature_b64 = base64.b64encode(signature).decode() + + key_id = f"{oci_tenancy}/{oci_user}/{oci_fingerprint}" + authorization = ( + 'Signature version="1",' + f'keyId="{key_id}",' + 'algorithm="rsa-sha256",' + f'headers="{" ".join(signed_header_names)}",' + f'signature="{signature_b64}"' + ) + + headers.update( + { + "authorization": authorization, + "date": date, + "host": host, + "content-type": content_type, + "content-length": content_length, + "x-content-sha256": x_content_sha256, + } + ) + return headers, body + + +def sign_oci_request( + headers: dict, + optional_params: dict, + request_data: dict, + api_base: str, + api_key: Optional[str] = None, + model: Optional[str] = None, + stream: Optional[bool] = None, + fake_stream: Optional[bool] = None, +) -> Tuple[dict, bytes]: + """ + Route to the appropriate OCI signing method based on what credentials are present. + + If ``oci_signer`` is in optional_params, use the OCI SDK signer object. + Otherwise use manual RSA-SHA256 signing with explicit credentials (which can + also be supplied via OCI_* environment variables). + + Returns: + Tuple of (signed_headers, signed_body_bytes) + """ + if optional_params.get("oci_signer") is not None: + return sign_with_oci_signer(headers, optional_params, request_data, api_base) + return sign_with_manual_credentials( + headers, optional_params, request_data, api_base + ) + + +def validate_oci_environment( + headers: dict, + optional_params: dict, + api_key: Optional[str] = None, +) -> dict: + """ + Populate common OCI request headers (content-type, user-agent). + + Full credential validation is deferred to signing time so that credentials + supplied via environment variables are resolved at call time rather than + at construction time. + """ + headers.setdefault("content-type", "application/json") + headers.setdefault("user-agent", f"litellm/{_litellm_version}") + return headers + + +# --------------------------------------------------------------------------- +# JSON schema utilities for OCI tool definitions +# +# OCI Generative AI does not support JSON Schema extensions ($ref, $defs, +# anyOf). Pydantic v2 emits all three for models with Optional fields or +# nested schemas. The helpers below are ported from the official +# langchain-oracle reference implementation so that tool schemas are always +# valid before they reach the OCI endpoint. +# --------------------------------------------------------------------------- + +# Mapping from JSON Schema type names to Python type names, as expected by +# the OCI Cohere API's CohereParameterDefinition.type field. +OCI_JSON_TO_PYTHON_TYPES: Dict[str, str] = { + "string": "str", + "number": "float", + "boolean": "bool", + "integer": "int", + "array": "List", + "object": "Dict", + "any": "any", +} + + +def resolve_oci_schema_refs(schema: Dict[str, Any]) -> Dict[str, Any]: + """Inline all ``$ref``/``$defs`` references — OCI does not support JSON Schema ``$ref``.""" + defs = schema.get("$defs", {}) + resolving_stack: set = set() + + def _resolve(obj: Any) -> Any: + if isinstance(obj, dict): + if "$ref" in obj: + ref = obj["$ref"] + if ref.startswith("#/$defs/"): + key = ref.split("/")[-1] + if key in resolving_stack: + return {"type": "object"} # break cycles + resolving_stack.add(key) + try: + return _resolve(defs.get(key, obj)) + finally: + resolving_stack.discard(key) + return obj # external $ref — leave unchanged + return {k: _resolve(v) for k, v in obj.items()} + if isinstance(obj, list): + return [_resolve(item) for item in obj] + return obj + + resolved = _resolve(schema) + if isinstance(resolved, dict): + resolved.pop("$defs", None) + return resolved + + +def resolve_oci_schema_anyof(obj: Any) -> Any: + """Resolve Pydantic v2 ``Optional[T]`` → ``anyOf`` patterns. + + Pydantic v2 emits ``{"anyOf": [{"type": "T"}, {"type": "null"}]}`` for + ``Optional[T]``. OCI models don't understand ``anyOf``, so we pick the + first non-null branch and merge top-level metadata into it. + """ + if isinstance(obj, dict): + if "anyOf" in obj and "type" not in obj: + non_null = [ + t + for t in obj["anyOf"] + if not (isinstance(t, dict) and t.get("type") == "null") + ] + if non_null: + resolved = {**obj, **non_null[0]} + resolved.pop("anyOf", None) + return resolve_oci_schema_anyof(resolved) + return {k: resolve_oci_schema_anyof(v) for k, v in obj.items()} + if isinstance(obj, list): + return [resolve_oci_schema_anyof(item) for item in obj] + return obj + + +def sanitize_oci_schema(schema: Any) -> Any: + """Recursively remove OCI-incompatible fields from a JSON schema. + + Strips ``title`` keys, removes ``None``-valued ``default`` entries, + normalises ``type: [T, "null"]`` list types, and ensures arrays carry an + ``items`` definition. + """ + if isinstance(schema, list): + return [sanitize_oci_schema(item) for item in schema] + if not isinstance(schema, dict): + return schema + + sanitized: Dict[str, Any] = {} + for key, value in schema.items(): + if key == "title": + continue + if key == "default" and value is None: + continue + if key == "type": + if value == "any": + sanitized[key] = "object" + continue + if isinstance(value, list): + non_null = [t for t in value if t != "null"] + sanitized[key] = non_null[0] if non_null else "string" + continue + sanitized[key] = sanitize_oci_schema(value) + + if sanitized.get("type") == "array" and "items" not in sanitized: + sanitized["items"] = {"type": "object"} + + required = sanitized.get("required") + properties = sanitized.get("properties") + if "required" in sanitized: + if isinstance(required, list) and isinstance(properties, dict): + sanitized["required"] = [ + f for f in required if isinstance(f, str) and f in properties + ] + elif not isinstance(required, list): + sanitized["required"] = [] + + return sanitized + + +def enrich_cohere_param_description( + description: str, param_schema: Dict[str, Any] +) -> str: + """Embed schema constraints into a Cohere parameter description. + + ``CohereParameterDefinition`` only has ``type``, ``description``, and + ``isRequired``. Rich constraints (``enum``, ``format``, ``minimum``, + ``maximum``, ``pattern``) are appended to the description string so the + model can still see and respect them. + """ + parts = [description] if description else [] + if "enum" in param_schema: + parts.append(f"Allowed values: {param_schema['enum']}") + if "format" in param_schema: + parts.append(f"Format: {param_schema['format']}") + if "minimum" in param_schema or "maximum" in param_schema: + range_parts = [] + if "minimum" in param_schema: + range_parts.append(f"min={param_schema['minimum']}") + if "maximum" in param_schema: + range_parts.append(f"max={param_schema['maximum']}") + parts.append(f"Range: {', '.join(range_parts)}") + if "pattern" in param_schema: + parts.append(f"Pattern: {param_schema['pattern']}") + return ". ".join(parts) if parts else "" diff --git a/litellm/llms/oci/embed/transformation.py b/litellm/llms/oci/embed/transformation.py index 1dcd8c5213c..6cfa85b4bc4 100644 --- a/litellm/llms/oci/embed/transformation.py +++ b/litellm/llms/oci/embed/transformation.py @@ -1,8 +1,14 @@ """ -OCI Generative AI Embedding Configuration +OCI Generative AI — Embedding transformation. -Supports embedding models available on Oracle Cloud Infrastructure Generative AI service. -Uses the same authentication mechanisms as OCI chat (manual signing or OCI SDK Signer). +Endpoint: POST /20231130/actions/embedText +Supported models: cohere.embed-english-v3.0, cohere.embed-multilingual-v3.0, +cohere.embed-v4.0, and all other Cohere embed variants available on OCI +(including dedicated endpoints). + +Authentication follows the same RSA-SHA256 / OCI SDK signer pattern as chat. +The base handler (base_llm_http_handler.embedding) calls sign_request after +building the body, so signing happens automatically. Supported models: - cohere.embed-english-v3.0 @@ -10,25 +16,45 @@ Supported models: - cohere.embed-multilingual-v3.0 - cohere.embed-multilingual-light-v3.0 - cohere.embed-english-image-v3.0 -- cohere.embed-english-light-image-v3.0 -- cohere.embed-multilingual-light-image-v3.0 +- cohere.embed-multilingual-image-v3.0 - cohere.embed-v4.0 Reference: https://docs.oracle.com/en-us/iaas/api/#/en/generative-ai-inference/latest/EmbedTextResult/EmbedText """ -from typing import Any, Dict, List, Optional, Union +from typing import TYPE_CHECKING, Any, List, Optional, Tuple, Union import httpx -from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +import litellm from litellm.llms.base_llm.chat.transformation import BaseLLMException from litellm.llms.base_llm.embedding.transformation import BaseEmbeddingConfig -from litellm.llms.oci.chat.transformation import OCIChatConfig -from litellm.llms.oci.common_utils import OCIError +from litellm.llms.oci.common_utils import ( + OCI_API_VERSION, + OCIError, + get_oci_base_url, + resolve_oci_credentials, + sign_oci_request, + validate_oci_environment, +) +from litellm.types.llms.oci import ( + OCIEmbedRequest, + OCIEmbedResponse, + OCIServingMode, +) from litellm.types.llms.openai import AllEmbeddingInputValues, AllMessageValues from litellm.types.utils import EmbeddingResponse, Usage +if TYPE_CHECKING: + from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj + + LiteLLMLoggingObj = _LiteLLMLoggingObj +else: + LiteLLMLoggingObj = Any + +# OCI sends up to 96 texts per embedText request (Cohere limit). +OCI_EMBED_BATCH_LIMIT = 96 + # Input type mapping from OpenAI conventions to OCI/Cohere conventions _INPUT_TYPE_MAP = { "search_document": "SEARCH_DOCUMENT", @@ -38,65 +64,43 @@ _INPUT_TYPE_MAP = { } -class OCIEmbeddingConfig(BaseEmbeddingConfig): +class OCIEmbedConfig(BaseEmbeddingConfig): """ - Configuration for OCI Generative AI Embedding API. + Transformation config for OCI Generative AI embeddings. - The OCI embedding endpoint uses the Cohere embed models hosted on OCI. - Authentication is handled via OCI request signing (manual credentials or OCI SDK Signer). + Supports both text and (on cohere.embed-v4.0) multimodal inputs. - Usage: - ```python - import litellm + Authentication — same two modes as chat: + - **OCI SDK signer**: pass ``oci_signer`` in optional_params. + - **Manual RSA-SHA256**: pass ``oci_user``, ``oci_fingerprint``, ``oci_tenancy``, + and ``oci_key`` or ``oci_key_file``, or set the corresponding ``OCI_*`` env vars. - response = litellm.embedding( - model="oci/cohere.embed-english-v3.0", - input=["Hello world", "Goodbye world"], - oci_compartment_id="ocid1.compartment.oc1..xxx", - oci_region="us-ashburn-1", - oci_user="ocid1.user.oc1..xxx", - oci_fingerprint="xx:xx:xx:xx", - oci_tenancy="ocid1.tenancy.oc1..xxx", - oci_key_file="~/.oci/key.pem", - ) - ``` + Required call-time params (via optional_params or env vars): + - ``oci_compartment_id`` / ``OCI_COMPARTMENT_ID`` + - ``oci_region`` / ``OCI_REGION`` (default: ``us-ashburn-1``) + + Optional call-time params: + - ``oci_serving_mode``: ``"ON_DEMAND"`` (default) or ``"DEDICATED"`` + - ``oci_endpoint_id``: endpoint OCID for dedicated serving mode + - ``input_type``: ``SEARCH_DOCUMENT``, ``SEARCH_QUERY``, ``CLASSIFICATION``, ``CLUSTERING`` + - ``truncate``: ``NONE``, ``START``, or ``END`` (default ``END``) + - ``dimensions``: output embedding dimensions (cohere.embed-v4.0+) """ - def __init__(self) -> None: - # We reuse OCIChatConfig for signing logic - self._chat_config = OCIChatConfig() - - def get_complete_url( - self, - api_base: Optional[str], - api_key: Optional[str], - model: str, - optional_params: dict, - litellm_params: dict, - stream: Optional[bool] = None, - ) -> str: - if api_base: - return api_base - - oci_region = optional_params.get("oci_region", "us-ashburn-1") - return f"https://inference.generativeai.{oci_region}.oci.oraclecloud.com/20231130/actions/embedText" - - def get_supported_openai_params(self, model: str) -> list: - return [ - "dimensions", - ] + def get_supported_openai_params(self, model: str) -> List[str]: + return ["dimensions"] def map_openai_params( self, non_default_params: dict, optional_params: dict, model: str, - drop_params: bool, + drop_params: bool = False, ) -> dict: - # Note: OCI Cohere embed does not support custom dimensions natively, - # but we pass it through in case future models support it - if "dimensions" in non_default_params: - optional_params["dimensions"] = non_default_params["dimensions"] + for key, value in non_default_params.items(): + if key == "dimensions": + # OCI API uses outputDimensions (cohere.embed-v4.0+) + optional_params["outputDimensions"] = value return optional_params def validate_environment( @@ -109,49 +113,42 @@ class OCIEmbeddingConfig(BaseEmbeddingConfig): api_key: Optional[str] = None, api_base: Optional[str] = None, ) -> dict: - """ - Validate OCI credentials for embedding requests. - Supports both OCI SDK Signer and manual credential signing. - """ - oci_signer = optional_params.get("oci_signer") - oci_region = optional_params.get("oci_region", "us-ashburn-1") - - api_base = ( - api_base - or f"https://inference.generativeai.{oci_region}.oci.oraclecloud.com" - ) - - if oci_signer is None: - oci_user = optional_params.get("oci_user") - oci_fingerprint = optional_params.get("oci_fingerprint") - oci_tenancy = optional_params.get("oci_tenancy") - oci_key = optional_params.get("oci_key") - oci_key_file = optional_params.get("oci_key_file") - oci_compartment_id = optional_params.get("oci_compartment_id") - - if ( - not oci_user - or not oci_fingerprint - or not oci_tenancy - or not (oci_key or oci_key_file) - or not oci_compartment_id - ): - raise Exception( - "Missing required parameters: oci_user, oci_fingerprint, oci_tenancy, oci_compartment_id " - "and at least one of oci_key or oci_key_file. " - "Alternatively, provide an oci_signer object from the OCI SDK." + if optional_params.get("oci_signer") is None: + creds = resolve_oci_credentials(optional_params) + missing = [ + k + for k in ( + "oci_user", + "oci_fingerprint", + "oci_tenancy", + "oci_compartment_id", ) + if not creds.get(k) + ] + if missing or not (creds.get("oci_key") or creds.get("oci_key_file")): + raise OCIError( + status_code=401, + message=( + "Missing required parameters: oci_user, oci_fingerprint, oci_tenancy, " + "oci_compartment_id and at least one of oci_key or oci_key_file. " + "These can be supplied via optional_params or via OCI_USER, OCI_FINGERPRINT, " + "OCI_TENANCY, OCI_COMPARTMENT_ID, OCI_KEY_FILE environment variables. " + "Alternatively, provide an oci_signer object from the OCI SDK." + ), + ) + return validate_oci_environment(headers, optional_params, api_key) - from litellm.llms.custom_httpx.http_handler import version - - headers.update( - { - "content-type": "application/json", - "user-agent": f"litellm/{version}", - } - ) - - return headers + def get_complete_url( + self, + api_base: Optional[str], + api_key: Optional[str], + model: str, + optional_params: dict, + litellm_params: dict, + stream: Optional[bool] = None, + ) -> str: + base = get_oci_base_url(optional_params, api_base or litellm.api_base) + return f"{base}/{OCI_API_VERSION}/actions/embedText" def sign_request( self, @@ -163,9 +160,8 @@ class OCIEmbeddingConfig(BaseEmbeddingConfig): model: Optional[str] = None, stream: Optional[bool] = None, fake_stream: Optional[bool] = None, - ): - """Delegate to OCIChatConfig's signing logic.""" - return self._chat_config.sign_request( + ) -> Tuple[dict, bytes]: + return sign_oci_request( headers=headers, optional_params=optional_params, request_data=request_data, @@ -182,91 +178,74 @@ class OCIEmbeddingConfig(BaseEmbeddingConfig): input: AllEmbeddingInputValues, optional_params: dict, headers: dict, - api_base: Optional[str] = None, ) -> dict: - """ - Transform the embedding request to OCI format. - - OCI embedText API expects: - { - "compartmentId": "...", - "servingMode": {"servingType": "ON_DEMAND", "modelId": "..."}, - "inputs": ["text1", "text2"], - "truncate": "END", - "inputType": "SEARCH_DOCUMENT" - } - """ - oci_compartment_id = optional_params.get("oci_compartment_id") - if not oci_compartment_id: - raise Exception( - "kwarg `oci_compartment_id` is required for OCI embedding requests" + creds = resolve_oci_credentials(optional_params) + compartment_id = creds["oci_compartment_id"] + if not compartment_id: + raise OCIError( + status_code=400, + message=( + "oci_compartment_id is required for OCI embedding requests. " + "Pass it as optional_params or set the OCI_COMPARTMENT_ID env var." + ), ) - # Build serving mode - oci_serving_mode = optional_params.get("oci_serving_mode", "ON_DEMAND") - if oci_serving_mode == "DEDICATED": - oci_endpoint_id = optional_params.get("oci_endpoint_id", model) - serving_mode = { - "servingType": "DEDICATED", - "endpointId": oci_endpoint_id, - } - else: - serving_mode = { - "servingType": "ON_DEMAND", - "modelId": model, - } - - # Normalize input to list of strings + # Normalise input to a flat list of strings if isinstance(input, str): - inputs = [input] + texts = [input] elif isinstance(input, list): - inputs = [] + texts = [] for item in input: - if isinstance(item, str): - inputs.append(item) - elif isinstance(item, list): - raise ValueError( - "OCI embedding does not support token-array inputs. " - "Please convert token lists to strings before calling embedding()." + if isinstance(item, list): + raise OCIError( + status_code=400, + message=( + "OCI embedText does not support token-array inputs. " + "Convert token lists to strings before calling embedding()." + ), ) - else: - inputs.append(str(item)) + texts.append(item if isinstance(item, str) else str(item)) else: - inputs = [str(input)] + texts = [str(input)] - # Build request data — OCI embedText API expects inputs, truncate, - # and inputType at the top level alongside compartmentId and servingMode - request_data: Dict[str, Any] = { - "compartmentId": oci_compartment_id, - "servingMode": serving_mode, - "inputs": inputs, - "truncate": optional_params.get("truncate", "END"), - } + if len(texts) > OCI_EMBED_BATCH_LIMIT: + raise OCIError( + status_code=400, + message=( + f"OCI embedText accepts at most {OCI_EMBED_BATCH_LIMIT} inputs per request " + f"(got {len(texts)}). Batch your requests." + ), + ) - # Map input_type if provided + serving_mode_type = optional_params.get("oci_serving_mode", "ON_DEMAND").upper() + if serving_mode_type not in {"ON_DEMAND", "DEDICATED"}: + raise OCIError( + status_code=400, + message="oci_serving_mode must be 'ON_DEMAND' or 'DEDICATED'.", + ) + + if serving_mode_type == "DEDICATED": + endpoint_id = optional_params.get("oci_endpoint_id", model) + serving_mode = OCIServingMode( + servingType="DEDICATED", endpointId=endpoint_id + ) + else: + serving_mode = OCIServingMode(servingType="ON_DEMAND", modelId=model) + + # Map input_type from OpenAI convention to OCI/Cohere convention input_type = optional_params.get("input_type") if input_type: - mapped_type = _INPUT_TYPE_MAP.get(input_type.lower(), input_type.upper()) - request_data["inputType"] = mapped_type + input_type = _INPUT_TYPE_MAP.get(input_type.lower(), input_type.upper()) - # Sign the request using the same URL the HTTP handler will POST to - signing_url = self.get_complete_url( - api_base=api_base, - api_key=None, - model=model, - optional_params=optional_params, - litellm_params={}, + request = OCIEmbedRequest( + compartmentId=compartment_id, + servingMode=serving_mode, + inputs=texts, + inputType=input_type, + truncate=optional_params.get("truncate", "END"), + outputDimensions=optional_params.get("outputDimensions"), ) - - signed_headers, body = self.sign_request( - headers=headers, - optional_params=optional_params, - request_data=request_data, - api_base=signing_url, - ) - headers.update(signed_headers) - - return request_data + return request.model_dump(exclude_none=True) def transform_embedding_response( self, @@ -274,63 +253,57 @@ class OCIEmbeddingConfig(BaseEmbeddingConfig): raw_response: httpx.Response, model_response: EmbeddingResponse, logging_obj: LiteLLMLoggingObj, - api_key: Optional[str] = None, - request_data: dict = {}, - optional_params: dict = {}, - litellm_params: dict = {}, + api_key: Optional[str], + request_data: dict, + optional_params: dict, + litellm_params: dict, ) -> EmbeddingResponse: - """ - Transform OCI embedding response to standard EmbeddingResponse format. - - OCI response format: - { - "embeddings": [[0.1, 0.2, ...], [0.3, 0.4, ...]], - "modelId": "cohere.embed-english-v3.0", - "modelVersion": "3.0", - "inputTextTokenCounts": [5, 4] - } - """ if raw_response.status_code != 200: raise OCIError( - message=raw_response.text, status_code=raw_response.status_code, + message=raw_response.text, ) try: - raw_response_json = raw_response.json() - except Exception: + json_response = raw_response.json() + except Exception as e: raise OCIError( - message=raw_response.text, status_code=raw_response.status_code, + message=f"Failed to parse OCI embed response as JSON: {e}", ) - embeddings = raw_response_json.get("embeddings", []) - model_id = raw_response_json.get("modelId", model) - - # Build response data in OpenAI format - embedding_data = [] - for idx, embedding in enumerate(embeddings): - embedding_data.append( - { - "object": "embedding", - "index": idx, - "embedding": embedding, - } + try: + parsed = OCIEmbedResponse(**json_response) + except Exception as e: + raise OCIError( + status_code=500, + message=f"OCI embed response does not match expected schema: {e}", ) - model_response.model = model_id - model_response.data = embedding_data - model_response.object = "list" + model_response.model = parsed.modelId + model_response.data = [ + { + "object": "embedding", + "index": i, + "embedding": embedding, + } + for i, embedding in enumerate(parsed.embeddings) + ] - # Calculate token usage - input_token_counts = raw_response_json.get("inputTextTokenCounts", []) - total_tokens = sum(input_token_counts) if input_token_counts else 0 - - usage = Usage( - prompt_tokens=total_tokens, - total_tokens=total_tokens, - ) - model_response.usage = usage + if parsed.inputTextTokenCounts is not None: + # Actual OCI API returns per-input token counts — sum for total usage + total = sum(parsed.inputTextTokenCounts) + model_response.usage = Usage(prompt_tokens=total, total_tokens=total) + elif parsed.usage is not None: + # Some deployments may return a usage object directly + model_response.usage = Usage( + prompt_tokens=parsed.usage.promptTokens, + total_tokens=parsed.usage.totalTokens, + ) + else: + # Neither field returned — default to zero so downstream consumers + # can always rely on usage being populated. + model_response.usage = Usage(prompt_tokens=0, total_tokens=0) return model_response @@ -340,8 +313,8 @@ class OCIEmbeddingConfig(BaseEmbeddingConfig): status_code: int, headers: Union[dict, httpx.Headers], ) -> BaseLLMException: - return OCIError( - message=error_message, - status_code=status_code, - headers=headers if isinstance(headers, httpx.Headers) else None, - ) + return OCIError(status_code=status_code, message=error_message) + + +# Alias for backwards compatibility with any code that imports OCIEmbeddingConfig +OCIEmbeddingConfig = OCIEmbedConfig diff --git a/litellm/llms/ollama/chat/transformation.py b/litellm/llms/ollama/chat/transformation.py index 48534799c97..e36150a4954 100644 --- a/litellm/llms/ollama/chat/transformation.py +++ b/litellm/llms/ollama/chat/transformation.py @@ -507,10 +507,10 @@ class OllamaChatCompletionResponseIterator(BaseModelResponseIterator): # PROCESS REASONING CONTENT reasoning_content: Optional[str] = None content: Optional[str] = None - if chunk["message"].get("thinking") is not None: + if chunk["message"].get("thinking"): reasoning_content = chunk["message"].get("thinking") self.started_reasoning_content = True - elif chunk["message"].get("content") is not None: + if chunk["message"].get("content"): if ( self.started_reasoning_content and not self.finished_reasoning_content diff --git a/litellm/llms/ollama/common_utils.py b/litellm/llms/ollama/common_utils.py index 8aedd9b3500..8ca8b7d383a 100644 --- a/litellm/llms/ollama/common_utils.py +++ b/litellm/llms/ollama/common_utils.py @@ -108,7 +108,7 @@ class OllamaModelInfo(BaseLLMModelInfo): continue nm = entry.get("name") or entry.get("model") if isinstance(nm, str): - names.add(nm) + names.add(nm if nm.startswith("ollama/") else f"ollama/{nm}") except Exception as e: verbose_logger.warning(f"Error retrieving ollama tag endpoint: {e}") # If tags endpoint fails, fall back to static list diff --git a/litellm/llms/openai/chat/gpt_transformation.py b/litellm/llms/openai/chat/gpt_transformation.py index 6b7ec4dfb1c..5464b5bb7ee 100644 --- a/litellm/llms/openai/chat/gpt_transformation.py +++ b/litellm/llms/openai/chat/gpt_transformation.py @@ -287,7 +287,13 @@ class OpenAIGPTConfig(BaseLLMModelInfo, BaseConfig): content_item["image_url"] = new_image_url_obj elif content_item.get("type") == "file": content_item = cast(ChatCompletionFileObject, content_item) - file_obj = content_item["file"] + file_obj = content_item.get("file") + if file_obj is None: + raise litellm.BadRequestError( + message="Content block has type='file' but is missing the required 'file' field", + model=None, + llm_provider="openai", + ) new_file_obj = ChatCompletionFileObjectFile( **{ # type: ignore k: v diff --git a/litellm/llms/openai/chat/o_series_transformation.py b/litellm/llms/openai/chat/o_series_transformation.py index 02ae2cc9750..8db7ecf7b3a 100644 --- a/litellm/llms/openai/chat/o_series_transformation.py +++ b/litellm/llms/openai/chat/o_series_transformation.py @@ -1,14 +1,14 @@ """ -Support for o1/o3 model family +Support for o1/o3 model family https://platform.openai.com/docs/guides/reasoning Translations handled by LiteLLM: -- modalities: image => drop param (if user opts in to dropping param) -- role: system ==> translate to role 'user' -- streaming => faked by LiteLLM -- Tools, response_format => drop param (if user opts in to dropping param) -- Logprobs => drop param (if user opts in to dropping param) +- modalities: image => drop param (if user opts in to dropping param) +- role: system ==> translate to role 'user' +- streaming => faked by LiteLLM +- Tools, response_format => drop param (if user opts in to dropping param) +- Logprobs => drop param (if user opts in to dropping param) """ from typing import Any, Coroutine, List, Literal, Optional, Union, cast, overload diff --git a/litellm/llms/openai/common_utils.py b/litellm/llms/openai/common_utils.py index c13a976c1b9..381f215a13f 100644 --- a/litellm/llms/openai/common_utils.py +++ b/litellm/llms/openai/common_utils.py @@ -201,7 +201,7 @@ class BaseOpenAILLM: @staticmethod def get_openai_client_initialization_param_fields( - client_type: Literal["openai", "azure"] + client_type: Literal["openai", "azure"], ) -> Tuple[str, ...]: """Returns a tuple of fields that are used to initialize the OpenAI client""" if client_type == "openai": diff --git a/litellm/llms/openai/cost_calculation.py b/litellm/llms/openai/cost_calculation.py index 32b71a43afa..6935cafd0d9 100644 --- a/litellm/llms/openai/cost_calculation.py +++ b/litellm/llms/openai/cost_calculation.py @@ -19,7 +19,10 @@ def cost_router(call_type: CallTypes) -> Literal["cost_per_token", "cost_per_sec def cost_per_token( - model: str, usage: Usage, service_tier: Optional[str] = None + model: str, + usage: Usage, + service_tier: Optional[str] = None, + data_residency: Optional[str] = None, ) -> Tuple[float, float]: """ Calculates the cost per token for a given model, prompt tokens, and completion tokens. @@ -27,6 +30,9 @@ def cost_per_token( Input: - model: str, the model name without provider prefix - usage: LiteLLM Usage block, containing anthropic caching information + - data_residency: optional OpenAI data-residency region (e.g. "eu", "us"), + inferred from api_base. Applies the model's regional-processing + uplift multiplier when set. Returns: Tuple[float, float] - prompt_cost_in_usd, completion_cost_in_usd @@ -37,6 +43,7 @@ def cost_per_token( usage=usage, custom_llm_provider="openai", service_tier=service_tier, + data_residency=data_residency, ) # ### Non-cached text tokens # non_cached_text_tokens = usage.prompt_tokens diff --git a/litellm/llms/openai/data_residency.py b/litellm/llms/openai/data_residency.py new file mode 100644 index 00000000000..7162f70ca5f --- /dev/null +++ b/litellm/llms/openai/data_residency.py @@ -0,0 +1,41 @@ +""" +Helpers for resolving OpenAI data-residency (regional processing) from an +api_base URL. + +OpenAI enforces hostname-per-region for projects with geography restrictions +enabled and rejects requests sent to the wrong host, so the api_base hostname +is the authoritative signal of which region a request was processed in. +""" + +from typing import Dict, Optional +from urllib.parse import urlparse + +# Mapping of OpenAI regional hostnames to the corresponding data-residency +# value used by the cost calculator. See +# https://developers.openai.com/api/docs/pricing for the regional-processing +# uplift these hostnames trigger. +_OPENAI_REGIONAL_HOSTS: Dict[str, str] = { + "eu.api.openai.com": "eu", + "us.api.openai.com": "us", +} + + +def infer_openai_data_residency( + custom_llm_provider: Optional[str], api_base: Optional[str] +) -> Optional[str]: + """ + Derive the OpenAI data-residency region from an api_base URL. + + Returns ``"eu"`` for the EU regional host, ``"us"`` for the US regional + host, and ``None`` for the default global host, any non-OpenAI provider, + or any non-OpenAI URL. + """ + if custom_llm_provider != "openai" or not api_base: + return None + try: + host = urlparse(api_base).hostname + except (TypeError, ValueError): + return None + if not host: + return None + return _OPENAI_REGIONAL_HOSTS.get(host.lower()) diff --git a/litellm/llms/openai/responses/transformation.py b/litellm/llms/openai/responses/transformation.py index b7d5340d8d4..5043d25ee37 100644 --- a/litellm/llms/openai/responses/transformation.py +++ b/litellm/llms/openai/responses/transformation.py @@ -126,9 +126,21 @@ class OpenAIResponsesAPIConfig(BaseResponsesAPIConfig): litellm_params: GenericLiteLLMParams, headers: dict, ) -> Dict: - """No transform applied since inputs are in OpenAI spec already""" + """Strip Anthropic-only `cache_control` markers before sending to OpenAI. + + OpenAI's Responses API rejects unknown fields on input content blocks + with HTTP 400 ("Unknown parameter: 'input[0].content[0].cache_control'"). + Chat Completions strips these in + `remove_cache_control_flag_from_messages_and_tools`; mirror that here. + """ input = self._validate_input_param(input) + tools = response_api_optional_request_params.get("tools") + input, tools = self.remove_cache_control_flag_from_input_and_tools( + model=model, input=input, tools=tools + ) + if tools is not None: + response_api_optional_request_params["tools"] = tools final_request_params = dict( ResponsesAPIRequestParams( model=model, input=input, **response_api_optional_request_params @@ -137,6 +149,38 @@ class OpenAIResponsesAPIConfig(BaseResponsesAPIConfig): return final_request_params + def remove_cache_control_flag_from_input_and_tools( + self, + model: str, # allows overrides to selectively run this + input: Union[str, ResponseInputParam], + tools: Optional[List[ALL_RESPONSES_API_TOOL_PARAMS]] = None, + ) -> Tuple[ + Union[str, ResponseInputParam], + Optional[List[ALL_RESPONSES_API_TOOL_PARAMS]], + ]: + """Sibling of `remove_cache_control_flag_from_messages_and_tools` on + the chat path. Strips Anthropic-only `cache_control` markers from + Responses API input content blocks and tools. + + `filter_value_from_dict` mutates each dict in place, so the same + objects are returned. + """ + from litellm.litellm_core_utils.prompt_templates.common_utils import ( + filter_value_from_dict, + ) + + if isinstance(input, list): + for item in input: + if isinstance(item, dict): + filter_value_from_dict(cast(dict, item), "cache_control") + + if tools is not None: + for tool in tools: + if isinstance(tool, dict): + filter_value_from_dict(cast(dict, tool), "cache_control") + + return input, tools + def _validate_input_param( self, input: Union[str, ResponseInputParam] ) -> Union[str, ResponseInputParam]: @@ -604,6 +648,12 @@ class OpenAIResponsesAPIConfig(BaseResponsesAPIConfig): url = str(parsed_url.copy_with(path=compact_path)) input = self._validate_input_param(input) + tools = response_api_optional_request_params.get("tools") + input, tools = self.remove_cache_control_flag_from_input_and_tools( + model=model, input=input, tools=tools + ) + if tools is not None: + response_api_optional_request_params["tools"] = tools data = dict( ResponsesAPIRequestParams( model=model, input=input, **response_api_optional_request_params diff --git a/litellm/llms/openai/videos/transformation.py b/litellm/llms/openai/videos/transformation.py index 2d165a7d7df..520a42e9dd1 100644 --- a/litellm/llms/openai/videos/transformation.py +++ b/litellm/llms/openai/videos/transformation.py @@ -534,6 +534,7 @@ class OpenAIVideoConfig(BaseVideoConfig): litellm_params: GenericLiteLLMParams, headers: dict, extra_body: Optional[Dict[str, Any]] = None, + prefetched_source_data: Optional[Dict[str, Any]] = None, ) -> Tuple[str, Dict]: original_video_id = extract_original_video_id(video_id) url = f"{api_base.rstrip('/')}/edits" @@ -547,6 +548,7 @@ class OpenAIVideoConfig(BaseVideoConfig): raw_response: httpx.Response, logging_obj: Any, custom_llm_provider: Optional[str] = None, + request_data: Optional[Dict] = None, ) -> VideoObject: video_obj = VideoObject(**raw_response.json()) if custom_llm_provider and video_obj.id: diff --git a/litellm/llms/openrouter/image_generation/transformation.py b/litellm/llms/openrouter/image_generation/transformation.py index a55716a5e50..9c2293eb3f1 100644 --- a/litellm/llms/openrouter/image_generation/transformation.py +++ b/litellm/llms/openrouter/image_generation/transformation.py @@ -49,7 +49,6 @@ from litellm.types.utils import ( ) from litellm.llms.openrouter.common_utils import OpenRouterException - if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj else: diff --git a/litellm/llms/reducto/__init__.py b/litellm/llms/reducto/__init__.py new file mode 100644 index 00000000000..8b137891791 --- /dev/null +++ b/litellm/llms/reducto/__init__.py @@ -0,0 +1 @@ + diff --git a/litellm/llms/reducto/common.py b/litellm/llms/reducto/common.py new file mode 100644 index 00000000000..4e7d96dbe87 --- /dev/null +++ b/litellm/llms/reducto/common.py @@ -0,0 +1,159 @@ +import base64 +import binascii +from collections import defaultdict +from typing import TYPE_CHECKING, Any, Dict, List, NoReturn, Optional, Tuple + +from litellm.constants import request_timeout + +REDUCTO_API_BASE = "https://platform.reducto.ai" +REDUCTO_ID_PREFIX = "reducto://" + +if TYPE_CHECKING: + from litellm.llms.base_llm.ocr.transformation import OCRPage + + +def _normalize_api_base(api_base: Optional[str]) -> str: + return (api_base or REDUCTO_API_BASE).rstrip("/") + + +def _raise_bad_request(message: str, model: str) -> NoReturn: + import litellm + + raise litellm.BadRequestError( + message=message, + model=model, + llm_provider="reducto", + ) + + +def extract_file_id_or_bytes( + source_url: str, + model: str, +) -> Tuple[Optional[str], Optional[bytes], Optional[str]]: + if source_url.startswith(REDUCTO_ID_PREFIX): + return source_url, None, None + + if source_url.startswith("http://") or source_url.startswith("https://"): + _raise_bad_request( + "Reducto requires type='file' (auto-uploaded) or a reducto:// id. Plain http(s) URLs are not supported; upload the file first.", + model=model, + ) + + if not source_url.startswith("data:"): + _raise_bad_request( + "Reducto requires a reducto:// id or a base64 data URI after OCR preprocessing.", + model=model, + ) + + try: + header, encoded = source_url.split(",", 1) + except ValueError: + _raise_bad_request("Invalid Reducto data URI provided.", model=model) + + if ";base64" not in header: + _raise_bad_request( + "Reducto only supports base64-encoded data URIs.", model=model + ) + + mime = header.removeprefix("data:").split(";")[0] or "application/octet-stream" + try: + raw_bytes = base64.b64decode(encoded, validate=True) + except (binascii.Error, ValueError): + _raise_bad_request("Invalid Reducto base64 payload provided.", model=model) + + return None, raw_bytes, mime + + +def _extract_file_id_from_upload_response(response: Any) -> str: + try: + payload = response.json() + except ValueError as exc: + raise ValueError( + "Reducto /upload returned a non-JSON 200 response: {}".format(response.text) + ) from exc + file_id = (payload or {}).get("file_id") if isinstance(payload, dict) else None + if not isinstance(file_id, str) or not file_id: + raise ValueError( + "Reducto /upload returned 200 without a file_id; got payload={}".format( + payload + ) + ) + return file_id + + +def upload_bytes_sync( + raw_bytes: bytes, + mime: Optional[str], + api_key: str, + api_base: Optional[str], +) -> str: + import litellm + + response = litellm.module_level_client.post( + url="{}{}".format(_normalize_api_base(api_base), "/upload"), + headers={"Authorization": f"Bearer {api_key}"}, + files={"file": ("document", raw_bytes, mime or "application/octet-stream")}, + timeout=request_timeout, + ) + response.raise_for_status() + return _extract_file_id_from_upload_response(response) + + +async def upload_bytes_async( + raw_bytes: bytes, + mime: Optional[str], + api_key: str, + api_base: Optional[str], +) -> str: + import litellm + + response = await litellm.module_level_aclient.post( + url="{}{}".format(_normalize_api_base(api_base), "/upload"), + headers={"Authorization": f"Bearer {api_key}"}, + files={"file": ("document", raw_bytes, mime or "application/octet-stream")}, + timeout=request_timeout, + ) + response.raise_for_status() + return _extract_file_id_from_upload_response(response) + + +def build_pages_from_reducto(result: Dict[str, Any]) -> List["OCRPage"]: + from litellm.llms.base_llm.ocr.transformation import OCRPage + + chunks = result.get("chunks", []) or [] + blocks_by_page: Dict[int, List[Dict[str, Any]]] = defaultdict(list) + + for chunk in chunks: + for block in chunk.get("blocks", []) or []: + page_no = (block.get("bbox") or {}).get("page") + if page_no is None: + continue + try: + normalized_page = int(page_no) + except (TypeError, ValueError): + continue + blocks_by_page[normalized_page].append(block) + + if not blocks_by_page: + fallback_markdown = "\n\n".join( + chunk.get("content", "") for chunk in chunks if chunk.get("content") + ) + if fallback_markdown == "": + return [] + return [OCRPage(index=0, markdown=fallback_markdown)] + + pages: List["OCRPage"] = [] + for page_no, blocks in sorted(blocks_by_page.items()): + markdown = "\n\n".join( + block.get("content", "") for block in blocks if block.get("content") + ) + page_index = max(page_no - 1, 0) + page = OCRPage( + index=page_index, + markdown=markdown, + ) + # OCRPage accepts extra keys at runtime; assign blocks after construction + # so static typing does not reject provider-specific metadata. + setattr(page, "blocks", blocks) + pages.append(page) + return pages diff --git a/litellm/llms/reducto/ocr/__init__.py b/litellm/llms/reducto/ocr/__init__.py new file mode 100644 index 00000000000..8b137891791 --- /dev/null +++ b/litellm/llms/reducto/ocr/__init__.py @@ -0,0 +1 @@ + diff --git a/litellm/llms/reducto/ocr/transformation.py b/litellm/llms/reducto/ocr/transformation.py new file mode 100644 index 00000000000..cc338ecc484 --- /dev/null +++ b/litellm/llms/reducto/ocr/transformation.py @@ -0,0 +1,241 @@ +from typing import Any, Dict, Optional, Tuple + +import httpx + +from litellm.llms.base_llm.ocr.transformation import ( + BaseOCRConfig, + DocumentType, + OCRRequestData, + OCRResponse, + OCRUsageInfo, +) +from litellm.llms.reducto.common import ( + REDUCTO_API_BASE, + build_pages_from_reducto, + extract_file_id_or_bytes, + upload_bytes_async, + upload_bytes_sync, +) + + +class _BaseReductoOCRConfig(BaseOCRConfig): + def map_ocr_params( + self, + non_default_params: dict, + optional_params: dict, + model: str, + ) -> dict: + mapped_params = dict(optional_params) + supported_params = self.get_supported_ocr_params(model=model) + for param, value in non_default_params.items(): + if param in supported_params: + mapped_params[param] = value + return mapped_params + + def validate_environment( + self, + headers: Dict, + model: str, + api_key: Optional[str] = None, + api_base: Optional[str] = None, + litellm_params: Optional[dict] = None, + **kwargs, + ) -> Dict: + from litellm.secret_managers.main import get_secret_str + + resolved_key = api_key or get_secret_str("REDUCTO_API_KEY") + if resolved_key is None: + raise ValueError( + "Missing REDUCTO_API_KEY - set it in the environment or pass api_key to litellm.ocr()/litellm.aocr()" + ) + + return { + "Authorization": f"Bearer {resolved_key}", + "Content-Type": "application/json", + **headers, + } + + def get_complete_url( + self, + api_base: Optional[str], + model: str, + optional_params: dict, + litellm_params: Optional[dict] = None, + **kwargs, + ) -> str: + return "{}/parse".format((api_base or REDUCTO_API_BASE).rstrip("/")) + + def _get_source_url(self, document: DocumentType, model: str) -> str: + source_url = document.get("document_url") or document.get("image_url") + if source_url is None: + raise ValueError( + "Reducto expected OCR preprocessing to produce document_url or image_url for model={}".format( + model + ) + ) + return source_url + + @staticmethod + def _resolve_credentials( + api_key: Optional[str], api_base: Optional[str] + ) -> Tuple[str, str]: + from litellm.secret_managers.main import get_secret_str + + resolved_key = api_key or get_secret_str("REDUCTO_API_KEY") + if resolved_key is None: + raise ValueError( + "Missing REDUCTO_API_KEY - set it in the environment or pass api_key to litellm.ocr()/litellm.aocr()" + ) + resolved_base = (api_base or REDUCTO_API_BASE).rstrip("/") + return resolved_key, resolved_base + + def _ensure_file_id_sync( + self, + model: str, + document: DocumentType, + api_key: Optional[str], + api_base: Optional[str], + ) -> str: + source_url = self._get_source_url(document=document, model=model) + file_id, raw_bytes, mime = extract_file_id_or_bytes(source_url, model=model) + if file_id is not None: + return file_id + resolved_key, resolved_base = self._resolve_credentials(api_key, api_base) + return upload_bytes_sync( + raw_bytes=raw_bytes or b"", + mime=mime, + api_key=resolved_key, + api_base=resolved_base, + ) + + async def _ensure_file_id_async( + self, + model: str, + document: DocumentType, + api_key: Optional[str], + api_base: Optional[str], + ) -> str: + source_url = self._get_source_url(document=document, model=model) + file_id, raw_bytes, mime = extract_file_id_or_bytes(source_url, model=model) + if file_id is not None: + return file_id + resolved_key, resolved_base = self._resolve_credentials(api_key, api_base) + return await upload_bytes_async( + raw_bytes=raw_bytes or b"", + mime=mime, + api_key=resolved_key, + api_base=resolved_base, + ) + + def transform_ocr_response( + self, + model: str, + raw_response: httpx.Response, + logging_obj: Any, + **kwargs, + ) -> OCRResponse: + response_json = raw_response.json() + result = response_json.get("result", response_json) or {} + usage = response_json.get("usage", {}) or {} + response = OCRResponse( + pages=build_pages_from_reducto(result), + model=model, + usage_info=OCRUsageInfo( + pages_processed=usage.get("num_pages"), + credits=usage.get("credits"), + ), + object="ocr", + ) + response._hidden_params["reducto_raw"] = response_json + return response + + +class ReductoParseV3Config(_BaseReductoOCRConfig): + def get_supported_ocr_params(self, model: str) -> list: + return ["formatting", "retrieval", "settings"] + + def transform_ocr_request( + self, + model: str, + document: DocumentType, + optional_params: dict, + headers: dict, + **kwargs, + ) -> OCRRequestData: + file_id = self._ensure_file_id_sync( + model=model, + document=document, + api_key=kwargs.get("api_key"), + api_base=kwargs.get("api_base"), + ) + return OCRRequestData(data={"input": file_id, **optional_params}, files=None) + + async def async_transform_ocr_request( + self, + model: str, + document: DocumentType, + optional_params: dict, + headers: dict, + **kwargs, + ) -> OCRRequestData: + file_id = await self._ensure_file_id_async( + model=model, + document=document, + api_key=kwargs.get("api_key"), + api_base=kwargs.get("api_base"), + ) + return OCRRequestData(data={"input": file_id, **optional_params}, files=None) + + +class ReductoParseLegacyConfig(_BaseReductoOCRConfig): + def get_supported_ocr_params(self, model: str) -> list: + return ["enhance"] + + def _build_legacy_body(self, file_id: str, optional_params: dict) -> Dict[str, Any]: + body: Dict[str, Any] = {"document_url": file_id} + enhance = optional_params.get("enhance") + if enhance is not None: + body["options"] = {"enhance": enhance} + return body + + def transform_ocr_request( + self, + model: str, + document: DocumentType, + optional_params: dict, + headers: dict, + **kwargs, + ) -> OCRRequestData: + file_id = self._ensure_file_id_sync( + model=model, + document=document, + api_key=kwargs.get("api_key"), + api_base=kwargs.get("api_base"), + ) + return OCRRequestData( + data=self._build_legacy_body( + file_id=file_id, optional_params=optional_params + ), + files=None, + ) + + async def async_transform_ocr_request( + self, + model: str, + document: DocumentType, + optional_params: dict, + headers: dict, + **kwargs, + ) -> OCRRequestData: + file_id = await self._ensure_file_id_async( + model=model, + document=document, + api_key=kwargs.get("api_key"), + api_base=kwargs.get("api_base"), + ) + return OCRRequestData( + data=self._build_legacy_body( + file_id=file_id, optional_params=optional_params + ), + files=None, + ) diff --git a/litellm/llms/runwayml/videos/transformation.py b/litellm/llms/runwayml/videos/transformation.py index 4f84816a2bc..b1723f494ec 100644 --- a/litellm/llms/runwayml/videos/transformation.py +++ b/litellm/llms/runwayml/videos/transformation.py @@ -623,12 +623,23 @@ class RunwayMLVideoConfig(BaseVideoConfig): raise NotImplementedError("video get character is not supported for RunwayML") def transform_video_edit_request( - self, prompt, video_id, api_base, litellm_params, headers, extra_body=None + self, + prompt, + video_id, + api_base, + litellm_params, + headers, + extra_body=None, + prefetched_source_data=None, ): raise NotImplementedError("video edit is not supported for RunwayML") def transform_video_edit_response( - self, raw_response, logging_obj, custom_llm_provider=None + self, + raw_response, + logging_obj, + custom_llm_provider=None, + request_data=None, ): raise NotImplementedError("video edit is not supported for RunwayML") diff --git a/litellm/llms/sagemaker/common_utils.py b/litellm/llms/sagemaker/common_utils.py index 50c8ee4220e..6c15d642f8c 100644 --- a/litellm/llms/sagemaker/common_utils.py +++ b/litellm/llms/sagemaker/common_utils.py @@ -1,3 +1,4 @@ +import functools import json from typing import AsyncIterator, Iterator, List, Optional, Union @@ -22,14 +23,22 @@ def _load_sagemaker_response_stream_shape(): ) except Exception as e: verbose_logger.warning( - "litellm: could not pre-load sagemaker-runtime response stream shape " + "litellm: could not load sagemaker-runtime response stream shape " "— SageMaker event-stream decoding will be unavailable. Error: %s", e, ) return None -SAGEMAKER_RESPONSE_STREAM_SHAPE = _load_sagemaker_response_stream_shape() +@functools.lru_cache(maxsize=1) +def get_sagemaker_response_stream_shape(): + """ + Lazily load and cache the sagemaker-runtime stream shape for the process. + + Avoids importing botocore (and logging warnings) unless SageMaker event-stream + decoding is actually needed. + """ + return _load_sagemaker_response_stream_shape() class SagemakerError(BaseLLMException): @@ -207,7 +216,8 @@ class AWSEventStreamDecoder: verbose_logger.error(f"Final error parsing accumulated JSON: {e}") def _parse_message_from_event(self, event) -> Optional[str]: - if SAGEMAKER_RESPONSE_STREAM_SHAPE is None: + response_stream_shape = get_sagemaker_response_stream_shape() + if response_stream_shape is None: raise SagemakerError( status_code=500, message=( @@ -216,9 +226,7 @@ class AWSEventStreamDecoder: ), ) response_dict = event.to_response_dict() - parsed_response = self.parser.parse( - response_dict, SAGEMAKER_RESPONSE_STREAM_SHAPE - ) + parsed_response = self.parser.parse(response_dict, response_stream_shape) if response_dict["status_code"] != 200: raise ValueError(f"Bad response code, expected 200: {response_dict}") diff --git a/litellm/llms/sagemaker/completion/handler.py b/litellm/llms/sagemaker/completion/handler.py index efbb218f575..de7be18e8ba 100644 --- a/litellm/llms/sagemaker/completion/handler.py +++ b/litellm/llms/sagemaker/completion/handler.py @@ -578,7 +578,7 @@ class SagemakerLLM(BaseAWSLLM): logger_fn=None, ): """ - Supports both Huggingface Jumpstart embeddings and Voyage models + Supports Hugging Face (TGI), Voyage, and Cohere embedding endpoints """ ### BOTO3 INIT import boto3 diff --git a/litellm/llms/sagemaker/completion/transformation.py b/litellm/llms/sagemaker/completion/transformation.py index 3e4e2460cdb..8fd32bc4460 100644 --- a/litellm/llms/sagemaker/completion/transformation.py +++ b/litellm/llms/sagemaker/completion/transformation.py @@ -1,7 +1,7 @@ """ Translate from OpenAI's `/v1/chat/completions` to Sagemaker's `/invoke` -In the Huggingface TGI format. +In the Huggingface TGI format. """ import json diff --git a/litellm/llms/sagemaker/embedding/cohere_transformation.py b/litellm/llms/sagemaker/embedding/cohere_transformation.py new file mode 100644 index 00000000000..fdb67202ebb --- /dev/null +++ b/litellm/llms/sagemaker/embedding/cohere_transformation.py @@ -0,0 +1,141 @@ +""" +Translate from OpenAI's `/v1/embeddings` to Sagemaker's `/invoke` + +In the native Cohere embed format for self-hosted Cohere endpoints +(AWS Marketplace / JumpStart). Cohere containers expect +`{"texts": [...], "input_type": "..."}` and reject the HuggingFace TGI shape +`{"inputs": [...]}` with `422 EmbedReqV2.inputs is of type string but should +be of type Object`. + +Reference: https://docs.cohere.com/v2/reference/embed +""" + +from typing import TYPE_CHECKING, Any, List, Optional, Union, cast + +if TYPE_CHECKING: + from litellm.types.llms.openai import AllEmbeddingInputValues + +from httpx._models import Headers, Response + +import litellm +from litellm.llms.base_llm.chat.transformation import BaseLLMException +from litellm.llms.base_llm.embedding.transformation import BaseEmbeddingConfig +from litellm.llms.bedrock.embed.cohere_transformation import ( + BedrockCohereEmbeddingConfig, +) +from litellm.llms.cohere.embed.v1_transformation import CohereEmbeddingConfig +from litellm.types.utils import EmbeddingResponse + +from ..common_utils import SagemakerError + + +class SagemakerCohereEmbeddingConfig(BaseEmbeddingConfig): + """ + SageMaker invoke payload for self-hosted Cohere embed models. + """ + + def __init__(self) -> None: + pass + + def get_supported_openai_params(self, model: str) -> List[str]: + return ["encoding_format", "dimensions", "input_type"] + + def map_openai_params( + self, + non_default_params: dict, + optional_params: dict, + model: str, + drop_params: bool, + ) -> dict: + optional_params = BedrockCohereEmbeddingConfig().map_openai_params( + non_default_params=non_default_params, + optional_params=optional_params, + ) + if "input_type" in non_default_params: + optional_params["input_type"] = non_default_params["input_type"] + return optional_params + + def get_error_class( + self, error_message: str, status_code: int, headers: Union[dict, Headers] + ) -> BaseLLMException: + return SagemakerError( + message=error_message, status_code=status_code, headers=headers + ) + + def transform_embedding_request( + self, + model: str, + input: "AllEmbeddingInputValues", + optional_params: dict, + headers: dict, + ) -> dict: + """ + Transform embedding request for Cohere models on SageMaker + """ + if isinstance(input, str): + input_list: List[str] = [input] + elif isinstance(input, list): + if input and (isinstance(input[0], list) or isinstance(input[0], int)): + raise ValueError("Input must be a list of strings") + input_list = cast(List[str], input) + else: + input_list = [str(input)] + + return dict( + BedrockCohereEmbeddingConfig()._transform_request( + model=model, + input=input_list, + inference_params=optional_params, + ) + ) + + def transform_embedding_response( + self, + model: str, + raw_response: Response, + model_response: "EmbeddingResponse", + logging_obj: Any, + api_key: Optional[str] = None, + request_data: dict = {}, + optional_params: dict = {}, + litellm_params: dict = {}, + ) -> "EmbeddingResponse": + """ + Transform embedding response for Cohere models on SageMaker. + + Uses `CohereEmbeddingConfig._populate_embedding_response` (not + `_transform_response`) so we do not log `post_call` a second time + — the SageMaker embedding handler already logs `post_call` before + invoking this transform. + """ + input_value = ( + logging_obj.model_call_details.get("input") + or request_data.get("texts") + or request_data.get("images") + or [] + ) + if isinstance(input_value, str): + input_value = [input_value] + + return CohereEmbeddingConfig()._populate_embedding_response( + response_json=raw_response.json(), + model_response=model_response, + model=model, + encoding=litellm.encoding, + input=input_value, + ) + + def validate_environment( + self, + headers: dict, + model: str, + messages: List[Any], + optional_params: dict, + litellm_params: dict, + api_key: Optional[str] = None, + api_base: Optional[str] = None, + ) -> dict: + """ + Validate environment for SageMaker Cohere embeddings + """ + return {"Content-Type": "application/json"} diff --git a/litellm/llms/sagemaker/embedding/transformation.py b/litellm/llms/sagemaker/embedding/transformation.py index 04430171187..5e2aa99534f 100644 --- a/litellm/llms/sagemaker/embedding/transformation.py +++ b/litellm/llms/sagemaker/embedding/transformation.py @@ -1,7 +1,7 @@ """ Translate from OpenAI's `/v1/embeddings` to Sagemaker's `/invoke` -In the Huggingface TGI format. +In the Huggingface TGI format. """ from typing import TYPE_CHECKING, Any, List, Optional, Union @@ -11,12 +11,13 @@ if TYPE_CHECKING: from httpx._models import Headers, Response -from litellm.llms.base_llm.embedding.transformation import BaseEmbeddingConfig from litellm.llms.base_llm.chat.transformation import BaseLLMException -from litellm.types.utils import Usage, EmbeddingResponse +from litellm.llms.base_llm.embedding.transformation import BaseEmbeddingConfig from litellm.llms.voyage.embedding.transformation import VoyageEmbeddingConfig +from litellm.types.utils import EmbeddingResponse, Usage from ..common_utils import SagemakerError +from .cohere_transformation import SagemakerCohereEmbeddingConfig class SagemakerEmbeddingConfig(BaseEmbeddingConfig): @@ -38,17 +39,20 @@ class SagemakerEmbeddingConfig(BaseEmbeddingConfig): Returns: Appropriate embedding config instance """ - if "voyage" in model.lower(): + model_lower = model.lower() + if "voyage" in model_lower: return VoyageEmbeddingConfig() - else: - return cls() + if "cohere" in model_lower: + return SagemakerCohereEmbeddingConfig() + return cls() def get_supported_openai_params(self, model: str) -> List[str]: - # Check if this is an embedding model - if "voyage" in model.lower(): + model_lower = model.lower() + if "voyage" in model_lower: return VoyageEmbeddingConfig().get_supported_openai_params(model) - else: - return [] + if "cohere" in model_lower: + return SagemakerCohereEmbeddingConfig().get_supported_openai_params(model) + return [] def map_openai_params( self, diff --git a/litellm/llms/sap/credentials.py b/litellm/llms/sap/credentials.py index 0ae351783e8..dd307ddf496 100644 --- a/litellm/llms/sap/credentials.py +++ b/litellm/llms/sap/credentials.py @@ -207,7 +207,7 @@ def resolve_resource_group(sources: List[Source]) -> Optional[str]: def _parse_service_key_once( - service_key: Optional[Union[str, dict]] + service_key: Optional[Union[str, dict]], ) -> Optional[Dict[str, Any]]: """ Pre-parse service_key if it's a string to avoid repeated JSON parsing. diff --git a/litellm/llms/snowflake/chat/transformation.py b/litellm/llms/snowflake/chat/transformation.py index 3e590680a75..23bb6f44757 100644 --- a/litellm/llms/snowflake/chat/transformation.py +++ b/litellm/llms/snowflake/chat/transformation.py @@ -14,7 +14,6 @@ from ...openai_like.chat.transformation import OpenAIGPTConfig from ..utils import SnowflakeBaseConfig - if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj diff --git a/litellm/llms/together_ai/chat.py b/litellm/llms/together_ai/chat.py index 7efb12fc1b2..238849cc1ec 100644 --- a/litellm/llms/together_ai/chat.py +++ b/litellm/llms/together_ai/chat.py @@ -1,5 +1,5 @@ """ -Support for OpenAI's `/v1/chat/completions` endpoint. +Support for OpenAI's `/v1/chat/completions` endpoint. Calls done in OpenAI/openai.py as TogetherAI is openai-compatible. diff --git a/litellm/llms/together_ai/embed.py b/litellm/llms/together_ai/embed.py index 577df0256cc..6a39b94acfc 100644 --- a/litellm/llms/together_ai/embed.py +++ b/litellm/llms/together_ai/embed.py @@ -1,5 +1,5 @@ """ -Support for OpenAI's `/v1/embeddings` endpoint. +Support for OpenAI's `/v1/embeddings` endpoint. Calls done in OpenAI/openai.py as TogetherAI is openai-compatible. diff --git a/litellm/llms/together_ai/rerank/transformation.py b/litellm/llms/together_ai/rerank/transformation.py index 63b593dfe42..f4d642bd25a 100644 --- a/litellm/llms/together_ai/rerank/transformation.py +++ b/litellm/llms/together_ai/rerank/transformation.py @@ -1,5 +1,5 @@ """ -Transformation logic from Cohere's /v1/rerank format to Together AI's `/v1/rerank` format. +Transformation logic from Cohere's /v1/rerank format to Together AI's `/v1/rerank` format. Why separate file? Make it easy to see how transformation works """ diff --git a/litellm/llms/vertex_ai/context_caching/transformation.py b/litellm/llms/vertex_ai/context_caching/transformation.py index 950edbeb478..f73eb220cc6 100644 --- a/litellm/llms/vertex_ai/context_caching/transformation.py +++ b/litellm/llms/vertex_ai/context_caching/transformation.py @@ -1,5 +1,5 @@ """ -Transformation logic for context caching. +Transformation logic for context caching. Why separate file? Make it easy to see how transformation works """ @@ -19,7 +19,7 @@ from ..gemini.transformation import ( def get_first_continuous_block_idx( - filtered_messages: List[Tuple[int, AllMessageValues]] # (idx, message) + filtered_messages: List[Tuple[int, AllMessageValues]], # (idx, message) ) -> int: """ Find the array index that ends the first continuous sequence of message blocks. @@ -174,7 +174,9 @@ def transform_openai_messages_to_gemini_context_caching( ) transformed_messages = _gemini_convert_messages_with_history( - messages=new_messages, model=model + messages=new_messages, + model=model, + custom_llm_provider=custom_llm_provider, ) model_name = "models/{}".format(model) diff --git a/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py b/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py index ac0f07b8e0b..3f945adca0d 100644 --- a/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py +++ b/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py @@ -41,7 +41,7 @@ class ContextCachingEndpoints(VertexBase): """ def __init__(self) -> None: - pass + super().__init__() def _get_token_and_url_context_caching( self, diff --git a/litellm/llms/vertex_ai/gemini/transformation.py b/litellm/llms/vertex_ai/gemini/transformation.py index 9afa5dec465..4f5846cc5b6 100644 --- a/litellm/llms/vertex_ai/gemini/transformation.py +++ b/litellm/llms/vertex_ai/gemini/transformation.py @@ -6,13 +6,16 @@ Why separate file? Make it easy to see how transformation works import json import os -from typing import TYPE_CHECKING, Dict, List, Literal, Optional, Tuple, Union, cast +import re +from typing import TYPE_CHECKING, Any, Dict, List, Literal, Optional, Tuple, Union, cast +from urllib.parse import quote import httpx from pydantic import BaseModel import litellm from litellm._logging import verbose_logger +from litellm.litellm_core_utils.asyncify import asyncify from litellm.litellm_core_utils.prompt_templates.common_utils import ( _get_image_mime_type_from_url, ) @@ -57,6 +60,45 @@ from ..common_utils import ( get_supports_system_message, ) +# Typed as Any to avoid introducing a module-load-time cyclic import to +# vertex_llm_base. The instance is lazily constructed by _get_vertex_base() +# the first time GCS metadata needs to be fetched. +_GCS_METADATA_VERTEX_BASE: Optional[Any] = None +# Shared sync client for GCS JSON API metadata reads so proxy/SSL settings +# from litellm's HTTP stack apply (see Greptile review on PR #27278). +_GCS_METADATA_HTTP_HANDLER: Optional[HTTPHandler] = None +_GEMINI_MIME_TYPE_ALIASES: Dict[str, str] = { + "image/jpg": "image/jpeg", +} + + +def _apply_gemini_mime_type_aliases(mime_type: str) -> str: + """Normalize known MIME aliases only; does not consult the file-type registry. + + Also strips MIME parameters (e.g. ``; charset=utf-8``) so that values + sourced from GCS object metadata (``contentType``) validate correctly. + """ + normalized = mime_type.split(";", 1)[0].strip().lower() + return _GEMINI_MIME_TYPE_ALIASES.get(normalized, normalized) + + +def _get_vertex_base() -> Any: + """Lazily return the shared VertexBase instance to avoid a module-load-time cyclic import.""" + global _GCS_METADATA_VERTEX_BASE + if _GCS_METADATA_VERTEX_BASE is None: + from ..vertex_llm_base import VertexBase + + _GCS_METADATA_VERTEX_BASE = VertexBase() + return _GCS_METADATA_VERTEX_BASE + + +def _get_gcs_metadata_http_handler() -> HTTPHandler: + global _GCS_METADATA_HTTP_HANDLER + if _GCS_METADATA_HTTP_HANDLER is None: + _GCS_METADATA_HTTP_HANDLER = HTTPHandler(timeout=5.0) + return _GCS_METADATA_HTTP_HANDLER + + if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj @@ -171,12 +213,299 @@ def _apply_gemini_metadata( return cast(PartType, part_dict) +def _parse_gs_uri(gs_uri: str) -> Tuple[str, str]: + if not gs_uri.startswith("gs://"): + raise ValueError(f"Invalid gs URI: {gs_uri}") + uri_without_scheme = gs_uri[5:] # drop gs:// + uri_parts = uri_without_scheme.split("/", 1) + if len(uri_parts) != 2 or not uri_parts[0] or not uri_parts[1]: + raise ValueError(f"Invalid gs URI: {gs_uri}") + return uri_parts[0], uri_parts[1] + + +def _is_valid_gcs_bucket_name(bucket: str) -> bool: + """ + Validate bucket name against core GCS naming constraints. + """ + bucket_length = len(bucket) + max_bucket_length = 222 if "." in bucket else 63 + if bucket_length < 3 or bucket_length > max_bucket_length: + return False + if "." in bucket and any( + len(label) == 0 or len(label) > 63 for label in bucket.split(".") + ): + return False + if not re.fullmatch(r"[a-z0-9][a-z0-9._-]*[a-z0-9]", bucket): + return False + if ".." in bucket: + return False + if re.fullmatch(r"\d+\.\d+\.\d+\.\d+", bucket): + return False + return True + + +def _gs_uri_requires_content_type_metadata(url: str) -> bool: + """ + True when _process_gemini_media would call _get_gcs_object_content_type + (extension-less gs:// and no explicit format passed into that helper). + """ + if "gs://" not in url: + return False + extension_with_dot = os.path.splitext(url)[-1] + extension = extension_with_dot[1:] if extension_with_dot else "" + return len(extension) == 0 + + +def _image_url_payload_may_need_sync_gcs_metadata_fetch( + raw_image_url: Any, +) -> bool: + """ + True when this image_url value (content-part image_url or assistant ``images[]`` + entry) can trigger a blocking GCS metadata read for MIME resolution. + """ + fmt: Optional[str] = None + url: Optional[str] = None + if isinstance(raw_image_url, dict): + url = raw_image_url.get("url") # type: ignore[assignment] + if not isinstance(url, str): + return False + fmt = ( + raw_image_url.get("format") + or raw_image_url.get("mime_type") + or raw_image_url.get("content_type") + ) + elif isinstance(raw_image_url, str): + url = raw_image_url + else: + return False + if "gs://" not in url or fmt: + return False + return _gs_uri_requires_content_type_metadata(url) + + +def _openai_messages_may_need_sync_gcs_metadata_fetch( + messages: List[AllMessageValues], +) -> bool: + """ + Heuristic: True if any message part can trigger a blocking GCS JSON + metadata read inside _transform_request_body (extension-less gs:// without + explicit MIME hints). Covers user/system ``content`` parts and assistant + ``images`` (same paths as ``_gemini_convert_messages_with_history``). Used + to decide whether ``async_transform_request_body`` should offload the sync + transform via ``asyncify``. + """ + for raw in messages: + msg: Any = raw + if not isinstance(msg, dict) and hasattr(msg, "model_dump"): + msg = msg.model_dump(exclude_none=False) + if not isinstance(msg, dict): + continue + images_field = msg.get("images") + if isinstance(images_field, list): + for image_item in images_field: + if not isinstance(image_item, dict): + continue + if _image_url_payload_may_need_sync_gcs_metadata_fetch( + image_item.get("image_url") + ): + return True + + content = msg.get("content") + if not isinstance(content, list): + continue + for item in content: + if not isinstance(item, dict): + continue + itype = item.get("type") + if itype == "image_url": + if _image_url_payload_may_need_sync_gcs_metadata_fetch( + item.get("image_url") + ): + return True + elif itype == "file": + file_obj = item.get("file") + if not isinstance(file_obj, dict): + continue + fmt = ( + file_obj.get("format") + or file_obj.get("mime_type") + or file_obj.get("content_type") + ) + passed = file_obj.get("file_id") or file_obj.get("file_data") + if ( + isinstance(passed, str) + and "gs://" in passed + and not fmt + and _gs_uri_requires_content_type_metadata(passed) + ): + return True + return False + + +def _get_gcs_object_content_type( + image_url: str, + vertex_project: Optional[str] = None, + vertex_credentials: Optional[Any] = None, +) -> Optional[str]: + """ + Resolve content type from GCS object metadata. + + Only attaches a Bearer token when the caller explicitly supplies Vertex + credentials, to avoid using the server's default Google credentials on + the Gemini API-key (Google AI Studio) path and being used as an oracle + for private GCS object metadata. Without explicit credentials we only + issue an anonymous request, which only succeeds for publicly-readable + objects. + """ + try: + bucket, object_name = _parse_gs_uri(image_url) + except ValueError: + return None + if not _is_valid_gcs_bucket_name(bucket): + return None + + headers: Dict[str, str] = {} + explicit_vertex_auth_provided = ( + vertex_project is not None or vertex_credentials is not None + ) + if explicit_vertex_auth_provided: + try: + access_token, _ = _get_vertex_base().get_access_token( + credentials=vertex_credentials, + project_id=vertex_project, + ) + headers["Authorization"] = f"Bearer {access_token}" + except Exception as e: + raise litellm.BadRequestError( + message=( + "Unable to fetch GCS metadata with provided Vertex credentials/project. " + f"Original error: {str(e)}" + ), + model=None, + llm_provider="vertex_ai", + ) + + # Build the URL via httpx.URL with a fixed scheme/host and URL-encode both + # bucket and object so CodeQL does not flag the interpolation as a + # potential SSRF that could resolve to an arbitrary host. + encoded_bucket = quote(bucket, safe="") + encoded_object = quote(object_name, safe="") + metadata_url = httpx.URL( + scheme="https", + host="storage.googleapis.com", + path=f"/storage/v1/b/{encoded_bucket}/o/{encoded_object}", + params={"fields": "contentType"}, + ) + try: + response = _get_gcs_metadata_http_handler().get( + url=str(metadata_url), + headers=headers or None, + ) + except httpx.RequestError as e: + if explicit_vertex_auth_provided: + raise litellm.BadRequestError( + message=( + "Unable to reach GCS JSON API for object metadata with provided " + f"Vertex credentials. {type(e).__name__}: {e}" + ), + model=None, + llm_provider="vertex_ai", + ) from e + return None + + if response.is_error: + if explicit_vertex_auth_provided: + preview = (response.text or "")[:1024] + raise litellm.BadRequestError( + message=( + "Unable to read GCS object metadata with provided Vertex credentials. " + f"HTTP {response.status_code}. Response body (truncated): {preview!r}" + ), + model=None, + llm_provider="vertex_ai", + ) + return None + + try: + payload = response.json() + except ValueError as e: + if explicit_vertex_auth_provided: + raise litellm.BadRequestError( + message=( + "GCS metadata response was not valid JSON when using provided " + f"Vertex credentials (HTTP {response.status_code}). Error: {e}" + ), + model=None, + llm_provider="vertex_ai", + ) from e + return None + + if not isinstance(payload, dict): + if explicit_vertex_auth_provided: + raise litellm.BadRequestError( + message=( + "GCS metadata response was not a JSON object when using provided " + f"Vertex credentials (HTTP {response.status_code})." + ), + model=None, + llm_provider="vertex_ai", + ) + return None + + content_type = payload.get("contentType") + if isinstance(content_type, str) and len(content_type) > 0: + return content_type + + if explicit_vertex_auth_provided: + preview = (response.text or "")[:1024] + raise litellm.BadRequestError( + message=( + "GCS metadata JSON did not include a non-empty contentType field when " + f"using provided Vertex credentials (HTTP {response.status_code}). " + f"Body (truncated): {preview!r}" + ), + model=None, + llm_provider="vertex_ai", + ) + return None + + +def _normalize_and_validate_gemini_mime_type( + mime_type: str, model: Optional[str] +) -> str: + # Import lazily to avoid a module-level cyclic-import alert with + # litellm.types.files. + from litellm.types.files import get_file_extension_from_mime_type + + normalized_mime_type = _apply_gemini_mime_type_aliases(mime_type) + try: + file_extension = get_file_extension_from_mime_type(normalized_mime_type) + file_type = get_file_type_from_extension(file_extension) + except ValueError: + raise litellm.BadRequestError( + message=f"File type not supported by gemini - {normalized_mime_type}", + model=model, + llm_provider="vertex_ai", + ) + + if not is_gemini_1_5_accepted_file_type(file_type): + raise litellm.BadRequestError( + message=f"File type not supported by gemini - {file_type}", + model=model, + llm_provider="vertex_ai", + ) + + return get_file_mime_type_for_file_type(file_type) + + def _process_gemini_media( image_url: str, format: Optional[str] = None, media_resolution_enum: Optional[Dict[str, str]] = None, model: Optional[str] = None, video_metadata: Optional[Dict[str, Any]] = None, + vertex_project: Optional[str] = None, + vertex_credentials: Optional[Any] = None, ) -> PartType: """ Given a media URL (image, audio, or video), return the appropriate PartType for Gemini @@ -193,20 +522,63 @@ def _process_gemini_media( try: # GCS URIs if "gs://" in image_url: - # Figure out file type extension_with_dot = os.path.splitext(image_url)[-1] # Ex: ".png" extension = extension_with_dot[1:] # Ex: "png" + explicit_gcs_format = False if not format: - file_type = get_file_type_from_extension(extension) + mime_type: Optional[str] = None + # For extension-less gs:// URIs, we cannot infer from path. + # If callers pass `format`/`mime_type`, this branch is skipped. + if extension: + file_type = get_file_type_from_extension(extension) - # Validate the file type is supported by Gemini - if not is_gemini_1_5_accepted_file_type(file_type): - raise Exception(f"File type not supported by gemini - {file_type}") + # Validate the file type is supported by Gemini + if not is_gemini_1_5_accepted_file_type(file_type): + raise litellm.BadRequestError( + message=f"File type not supported by gemini - {file_type}", + model=model, + llm_provider="vertex_ai", + ) - mime_type = get_file_mime_type_for_file_type(file_type) + mime_type = get_file_mime_type_for_file_type(file_type) + else: + mime_type = _get_gcs_object_content_type( + image_url=image_url, + vertex_project=vertex_project, + vertex_credentials=vertex_credentials, + ) + if mime_type is None: + raise litellm.BadRequestError( + message=( + f"Unable to determine mime type for gs URI: {image_url}. " + "This gs:// URI has no file extension and GCS metadata " + "lookup failed. Set it explicitly using image_url.format " + "(or image_url.mime_type/content_type) or " + "message.content[].file.format." + ), + model=model, + llm_provider="vertex_ai", + ) else: mime_type = format + explicit_gcs_format = True + if mime_type is None: + raise litellm.BadRequestError( + message=f"File type not supported by gemini - {image_url}", + model=model, + llm_provider="vertex_ai", + ) + if explicit_gcs_format: + # Callers who pass format/mime_type explicitly for gs:// URIs + # rely on pass-through to Gemini (pre-PR behavior). Only apply + # known MIME aliases; skip litellm's file-type registry. + mime_type = _apply_gemini_mime_type_aliases(mime_type) + else: + mime_type = _normalize_and_validate_gemini_mime_type( + mime_type=mime_type, + model=model, + ) file_data = FileDataType(mime_type=mime_type, file_uri=image_url) part: PartType = {"file_data": file_data} return _apply_gemini_metadata( @@ -258,8 +630,6 @@ def _snake_to_camel(snake_str: str) -> str: def _camel_to_snake(camel_str: str) -> str: """Convert camelCase to snake_case""" - import re - return re.sub(r"(? List[ContentType]: """ Converts given messages from OpenAI format to Gemini format @@ -326,6 +698,16 @@ def _gemini_convert_messages_with_history( # noqa: PLR0915 msg_i = 0 tool_call_responses = [] + vertex_project = None + vertex_credentials = None + if litellm_params: + vertex_project = litellm_params.get("vertex_project") or litellm_params.get( + "vertex_ai_project" + ) + vertex_credentials = litellm_params.get( + "vertex_credentials" + ) or litellm_params.get("vertex_ai_credentials") + try: while msg_i < len(messages): user_content: List[PartType] = [] @@ -351,20 +733,42 @@ def _gemini_convert_messages_with_history( # noqa: PLR0915 img_element = element format: Optional[str] = None media_resolution_enum: Optional[Dict[str, str]] = None - if isinstance(img_element["image_url"], dict): - image_url = img_element["image_url"]["url"] - format = img_element["image_url"].get("format") - detail = img_element["image_url"].get("detail") + raw_image_url = img_element.get("image_url") + if raw_image_url is None: + raise litellm.BadRequestError( + message="Invalid message content: element type is 'image_url' but 'image_url' field is missing ", + model=model, + llm_provider="vertex_ai", + ) + if isinstance(raw_image_url, dict): + image_url = raw_image_url.get("url") + if image_url is None: + raise litellm.BadRequestError( + message="Invalid message content: element type is 'image_url' but 'url' field is missing inside 'image_url' ", + model=model, + llm_provider="vertex_ai", + ) + # TypedDict does not declare mime_type/content_type; + # read via Dict[str, Any] for caller-provided MIME fields. + image_url_dict = cast(Dict[str, Any], raw_image_url) + format = ( + image_url_dict.get("format") + or image_url_dict.get("mime_type") + or image_url_dict.get("content_type") + ) + detail = image_url_dict.get("detail") media_resolution_enum = ( _convert_detail_to_media_resolution_enum(detail) ) else: - image_url = img_element["image_url"] + image_url = raw_image_url _part = _process_gemini_media( image_url=image_url, format=format, media_resolution_enum=media_resolution_enum, model=model, + vertex_project=vertex_project, + vertex_credentials=vertex_credentials, ) _parts.append(_part) elif element["type"] == "input_audio": @@ -390,15 +794,31 @@ def _gemini_convert_messages_with_history( # noqa: PLR0915 image_url=openai_image_str, format=audio_format_modified, model=model, + vertex_project=vertex_project, + vertex_credentials=vertex_credentials, ) _parts.append(_part) elif element["type"] == "file": file_element = cast(ChatCompletionFileObject, element) - file_id = file_element["file"].get("file_id") - format = file_element["file"].get("format") - file_data = file_element["file"].get("file_data") - detail = file_element["file"].get("detail") - video_metadata = file_element["file"].get("video_metadata") + _file_field = file_element.get("file") + if _file_field is None: + raise litellm.BadRequestError( + message="Content block has type='file' but is missing the required 'file' field", + model=model, + llm_provider="vertex_ai", + ) + # TypedDict does not declare mime_type/content_type; + # read via Dict[str, Any] for caller-provided MIME fields. + file_dict = cast(Dict[str, Any], _file_field) + file_id = file_dict.get("file_id") + format = ( + file_dict.get("format") + or file_dict.get("mime_type") + or file_dict.get("content_type") + ) + file_data = file_dict.get("file_data") + detail = file_dict.get("detail") + video_metadata = file_dict.get("video_metadata") passed_file = file_id or file_data if passed_file is None: raise Exception( @@ -417,13 +837,23 @@ def _gemini_convert_messages_with_history( # noqa: PLR0915 model=model, media_resolution_enum=media_resolution_enum, video_metadata=video_metadata, + vertex_project=vertex_project, + vertex_credentials=vertex_credentials, ) _parts.append(_part) - except Exception: - raise Exception( - "Unable to determine mime type for file_id: {}, set this explicitly using message[{}].content[{}].file.format".format( - file_id, msg_i, element_idx - ) + except litellm.BadRequestError: + raise + except Exception as e: + raise litellm.BadRequestError( + message=( + f"Unable to determine mime type for file: " + f"{file_id or 'provided data'}, set this explicitly " + f"using message[{msg_i}].content[{element_idx}].file.format " + f"(or file.mime_type/content_type). " + f"Original error: {str(e)}" + ), + model=model, + llm_provider="vertex_ai", ) user_content.extend(_parts) elif _message_content is not None and isinstance(_message_content, str): @@ -528,7 +958,11 @@ def _gemini_convert_messages_with_history( # noqa: PLR0915 image_url_obj = image_item.get("image_url") if isinstance(image_url_obj, dict): assistant_image_url = image_url_obj.get("url") - format = image_url_obj.get("format") + format = ( + image_url_obj.get("format") + or image_url_obj.get("mime_type") + or image_url_obj.get("content_type") + ) detail = image_url_obj.get("detail") media_resolution_enum = ( _convert_detail_to_media_resolution_enum(detail) @@ -539,6 +973,8 @@ def _gemini_convert_messages_with_history( # noqa: PLR0915 format=format, media_resolution_enum=media_resolution_enum, model=model, + vertex_project=vertex_project, + vertex_credentials=vertex_credentials, ) assistant_content.append(_part) @@ -548,7 +984,9 @@ def _gemini_convert_messages_with_history( # noqa: PLR0915 or assistant_msg.get("function_call") is not None ): # support assistant tool invoke conversion gemini_tool_call_parts = convert_to_gemini_tool_call_invoke( - assistant_msg, model=model + assistant_msg, + model=model, + custom_llm_provider=custom_llm_provider, ) ## check if gemini_tool_call already exists in assistant_content for gemini_tool_call_part in gemini_tool_call_parts: @@ -607,7 +1045,10 @@ def _gemini_convert_messages_with_history( # noqa: PLR0915 and messages[msg_i]["role"] in tool_call_message_roles ): _part = convert_to_gemini_tool_call_result( - messages[msg_i], last_message_with_tool_calls # type: ignore + messages[msg_i], # type: ignore + last_message_with_tool_calls, # type: ignore + model=model, + custom_llm_provider=custom_llm_provider, ) msg_i += 1 # Handle both single part and list of parts (for Computer Use with images) @@ -632,16 +1073,14 @@ def _gemini_convert_messages_with_history( # noqa: PLR0915 contents.append(ContentType(role="user", parts=tool_call_responses)) if len(contents) == 0: - verbose_logger.warning( - """ + verbose_logger.warning(""" No contents in messages. Contents are required. See https://cloud.google.com/vertex-ai/docs/reference/rest/v1/projects.locations.publishers.models/generateContent#request-body. If the original request did not comply to OpenAI API requirements it should have failed by now, but LiteLLM does not check for missing messages. Setting an empty content to prevent an 400 error. Relevant Issue - https://github.com/BerriAI/litellm/issues/9733 - """ - ) + """) contents.append(ContentType(role="user", parts=[PartType(text=" ")])) return contents except Exception as e: @@ -713,11 +1152,11 @@ def _transform_request_body( # noqa: PLR0915 try: if custom_llm_provider == "gemini": content = litellm.GoogleAIStudioGeminiConfig()._transform_messages( - messages=messages, model=model + messages=messages, model=model, litellm_params=litellm_params ) else: content = litellm.VertexGeminiConfig()._transform_messages( - messages=messages, model=model + messages=messages, model=model, litellm_params=litellm_params ) tools: Optional[Tools] = optional_params.pop("tools", None) tool_choice: Optional[ToolConfig] = optional_params.pop("tool_choice", None) @@ -893,6 +1332,20 @@ async def async_transform_request_body( vertex_auth_header=vertex_auth_header, ) + if _openai_messages_may_need_sync_gcs_metadata_fetch(messages): + # _transform_request_body may issue a sync httpx.get (up to 5s timeout) + # via _get_gcs_object_content_type to fetch GCS object metadata. Run the + # whole sync transformation on a worker thread so it does not block the + # async event loop. + return await asyncify(_transform_request_body)( + messages=messages, + model=model, + custom_llm_provider=custom_llm_provider, + litellm_params=litellm_params, + cached_content=cached_content, + optional_params=optional_params, + ) + return _transform_request_body( messages=messages, model=model, diff --git a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py index 6278de662f8..189ac7a7f6a 100644 --- a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py +++ b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py @@ -280,6 +280,8 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): - gemini-3-pro-preview - gemini-3-flash - gemini-3-flash-preview (Gemini 3 Flash) + - gemini-3.1-pro-preview, gemini-3.1-flash, gemini-3.1-flash-lite-preview + - gemini-3.5-flash - Any future Gemini 3.x models """ # Check for Gemini 3 models @@ -287,6 +289,20 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): return True return False + @staticmethod + def _forward_gemini_function_call_id( + model: str, custom_llm_provider: Optional[str] = None + ) -> bool: + """ + Whether to include `id` on function_call / function_response parts. + + Gemini 3+ on Google AI Studio accepts (and returns) `id` for strict + tool-call matching. Vertex AI rejects the field with HTTP 400. + """ + if custom_llm_provider != "gemini": + return False + return VertexGeminiConfig._is_gemini_3_or_newer(model) + def _supports_penalty_parameters(self, model: str) -> bool: # Gemini 3 models do not support penalty parameters if VertexGeminiConfig._is_gemini_3_or_newer(model): @@ -300,6 +316,7 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): supported_params = [ "temperature", "top_p", + "top_k", "max_tokens", "max_completion_tokens", "stream", @@ -363,6 +380,66 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): """ return Tools(googleSearch={}) + @staticmethod + def _search_tool_keys() -> set: + return { + VertexToolName.GOOGLE_SEARCH.value, + VertexToolName.GOOGLE_SEARCH_RETRIEVAL.value, + VertexToolName.ENTERPRISE_WEB_SEARCH.value, + VertexToolName.URL_CONTEXT.value, + "google_search", + "google_search_retrieval", + "enterprise_web_search", + "urlContext", + } + + @classmethod + def _drop_search_tools_mixed_with_functions(cls, optional_params: dict) -> None: + """ + Drop search tools from optional_params when mixed with function declarations + and include_server_side_tool_invocations is not enabled. + + Runs after map_openai_params merges tools and web_search_options so both + code paths (single _map_function call vs split tools + web_search_options) + get the same conflict resolution. + """ + if optional_params.get("include_server_side_tool_invocations"): + return + + tools = optional_params.get("tools") + if not isinstance(tools, list) or not tools: + return + + search_tool_keys = cls._search_tool_keys() + has_function_declarations = any( + isinstance(tool, dict) and tool.get("function_declarations") + for tool in tools + ) + if not has_function_declarations: + return + + has_search_tools = any( + isinstance(tool, dict) and any(key in tool for key in search_tool_keys) + for tool in tools + ) + if not has_search_tools: + return + + verbose_logger.warning( + "Vertex AI does not support mixing function declarations with " + "search tools (googleSearch, enterpriseWebSearch, urlContext, " + "googleSearchRetrieval) in the same request. Dropping search " + "tools and keeping function declarations. To use search tools, " + "send a request without function calling tools." + ) + optional_params["tools"] = [ + tool + for tool in tools + if not ( + isinstance(tool, dict) and any(key in tool for key in search_tool_keys) + ) + ] + def _map_service_tier_param(self, value: str, optional_params: dict) -> None: """ Map OpenAI service_tier (string) to Gemini serviceTier. @@ -884,9 +961,10 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): GeminiThinkingConfig with thinkingLevel and includeThoughts """ # Check if this is gemini-3-flash which supports MINIMAL thinking level - # Covers gemini-3-flash, gemini-3-flash-preview, gemini-3.1-flash, gemini-3.1-flash-lite-preview, etc. + # Covers gemini-3-flash, gemini-3-flash-preview, gemini-3.1-flash, gemini-3.1-flash-lite-preview, + # gemini-3.5-flash, and any future 3.x-flash variants. is_gemini3flash = model and ( - "gemini-3-flash" in model.lower() or "gemini-3.1-flash" in model.lower() + "flash" in model.lower() and "gemini-3" in model.lower() ) is_gemini31pro = model and ("gemini-3.1-pro-preview" in model.lower()) if reasoning_effort == "minimal": @@ -982,8 +1060,7 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): # Follow provider defaults unless explicitly opted into legacy behavior. if litellm.enable_gemini_default_thinking_level_low is True: is_gemini3flash = ( - "gemini-3-flash-preview" in model.lower() - or "gemini-3-flash" in model.lower() + "gemini-3" in model.lower() and "flash" in model.lower() ) params["thinkingLevel"] = ( "minimal" if is_gemini3flash else "low" @@ -1077,6 +1154,7 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): model: str, drop_params: bool, ) -> Dict: + gemini_sampling_params_warned: bool = False for param, value in non_default_params.items(): if param == "temperature": if VertexGeminiConfig._is_gemini_3_or_newer(model): @@ -1086,9 +1164,41 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): "can cause infinite loops, degraded reasoning performance, and failure on complex tasks. " "Strongly recommended to use temperature = 1.0 (default)." ) + if not gemini_sampling_params_warned: + verbose_logger.warning( + "DeprecationWarning: `temperature`, `top_p`, and `top_k` continue to " + f"function for Gemini 3+ ({model}) but are planned for removal in a " + "future release. Move sampling guidance into the `system` " + "instructions instead." + ) + gemini_sampling_params_warned = True optional_params["temperature"] = value elif param == "top_p": + if ( + VertexGeminiConfig._is_gemini_3_or_newer(model) + and not gemini_sampling_params_warned + ): + verbose_logger.warning( + "DeprecationWarning: `temperature`, `top_p`, and `top_k` continue to " + f"function for Gemini 3+ ({model}) but are planned for removal in a " + "future release. Move sampling guidance into the `system` " + "instructions instead." + ) + gemini_sampling_params_warned = True optional_params["top_p"] = value + elif param == "top_k": + if ( + VertexGeminiConfig._is_gemini_3_or_newer(model) + and not gemini_sampling_params_warned + ): + verbose_logger.warning( + "DeprecationWarning: `temperature`, `top_p`, and `top_k` continue to " + f"function for Gemini 3+ ({model}) but are planned for removal in a " + "future release. Move sampling guidance into the `system` " + "instructions instead." + ) + gemini_sampling_params_warned = True + optional_params["top_k"] = value elif ( param == "stream" and value is True ): # sending stream = False, can cause it to get passed unchecked and raise issues @@ -1139,11 +1249,14 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): if _tool_choice_value is not None: optional_params["tool_choice"] = _tool_choice_value elif param == "parallel_tool_calls": - if value is False and not ( - drop_params or litellm.drop_params - ): # if drop params is True, then we should just ignore this - self.validate_parallel_tool_calls(value, non_default_params) - else: + tools_list = non_default_params.get( + "tools", non_default_params.get("functions") + ) + num_tools = len(tools_list) if isinstance(tools_list, list) else 0 + # Gemini does not support parallel_tool_calls=False with multiple + # tools. Drop the param instead of failing — Responses API clients + # often send parallel_tool_calls=false by default. + if not (value is False and num_tools > 1): optional_params["parallel_tool_calls"] = value elif param == "seed": optional_params["seed"] = value @@ -1216,6 +1329,8 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): if "temperature" not in optional_params: optional_params["temperature"] = 1.0 + self._drop_search_tools_mixed_with_functions(optional_params) + return optional_params def get_mapped_special_auth_params(self) -> dict: @@ -1588,6 +1703,10 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): } # Extract thought signature if present thought_signature = part.get("thoughtSignature") + # Gemini 3.5+ returns a stable `id` per function call to enable + # strict response matching. Preserve it as the OpenAI + # tool_call_id so it can be echoed back unchanged. + gemini_call_id = part["functionCall"].get("id") if is_function_call is True: function_dict: Dict[str, Any] = dict(_function_chunk) @@ -1605,6 +1724,11 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): "function": _function_chunk, "index": cumulative_tool_call_idx, } + # Gemini 3.5+ returns a stable native `id`; prefer it over + # the synthetic call_ so the same value can be echoed + # back on the matching `functionResponse`. + if gemini_call_id: + _tool_response_chunk["id"] = gemini_call_id # Embed thought signature in ID for OpenAI client compatibility if thought_signature: _tool_response_chunk["provider_specific_fields"] = { # type: ignore @@ -2533,9 +2657,17 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig): return model_response def _transform_messages( - self, messages: List[AllMessageValues], model: Optional[str] = None + self, + messages: List[AllMessageValues], + model: Optional[str] = None, + litellm_params: Optional[dict] = None, ) -> List[ContentType]: - return _gemini_convert_messages_with_history(messages=messages, model=model) + return _gemini_convert_messages_with_history( + messages=messages, + model=model, + litellm_params=litellm_params, + custom_llm_provider="vertex_ai", + ) def get_error_class( self, error_message: str, status_code: int, headers: Union[Dict, httpx.Headers] @@ -3139,6 +3271,31 @@ class ModelResponseIterator: self.cumulative_tool_call_index: int = 0 self.has_seen_tool_calls: bool = False + @staticmethod + def _check_streaming_error(chunk: dict) -> None: + """Detect embedded errors (e.g. 429 RESOURCE_EXHAUSTED) in streaming chunks and raise VertexAIError.""" + if "error" not in chunk: + return + error_data = chunk["error"] + if not isinstance(error_data, dict): + raise VertexAIError( + status_code=500, + message=f"Unexpected error format in mid-stream chunk: {error_data}", + ) + raw_code = error_data.get("code", 500) + if raw_code is None: + raw_code = 500 + try: + error_code = int(raw_code) + except (TypeError, ValueError): + error_code = 500 + error_message = error_data.get("message", "Unknown error") + error_status = error_data.get("status", "UNKNOWN") + raise VertexAIError( + status_code=error_code, + message=f"{error_status} - {error_message}", + ) + def _apply_stream_candidates( self, _candidates: List[Candidates], @@ -3256,6 +3413,11 @@ class ModelResponseIterator: def chunk_parser(self, chunk: dict) -> Optional["ModelResponseStream"]: try: verbose_logger.debug(f"RAW GEMINI CHUNK: {chunk}") + + # Detect mid-stream error chunks (e.g. 429 RESOURCE_EXHAUSTED). + # Vertex AI can return errors as HTTP 200 but with an "error" field in the SSE body. + self._check_streaming_error(chunk) + from litellm.types.utils import ModelResponseStream processed_chunk = GenerateContentResponseBody(**chunk) # type: ignore diff --git a/litellm/llms/vertex_ai/gemini_embeddings/batch_embed_content_transformation.py b/litellm/llms/vertex_ai/gemini_embeddings/batch_embed_content_transformation.py index e1b365c9f42..ba6e6f0c056 100644 --- a/litellm/llms/vertex_ai/gemini_embeddings/batch_embed_content_transformation.py +++ b/litellm/llms/vertex_ai/gemini_embeddings/batch_embed_content_transformation.py @@ -1,5 +1,5 @@ """ -Transformation logic from OpenAI /v1/embeddings format to Google AI Studio /batchEmbedContents format. +Transformation logic from OpenAI /v1/embeddings format to Google AI Studio /batchEmbedContents format. Why separate file? Make it easy to see how transformation works """ diff --git a/litellm/llms/vertex_ai/google_genai/transformation.py b/litellm/llms/vertex_ai/google_genai/transformation.py index d7a4ceeb3e7..c1120d9ab8b 100644 --- a/litellm/llms/vertex_ai/google_genai/transformation.py +++ b/litellm/llms/vertex_ai/google_genai/transformation.py @@ -79,6 +79,9 @@ class VertexAIGoogleGenAIConfig(GoogleGenAIConfig): Transform the generate content request for Vertex AI. Since Vertex AI natively supports Google GenAI format, we can pass most fields directly. """ + if generate_content_config_dict: + self._normalize_response_schema(generate_content_config_dict, model) + # Build the request in Google GenAI format that Vertex AI expects result = { "model": model, diff --git a/litellm/llms/vertex_ai/realtime/transformation.py b/litellm/llms/vertex_ai/realtime/transformation.py index 2b4746b174e..ea4dbccc8c8 100644 --- a/litellm/llms/vertex_ai/realtime/transformation.py +++ b/litellm/llms/vertex_ai/realtime/transformation.py @@ -14,6 +14,7 @@ Auth: OAuth2 Bearer token (not an API key). import json from typing import List, Optional +from litellm import verbose_logger from litellm.llms.gemini.realtime.transformation import GeminiRealtimeConfig @@ -26,6 +27,7 @@ class VertexAIRealtimeConfig(GeminiRealtimeConfig): """ def __init__(self, access_token: str, project: str, location: str) -> None: + super().__init__() self._access_token = access_token self._project = project self._location = location @@ -138,6 +140,62 @@ class VertexAIRealtimeConfig(GeminiRealtimeConfig): # Request translation # ------------------------------------------------------------------ + def _vertex_model_path(self, model: str) -> str: + """Return the fully-qualified Vertex AI model resource path.""" + return ( + f"projects/{self._project}" + f"/locations/{self._location}" + f"/publishers/google/models/{model}" + ) + + def _build_vertex_ai_setup_config(self, model: str, session_params: dict) -> dict: + """Build Vertex AI setup configuration with proper model path and defaults.""" + # Normalize GA-remapped fields (``output_modalities``, nested + # ``audio.input.transcription``, ``audio.input.turn_detection``) back to + # their flat beta keys so ``map_openai_params`` picks them up. Without + # this, GA clients' explicit modality / transcription / turn-detection + # settings would be silently dropped because ``map_openai_params`` only + # recognises the flat OpenAI-beta key names. + session_params = self._normalize_session_payload_for_mapping(session_params) + setup_config = self.map_openai_params( + optional_params={}, non_default_params=session_params + ) + + # Use full Vertex AI model path + setup_config["model"] = self._vertex_model_path(model) + + # Add Vertex AI specific defaults if not provided + generation_config = setup_config.setdefault("generationConfig", {}) + generation_config.setdefault("responseModalities", ["AUDIO"]) + + # Ensure Vertex defaults for realtimeInputConfig apply even when + # the client provided a partial ``turn_detection`` (e.g. only + # ``silence_duration_ms``). ``map_automatic_turn_detection`` sets + # ``disabled=True`` whenever ``create_response`` is absent or + # ``False``. Force ``disabled=False`` only when the client did + # not explicitly request ``create_response: False`` — that path + # is how transcription guardrails suppress automatic responses, + # and overriding it here would silently bypass the guardrail. + # Vertex Live has no "VAD on, no auto-response" mode, so callers + # that need that behaviour must accept that VAD is off. + client_turn_detection = session_params.get("turn_detection") + client_disabled_auto_response = ( + isinstance(client_turn_detection, dict) + and client_turn_detection.get("create_response") is False + ) + realtime_input_config = setup_config.setdefault("realtimeInputConfig", {}) + automatic_detection = realtime_input_config.setdefault( + "automaticActivityDetection", {} + ) + if not client_disabled_auto_response: + automatic_detection["disabled"] = False + automatic_detection.setdefault("silenceDurationMs", 800) + + setup_config.setdefault("inputAudioTranscription", {}) + setup_config.setdefault("outputAudioTranscription", {}) + + return setup_config + def transform_realtime_request( self, message: str, @@ -147,16 +205,50 @@ class VertexAIRealtimeConfig(GeminiRealtimeConfig): """ Translate OpenAI realtime client messages to Vertex AI format. - ``session.update`` is intentionally ignored (returns []) because - Vertex AI only accepts a single ``setup`` message at the start of - the connection — sending a second one causes a 1007 close error. - The initial setup (sent automatically before bidirectional_forward) - already includes AUDIO modality and server VAD, so there is nothing - more to configure. + On the first ``session.update`` (when no setup has been sent yet) the + full ``BidiGenerateContentSetup`` is built with Vertex AI's model path + and forwarded. Any later ``session.update`` is dropped: Vertex AI + documents ``setup`` as the first-and-only client message, and a second + ``setup`` closes the connection with a 1007 policy error. """ json_message = json.loads(message) - if json_message.get("type") == "session.update": - # Do not forward as a second setup — Vertex AI rejects it. + msg_type = json_message.get("type") + + if msg_type == "session.update": + if session_configuration_request is None: + setup_config = self._build_vertex_ai_setup_config( + model, json_message.get("session") or {} + ) + gemini_setup_msg = json.dumps({"setup": setup_config}) + + verbose_logger.debug( + "Vertex AI Realtime: Sending initial setup with tools to backend" + ) + return [gemini_setup_msg] + + # A follow-up session.update can't be forwarded as a second setup + # (Vertex Live closes the WebSocket with 1007). If this drop is + # silencing the audio-transcription guardrail's create_response + # disable, surface a warning so operators know the model will + # auto-respond before the guardrail can gate it on Vertex AI. + client_turn_detection = GeminiRealtimeConfig._extract_turn_detection( + json_message.get("session") or {} + ) + if ( + isinstance(client_turn_detection, dict) + and client_turn_detection.get("create_response") is False + ): + verbose_logger.warning( + "Vertex AI Realtime: Dropping subsequent session.update " + "(turn_detection.create_response=False) — Vertex Live " + "rejects a second setup message. Audio-transcription " + "guardrails cannot suppress the model's auto-response on " + "Vertex AI in non-deferred mode." + ) + else: + verbose_logger.debug( + "Vertex AI Realtime: Ignoring session.update (setup already sent)" + ) return [] return super().transform_realtime_request( diff --git a/litellm/llms/vertex_ai/text_to_speech/text_to_speech_handler.py b/litellm/llms/vertex_ai/text_to_speech/text_to_speech_handler.py index 9d9015c2b91..b835ad7d8fa 100644 --- a/litellm/llms/vertex_ai/text_to_speech/text_to_speech_handler.py +++ b/litellm/llms/vertex_ai/text_to_speech/text_to_speech_handler.py @@ -139,7 +139,7 @@ class VertexTextToSpeechAPI(VertexLLM): ########## End of logging ############ ####### Send the request ################### if _is_async is True: - return self.async_audio_speech( # type:ignore + return self.async_audio_speech( # type: ignore logging_obj=logging_obj, url=url, headers=headers, request=request ) sync_handler = _get_httpx_client() diff --git a/litellm/llms/vertex_ai/vertex_ai_partner_models/main.py b/litellm/llms/vertex_ai/vertex_ai_partner_models/main.py index 123d925f7c1..13aa2a5350e 100644 --- a/litellm/llms/vertex_ai/vertex_ai_partner_models/main.py +++ b/litellm/llms/vertex_ai/vertex_ai_partner_models/main.py @@ -45,7 +45,7 @@ class PartnerModelPrefixes(str, Enum): class VertexAIPartnerModels(VertexBase): def __init__(self) -> None: - pass + super().__init__() @staticmethod def is_vertex_partner_model(model: str): @@ -116,9 +116,6 @@ class VertexAIPartnerModels(VertexBase): CodestralTextCompletion, ) from litellm.llms.openai_like.chat.handler import OpenAILikeChatHandler - from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import ( - VertexLLM, - ) except Exception as e: raise VertexAIError( status_code=400, @@ -133,9 +130,7 @@ class VertexAIPartnerModels(VertexBase): message="""Upgrade vertex ai. Run `pip install "google-cloud-aiplatform>=1.38"`""", ) try: - vertex_httpx_logic = VertexLLM() - - access_token, project_id = vertex_httpx_logic._ensure_access_token( + access_token, project_id = self._ensure_access_token( credentials=vertex_credentials, project_id=vertex_project, custom_llm_provider="vertex_ai", @@ -292,22 +287,15 @@ class VertexAIPartnerModels(VertexBase): Returns: Dict containing token count information """ - try: - import vertexai - except Exception as e: - raise VertexAIError( - status_code=400, - message=f"""vertexai import failed please run `pip install -U "google-cloud-aiplatform>=1.38"`. Got error: {e}""", - ) - - if not ( - hasattr(vertexai, "preview") or hasattr(vertexai.preview, "language_models") - ): - raise VertexAIError( - status_code=400, - message="""Upgrade vertex ai. Run `pip install "google-cloud-aiplatform>=1.38"`""", - ) - + # Note: we intentionally do not import `vertexai` (the Gemini SDK shipped + # by `google-cloud-aiplatform`) on this path. Partner models such as + # Claude on Vertex use the Anthropic Messages API protocol directly via + # `:rawPredict`, and `VertexAIPartnerModelsTokenCounter` reaches that + # endpoint with an authenticated httpx client — it never touches the + # Gemini SDK. Requiring `google-cloud-aiplatform>=1.38` here turned a + # SDK-free Anthropic-protocol call into a hard dependency on the Gemini + # SDK (see #28084), breaking `/v1/messages/count_tokens` for Claude-on- + # Vertex on any LiteLLM install without that extra. Stay SDK-free. try: from litellm.llms.vertex_ai.vertex_ai_partner_models.count_tokens.handler import ( VertexAIPartnerModelsTokenCounter, diff --git a/litellm/llms/vertex_ai/vertex_gemma_models/main.py b/litellm/llms/vertex_ai/vertex_gemma_models/main.py index 82cfe6de984..b6bf2f73b72 100644 --- a/litellm/llms/vertex_ai/vertex_gemma_models/main.py +++ b/litellm/llms/vertex_ai/vertex_gemma_models/main.py @@ -31,7 +31,7 @@ from ..vertex_llm_base import VertexBase class VertexAIGemmaModels(VertexBase): def __init__(self) -> None: - pass + super().__init__() def completion( self, @@ -62,9 +62,6 @@ class VertexAIGemmaModels(VertexBase): try: import vertexai - from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import ( - VertexLLM, - ) from litellm.llms.vertex_ai.vertex_gemma_models.transformation import ( VertexGemmaConfig, ) @@ -83,9 +80,8 @@ class VertexAIGemmaModels(VertexBase): ) try: model = get_vertex_base_model_name(model=model) - vertex_httpx_logic = VertexLLM() - access_token, project_id = vertex_httpx_logic._ensure_access_token( + access_token, project_id = self._ensure_access_token( credentials=vertex_credentials, project_id=vertex_project, custom_llm_provider="vertex_ai", diff --git a/litellm/llms/vertex_ai/vertex_gemma_models/transformation.py b/litellm/llms/vertex_ai/vertex_gemma_models/transformation.py index 6c6446958bc..35cd54d65f6 100644 --- a/litellm/llms/vertex_ai/vertex_gemma_models/transformation.py +++ b/litellm/llms/vertex_ai/vertex_gemma_models/transformation.py @@ -91,6 +91,10 @@ class VertexGemmaConfig(OpenAIGPTConfig): "stream", None ) # Streaming not supported, will be faked client-side openai_request.pop("stream_options", None) # Stream options not supported + # Vertex Gemma's chatCompletions wrapper does not understand + # `context_management` (an Anthropic/Responses API concept). Strip it + # so the upstream endpoint does not 400 on the unknown field. + openai_request.pop("context_management", None) # Wrap in Vertex Gemma format return { diff --git a/litellm/llms/vertex_ai/vertex_llm_base.py b/litellm/llms/vertex_ai/vertex_llm_base.py index 6f687dae7e8..990063bb9fb 100644 --- a/litellm/llms/vertex_ai/vertex_llm_base.py +++ b/litellm/llms/vertex_ai/vertex_llm_base.py @@ -4,8 +4,10 @@ Base Vertex, Google AI Studio LLM Class Handles Authentication and generating request urls for Vertex AI and Google AI Studio """ +import asyncio import json import os +import threading from typing import TYPE_CHECKING, Any, Dict, Literal, Optional, Tuple import litellm @@ -30,6 +32,7 @@ GOOGLE_IMPORT_ERROR_MESSAGE = ( if TYPE_CHECKING: from google.auth.credentials import Credentials as GoogleCredentialsObject + from google.auth.credentials import TokenState else: GoogleCredentialsObject = Any @@ -42,10 +45,28 @@ class VertexBase: self._credentials: Optional[GoogleCredentialsObject] = None self._credentials_project_mapping: Dict[ Tuple[Optional[VERTEX_CREDENTIALS_TYPES], Optional[str]], - Tuple[GoogleCredentialsObject, str], + Tuple[GoogleCredentialsObject, Optional[str]], ] = {} self.project_id: Optional[str] = None self.async_handler: Optional[AsyncHTTPHandler] = None + # Per-credential-key asyncio.Lock for single-flight async refresh. + # Prevents thundering herd when token expires under high concurrency. + # Uses a regular dict (not WeakValueDictionary) so the lock identity is + # stable across concurrent callers — a weak reference can be GC'd + # between two coroutines arriving at the lock, breaking single-flight. + # An explicit refcount tracks the number of coroutines currently using + # each lock; the entry is pruned when the count reaches zero, so the + # dict stays bounded even in long-running high-cardinality deployments + # without depending on any private asyncio internals. + self._async_refresh_locks: Dict[tuple, asyncio.Lock] = {} + self._async_refresh_lock_refcounts: Dict[tuple, int] = {} + # Tracks in-flight background refresh tasks to avoid duplicate refreshes. + self._background_refresh_tasks: Dict[tuple, asyncio.Task] = {} + # Protects the sync get_access_token refresh path. + # Use RLock so that the reauthentication retry path (which calls + # back into get_access_token while still holding the lock) can + # re-acquire it without deadlocking the current thread. + self._sync_refresh_lock = threading.RLock() def get_vertex_region(self, vertex_region: Optional[str], model: str) -> str: import litellm @@ -77,7 +98,9 @@ class VertexBase: return vertex_region or "us-central1" def load_auth( - self, credentials: Optional[VERTEX_CREDENTIALS_TYPES], project_id: Optional[str] + self, + credentials: Optional[VERTEX_CREDENTIALS_TYPES], + project_id: Optional[str], ) -> Tuple[Any, str]: if credentials is not None: if isinstance(credentials, str): @@ -343,7 +366,241 @@ class VertexBase: except ImportError: raise ImportError(GOOGLE_IMPORT_ERROR_MESSAGE) - credentials.refresh(Request()) + # Serialize all refreshes on this VertexBase across threads. + # ``credentials.refresh()`` is not safe to call concurrently on the + # same credentials object, and this method is invoked from three + # places that can run on different threads: + # - sync ``get_access_token`` (already holds ``_sync_refresh_lock``) + # - the async slow path (via ``asyncify`` in a worker thread) + # - the background proactive refresh task (via ``asyncify``) + # ``_sync_refresh_lock`` is an ``RLock`` so reentrant acquisition + # from the sync path is safe. + with self._sync_refresh_lock: + credentials.refresh(Request()) + + def _acquire_async_refresh_lock(self, credential_cache_key: tuple) -> asyncio.Lock: + """Increment the refcount and return the lock for ``credential_cache_key``. + + Every call must be paired with ``_release_async_refresh_lock`` once the + caller is done with the lock so the entry can be pruned when no other + coroutine is holding or waiting on it. + """ + lock = self._async_refresh_locks.setdefault( + credential_cache_key, asyncio.Lock() + ) + self._async_refresh_lock_refcounts[credential_cache_key] = ( + self._async_refresh_lock_refcounts.get(credential_cache_key, 0) + 1 + ) + return lock + + def _release_async_refresh_lock( + self, credential_cache_key: tuple, lock: asyncio.Lock + ) -> None: + """Decrement the refcount and drop the lock entry when it reaches zero. + + Must be called only after the caller has released ``lock`` (i.e. once + the surrounding ``async with`` has exited). asyncio is cooperative, so + the decrement-then-pop sequence below runs atomically with respect to + other coroutines. + """ + remaining = self._async_refresh_lock_refcounts.get(credential_cache_key, 0) - 1 + if remaining > 0: + self._async_refresh_lock_refcounts[credential_cache_key] = remaining + return + self._async_refresh_lock_refcounts.pop(credential_cache_key, None) + if self._async_refresh_locks.get(credential_cache_key) is lock: + self._async_refresh_locks.pop(credential_cache_key, None) + + def _try_get_cached_token( + self, + credential_cache_key: tuple, + project_id: Optional[str], + ) -> Optional[Tuple[str, str]]: + """ + Look up cached credentials and return (token, project_id) if the token + is FRESH. Returns None if not cached or not fresh. + """ + from google.auth.credentials import TokenState + + creds, cached_project_id = self._unpack_cached_credentials(credential_cache_key) + if ( + creds is not None + and self._get_token_state(creds) == TokenState.FRESH + and creds.token is not None + and isinstance(creds.token, str) + ): + resolved_project = project_id or cached_project_id + if resolved_project: + return creds.token, resolved_project + return None + + def _try_get_usable_cached_token( + self, + credential_cache_key: tuple, + project_id: Optional[str], + ) -> Optional[Tuple[str, str, "TokenState", Any, Optional[str]]]: + """ + Look up cached credentials and return usable token info for FRESH or + STALE tokens (both are still valid for outbound requests). STALE + tokens are returned along with their state and the underlying + credentials object so the caller can schedule a background refresh + without holding the per-key async lock. + """ + from google.auth.credentials import TokenState + + creds, cached_project_id = self._unpack_cached_credentials(credential_cache_key) + if creds is None: + return None + token_state = self._get_token_state(creds) + if token_state not in (TokenState.FRESH, TokenState.STALE): + return None + if creds.token is None or not isinstance(creds.token, str): + return None + resolved_project = project_id or cached_project_id + if not resolved_project: + return None + return creds.token, resolved_project, token_state, creds, cached_project_id + + def _unpack_cached_credentials( + self, credential_cache_key: tuple + ) -> Tuple[Any, Optional[str]]: + """ + Return (credentials, project_id) from the cache, or (None, None) if + not cached. Handles both tuple and legacy cache formats. + """ + if credential_cache_key not in self._credentials_project_mapping: + return None, None + cached_entry = self._credentials_project_mapping[credential_cache_key] + if isinstance(cached_entry, tuple): + return cached_entry + return cached_entry, cached_entry.quota_project_id or getattr( + cached_entry, "project_id", None + ) + + def _get_token_state(self, credentials: Any) -> "TokenState": + """ + Return the token state using google-auth's TokenState enum. + + Falls back to expired/valid checks if token_state is unavailable + (e.g. older google-auth versions or mock objects in tests). + """ + from google.auth.credentials import TokenState as _TokenState + + token_state = getattr(credentials, "token_state", None) + if isinstance(token_state, _TokenState): + return token_state + # Fallback for credentials without a real token_state (e.g. mocks) + if getattr(credentials, "expired", True): + return _TokenState.INVALID + if getattr(credentials, "valid", False): + return _TokenState.FRESH + return _TokenState.INVALID + + async def _load_and_cache_credentials( + self, + credentials: Optional[VERTEX_CREDENTIALS_TYPES], + project_id: Optional[str], + credential_cache_key: tuple, + ) -> Tuple[Any, Optional[str]]: + """Load credentials via load_auth (in thread) and cache the result.""" + try: + _credentials, credential_project_id = await asyncify(self.load_auth)( + credentials=credentials, + project_id=project_id, + ) + except Exception as e: + verbose_logger.exception("Failed to load vertex credentials: %s", str(e)) + raise + if _credentials is None: + raise ValueError("Could not resolve credentials") + self._credentials_project_mapping[credential_cache_key] = ( + _credentials, + credential_project_id, + ) + return _credentials, credential_project_id + + async def _background_refresh_credentials( + self, + credentials: Any, + credential_cache_key: tuple, + credential_project_id: Optional[str], + ) -> None: + """ + Refresh credentials in the background without blocking the calling request. + + Called when the token is still valid but nearing expiry (proactive refresh). + Errors are logged but not raised — the current token is still usable. + """ + try: + verbose_logger.debug("Background proactive credential refresh") + await asyncify(self.refresh_auth)(credentials) + # Only update the cache if it still points at the credentials + # object we just refreshed. The per-key async lock is not held + # here, so a concurrent INVALID path may have already replaced + # this entry (e.g. via _handle_reauthentication_async, which + # creates a fresh credentials object). In that case our write + # would clobber the newer entry with a stale reference. + cached_creds, _ = self._unpack_cached_credentials(credential_cache_key) + if cached_creds is credentials: + self._credentials_project_mapping[credential_cache_key] = ( + credentials, + credential_project_id, + ) + except Exception: + verbose_logger.debug( + "Background credential refresh failed, will retry on next request", + exc_info=True, + ) + + async def _await_in_flight_background_refresh( + self, credential_cache_key: tuple + ) -> None: + """Wait for an in-flight background refresh to finish, if any. + + google-auth's ``Credentials.refresh()`` is not safe to invoke + concurrently on the same credentials object. Coroutines that need a + blocking refresh must first drain any background refresh that was + scheduled while a previous STALE token was being served. + """ + existing_task = self._background_refresh_tasks.get(credential_cache_key) + if existing_task is None or existing_task.done(): + return + try: + await existing_task + except Exception: + # Background refresh failures are already logged inside + # _background_refresh_credentials; the caller will fall through + # to its own blocking refresh. + pass + + def _schedule_background_refresh( + self, + credentials: Any, + credential_cache_key: tuple, + credential_project_id: Optional[str], + ) -> None: + """Kick off a single background refresh for ``credential_cache_key``. + + Skips scheduling if a refresh is already in flight. The done-callback + guards against removing a newer task that has replaced this one in the + tracking dict (done_callbacks are scheduled via ``call_soon``). + """ + existing = self._background_refresh_tasks.get(credential_cache_key) + if existing is not None and not existing.done(): + return + self._background_refresh_tasks.pop(credential_cache_key, None) + task = asyncio.create_task( + self._background_refresh_credentials( + credentials, credential_cache_key, credential_project_id + ) + ) + + def _drop_background_refresh_task(_fut: asyncio.Future[Any]) -> None: + if self._background_refresh_tasks.get(credential_cache_key) is _fut: + self._background_refresh_tasks.pop(credential_cache_key, None) + + task.add_done_callback(_drop_background_refresh_task) + self._background_refresh_tasks[credential_cache_key] = task def _ensure_access_token( self, @@ -563,6 +820,65 @@ class VertexBase: # Re-raise the original error for better context raise error + async def _handle_reauthentication_async( + self, + credentials: Optional[VERTEX_CREDENTIALS_TYPES], + project_id: Optional[str], + credential_cache_key: Tuple, + error: Exception, + ) -> Tuple[str, str]: + """ + Async reauthentication retry that stays within the per-key async lock. + """ + verbose_logger.debug( + f"Handling async reauthentication for project_id: {project_id}. " + f"Clearing cache and retrying once." + ) + + self._credentials_project_mapping.pop(credential_cache_key, None) + + try: + _credentials, credential_project_id = ( + await self._load_and_cache_credentials( + credentials=credentials, + project_id=project_id, + credential_cache_key=credential_cache_key, + ) + ) + if project_id is None and isinstance(credential_project_id, str): + project_id = credential_project_id + cache_credentials = ( + json.dumps(credentials) + if isinstance(credentials, dict) + else credentials + ) + resolved_cache_key = (cache_credentials, project_id) + # Always overwrite — any pre-existing entry at the resolved key + # references the OLD credentials object we just replaced, and + # leaving it would force the next request to do a redundant + # refresh/reauth before realizing the cached creds are stale. + self._credentials_project_mapping[resolved_cache_key] = ( + _credentials, + credential_project_id, + ) + + if _credentials.token is None or not isinstance(_credentials.token, str): + raise ValueError( + "Could not resolve credentials token. Got None or non-string token (type={})".format( + type(_credentials.token).__name__ + ) + ) + if project_id is None: + raise ValueError("Could not resolve project_id") + + return _credentials.token, project_id + except Exception as retry_error: + verbose_logger.error( + f"Async reauthentication retry failed for project_id: {project_id}. " + f"Original error: {str(error)}. Retry error: {str(retry_error)}" + ) + raise error + def get_access_token( self, credentials: Optional[VERTEX_CREDENTIALS_TYPES], @@ -646,7 +962,7 @@ class VertexBase: ) ## VALIDATE CREDENTIALS - verbose_logger.debug(f"Validating credentials for project_id: {project_id}") + verbose_logger.debug("Validating credentials") if ( project_id is None and credential_project_id is not None @@ -666,26 +982,27 @@ class VertexBase: raise ValueError("Credentials are None after loading") if _credentials.expired: - try: - verbose_logger.debug( - f"Credentials expired, refreshing for project_id: {project_id}" - ) - self.refresh_auth(_credentials) - self._credentials_project_mapping[credential_cache_key] = ( - _credentials, - credential_project_id, - ) - except Exception as e: - # if refresh fails, it's possible the user has re-authenticated via `gcloud auth application-default login` - # in this case, we should try to reload the credentials by clearing the cache and retrying - if "Reauthentication is needed" in str(e) and not _retry_reauth: - return self._handle_reauthentication( - credentials=credentials, - project_id=project_id, - credential_cache_key=credential_cache_key, - error=e, - ) - raise e + with self._sync_refresh_lock: + # Double-check after acquiring lock + if _credentials.expired: + try: + verbose_logger.debug("Credentials expired, refreshing") + self.refresh_auth(_credentials) + self._credentials_project_mapping[credential_cache_key] = ( + _credentials, + credential_project_id, + ) + except Exception as e: + # if refresh fails, it's possible the user has re-authenticated via `gcloud auth application-default login` + # in this case, we should try to reload the credentials by clearing the cache and retrying + if "Reauthentication is needed" in str(e) and not _retry_reauth: + return self._handle_reauthentication( + credentials=credentials, + project_id=project_id, + credential_cache_key=credential_cache_key, + error=e, + ) + raise e ## VALIDATION STEP if _credentials.token is None or not isinstance(_credentials.token, str): @@ -700,6 +1017,149 @@ class VertexBase: return _credentials.token, project_id + async def get_access_token_async( + self, + credentials: Optional[VERTEX_CREDENTIALS_TYPES], + project_id: Optional[str], + ) -> Tuple[str, str]: + """ + Async version of get_access_token with single-flight refresh coordination. + + Prevents thundering herd: when credentials expire under high concurrency, + only one coroutine refreshes while others wait on the lock. Uses native + async refresh for service_account and authorized_user credentials. + """ + from google.auth.credentials import TokenState + + cache_credentials = ( + json.dumps(credentials) if isinstance(credentials, dict) else credentials + ) + credential_cache_key = (cache_credentials, project_id) + + # === FAST PATH (no lock) === + # If credentials are FRESH or STALE, return immediately without + # touching the per-key async lock. STALE tokens are still usable; + # we kick off a deduplicated background refresh so subsequent + # requests get a fresh token, but we must not serialize concurrent + # callers on the lock just to schedule that refresh. + usable = self._try_get_usable_cached_token(credential_cache_key, project_id) + if usable is not None: + cached_token, resolved_project, token_state, creds, cached_project_id = ( + usable + ) + if token_state == TokenState.STALE: + self._schedule_background_refresh( + creds, credential_cache_key, cached_project_id + ) + return cached_token, resolved_project + + # === SLOW PATH (per-key lock) === + lock = self._acquire_async_refresh_lock(credential_cache_key) + try: + async with lock: + # Double-check after acquiring lock — another coroutine may have refreshed. + cached = self._try_get_cached_token(credential_cache_key, project_id) + if cached is not None: + return cached + + _credentials, credential_project_id = self._unpack_cached_credentials( + credential_cache_key + ) + + # Load credentials if not cached + if _credentials is None: + _credentials, credential_project_id = ( + await self._load_and_cache_credentials( + credentials, project_id, credential_cache_key + ) + ) + + # Resolve project_id from credentials if not provided + if project_id is None and isinstance(credential_project_id, str): + project_id = credential_project_id + resolved_cache_key = (cache_credentials, project_id) + # Always overwrite — a pre-existing entry at the resolved + # key may reference stale credentials (e.g. from before a + # reauth that only repopulated the unresolved key), which + # would force the next request through an unnecessary + # refresh/reauth cycle. + self._credentials_project_mapping[resolved_cache_key] = ( + _credentials, + credential_project_id, + ) + + # Use google-auth's token_state to decide refresh strategy: + # - STALE: token is usable but within REFRESH_THRESHOLD (3:45) of + # expiry — return it immediately and refresh in the background. + # - INVALID: token is expired or missing — must block on refresh. + token_state = self._get_token_state(_credentials) + + if token_state == TokenState.STALE: + if project_id is None: + raise ValueError("Could not resolve project_id") + current_token = _credentials.token + if current_token is None or not isinstance(current_token, str): + # Token is malformed despite STALE state — block on a full + # refresh using the same path as INVALID credentials. + token_state = TokenState.INVALID + else: + self._schedule_background_refresh( + _credentials, + credential_cache_key, + credential_project_id, + ) + return current_token, project_id + + if token_state == TokenState.INVALID: + # Drain any in-flight background refresh before invoking + # refresh_auth ourselves; google-auth's + # Credentials.refresh() is not safe to call concurrently + # on the same credentials object, and the background task + # runs outside this lock. + await self._await_in_flight_background_refresh(credential_cache_key) + cached = self._try_get_cached_token( + credential_cache_key, project_id + ) + if cached is not None: + return cached + + # Token is expired or missing — must block until refresh completes. + try: + verbose_logger.debug("Credentials expired, refreshing") + await asyncify(self.refresh_auth)(_credentials) + self._credentials_project_mapping[credential_cache_key] = ( + _credentials, + credential_project_id, + ) + except Exception as e: + if "Reauthentication is needed" in str(e): + verbose_logger.debug( + "Reauthentication needed, clearing cache and retrying" + ) + return await self._handle_reauthentication_async( + credentials=credentials, + project_id=project_id, + credential_cache_key=credential_cache_key, + error=e, + ) + raise + + # Final validation + if _credentials.token is None or not isinstance( + _credentials.token, str + ): + raise ValueError( + "Could not resolve credentials token. Got None or non-string token (type={})".format( + type(_credentials.token).__name__ + ) + ) + if project_id is None: + raise ValueError("Could not resolve project_id") + + return _credentials.token, project_id + finally: + self._release_async_refresh_lock(credential_cache_key, lock) + async def _ensure_access_token_async( self, credentials: Optional[VERTEX_CREDENTIALS_TYPES], @@ -714,13 +1174,10 @@ class VertexBase: if custom_llm_provider == "gemini": return "", "" else: - try: - return await asyncify(self.get_access_token)( - credentials=credentials, - project_id=project_id, - ) - except Exception as e: - raise e + return await self.get_access_token_async( + credentials=credentials, + project_id=project_id, + ) def set_headers( self, auth_header: Optional[str], extra_headers: Optional[dict] diff --git a/litellm/llms/vertex_ai/vertex_model_garden/main.py b/litellm/llms/vertex_ai/vertex_model_garden/main.py index 7240d9dce57..732d5f90dc2 100644 --- a/litellm/llms/vertex_ai/vertex_model_garden/main.py +++ b/litellm/llms/vertex_ai/vertex_model_garden/main.py @@ -57,7 +57,7 @@ def create_vertex_url( class VertexAIModelGardenModels(VertexBase): def __init__(self) -> None: - pass + super().__init__() def completion( self, @@ -89,9 +89,6 @@ class VertexAIModelGardenModels(VertexBase): import vertexai from litellm.llms.openai_like.chat.handler import OpenAILikeChatHandler - from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import ( - VertexLLM, - ) except Exception as e: raise VertexAIError( status_code=400, @@ -107,9 +104,8 @@ class VertexAIModelGardenModels(VertexBase): ) try: model = get_vertex_base_model_name(model=model) - vertex_httpx_logic = VertexLLM() - access_token, project_id = vertex_httpx_logic._ensure_access_token( + access_token, project_id = self._ensure_access_token( credentials=vertex_credentials, project_id=vertex_project, custom_llm_provider="vertex_ai", diff --git a/litellm/llms/vertex_ai/videos/transformation.py b/litellm/llms/vertex_ai/videos/transformation.py index ed6176cef05..b84966354b8 100644 --- a/litellm/llms/vertex_ai/videos/transformation.py +++ b/litellm/llms/vertex_ai/videos/transformation.py @@ -40,6 +40,29 @@ else: BaseLLMException = Any +def _build_vertex_video_usage_from_request_data( + request_data: Optional[Dict[str, Any]], +) -> Dict[str, Any]: + """Build usage metadata (duration, resolution) for video cost calculation.""" + usage_data: Dict[str, Any] = {} + if not request_data: + return usage_data + + parameters = request_data.get("parameters", {}) + duration = ( + parameters.get("durationSeconds") or DEFAULT_GOOGLE_VIDEO_DURATION_SECONDS + ) + if duration is not None: + try: + usage_data["duration_seconds"] = float(duration) + except (ValueError, TypeError): + pass + res = parameters.get("resolution") + if res is not None and str(res).strip() != "": + usage_data["video_resolution"] = str(res).strip().lower() + return usage_data + + def _convert_image_to_vertex_format(image_file) -> Dict[str, str]: """ Convert image file to Vertex AI format with base64 encoding and MIME type. @@ -363,23 +386,7 @@ class VertexAIVideoConfig(BaseVideoConfig, VertexBase): id=video_id, object="video", status="processing", model=model ) - usage_data: Dict[str, Any] = {} - if request_data: - parameters = request_data.get("parameters", {}) - duration = ( - parameters.get("durationSeconds") - or DEFAULT_GOOGLE_VIDEO_DURATION_SECONDS - ) - if duration is not None: - try: - usage_data["duration_seconds"] = float(duration) - except (ValueError, TypeError): - pass - res = parameters.get("resolution") - if res is not None and str(res).strip() != "": - usage_data["video_resolution"] = str(res).strip().lower() - - video_obj.usage = usage_data + video_obj.usage = _build_vertex_video_usage_from_request_data(request_data) return video_obj def transform_video_status_retrieve_request( @@ -647,15 +654,123 @@ class VertexAIVideoConfig(BaseVideoConfig, VertexBase): def transform_video_get_character_response(self, raw_response, logging_obj): raise NotImplementedError("video get character is not supported for Vertex AI") + def get_video_edit_prefetch_params( + self, + video_id: str, + api_base: str, + litellm_params: GenericLiteLLMParams, + headers: dict, + ) -> Tuple[str, Dict]: + """Return the fetchPredictOperation URL and body needed to retrieve the source video.""" + return self.transform_video_status_retrieve_request( + video_id=video_id, + api_base=api_base, + litellm_params=litellm_params, + headers=headers, + ) + def transform_video_edit_request( - self, prompt, video_id, api_base, litellm_params, headers, extra_body=None - ): - raise NotImplementedError("video edit is not supported for Vertex AI") + self, + prompt: str, + video_id: str, + api_base: str, + litellm_params: GenericLiteLLMParams, + headers: dict, + extra_body: Optional[Dict[str, Any]] = None, + prefetched_source_data: Optional[Dict[str, Any]] = None, + ) -> Tuple[str, Dict]: + """ + Build a predictLongRunning edit request from the pre-fetched source video. + + The actual fetchPredictOperation HTTP call is hoisted into the handler so + it can use the shared async/sync httpx client instead of blocking the loop. + """ + if prefetched_source_data is None: + raise ValueError( + "prefetched_source_data is required for Vertex AI video edit. " + "Ensure get_video_edit_prefetch_params is called by the handler." + ) + + if not prefetched_source_data.get("done", False): + raise ValueError( + "Source video generation is not complete yet. " + "Check the video status before editing." + ) + + videos = prefetched_source_data.get("response", {}).get("videos", []) + if not videos: + raise ValueError("No videos found in the completed operation. Cannot edit.") + + source_video = videos[0] + video_input: Dict[str, Any] = {} + if "gcsUri" in source_video: + video_input["gcsUri"] = source_video["gcsUri"] + elif "bytesBase64Encoded" in source_video: + video_input["bytesBase64Encoded"] = source_video["bytesBase64Encoded"] + video_input["mimeType"] = source_video.get("mimeType", "video/mp4") + else: + raise ValueError( + "Source video has neither gcsUri nor bytesBase64Encoded. Cannot edit." + ) + + operation_name = extract_original_video_id(video_id) + model = self.extract_model_from_operation_name(operation_name) or "" + + instance_dict: Dict[str, Any] = {"prompt": prompt, "video": video_input} + request_data: Dict[str, Any] = {"instances": [instance_dict]} + + if extra_body: + extra_body_copy = dict(extra_body) + nested_params = extra_body_copy.pop("parameters", None) + vertex_params: Dict[str, Any] = {} + if isinstance(nested_params, dict): + vertex_params.update(nested_params) + vertex_params.update(extra_body_copy) + if vertex_params: + request_data["parameters"] = vertex_params + + edit_url = f"{api_base.rstrip('/')}/{model}:predictLongRunning" + return edit_url, request_data def transform_video_edit_response( - self, raw_response, logging_obj, custom_llm_provider=None - ): - raise NotImplementedError("video edit is not supported for Vertex AI") + self, + raw_response: httpx.Response, + logging_obj: LiteLLMLoggingObj, + custom_llm_provider: Optional[str] = None, + request_data: Optional[Dict] = None, + ) -> VideoObject: + """ + Transform the Veo video edit response. + + Veo returns the same operation response as video generation: + {"name": "projects/.../operations/OPERATION_ID"} + + usage includes duration_seconds and optional video_resolution from the + edit request parameters for cost calculation. + """ + response_data = raw_response.json() + + operation_name = response_data.get("name") + if not operation_name: + raise ValueError(f"No operation name in Veo edit response: {response_data}") + + model = self.extract_model_from_operation_name(operation_name) or "" + + if custom_llm_provider: + video_id = encode_video_id_with_provider( + operation_name, custom_llm_provider, model + ) + else: + video_id = operation_name + + video_obj = VideoObject( + id=video_id, + object="video", + status="processing", + model=model, + ) + video_obj.usage = _build_vertex_video_usage_from_request_data(request_data) + return video_obj def transform_video_extension_request( self, diff --git a/litellm/llms/vllm/completion/transformation.py b/litellm/llms/vllm/completion/transformation.py index ec4c07e95d8..e03b07f9897 100644 --- a/litellm/llms/vllm/completion/transformation.py +++ b/litellm/llms/vllm/completion/transformation.py @@ -1,5 +1,5 @@ """ -Translates from OpenAI's `/v1/chat/completions` to the VLLM sdk `llm.generate`. +Translates from OpenAI's `/v1/chat/completions` to the VLLM sdk `llm.generate`. NOT RECOMMENDED FOR PRODUCTION USE. Use `hosted_vllm/` instead. """ diff --git a/litellm/llms/voyage/embedding/transformation_contextual.py b/litellm/llms/voyage/embedding/transformation_contextual.py index 40328062e09..1f5ca99f47d 100644 --- a/litellm/llms/voyage/embedding/transformation_contextual.py +++ b/litellm/llms/voyage/embedding/transformation_contextual.py @@ -1,6 +1,6 @@ """ -This module is used to transform the request and response for the Voyage contextualized embeddings API. -This would be used for all the contextualized embeddings models in Voyage. +This module is used to transform the request and response for the Voyage contextualized embeddings API. +This would be used for all the contextualized embeddings models in Voyage. """ from typing import List, Optional, Union diff --git a/litellm/llms/xai/chat/transformation.py b/litellm/llms/xai/chat/transformation.py index 6300868a641..7325c0596a6 100644 --- a/litellm/llms/xai/chat/transformation.py +++ b/litellm/llms/xai/chat/transformation.py @@ -1,4 +1,4 @@ -from typing import Any, AsyncIterator, Iterator, List, Optional, Tuple, Union +from typing import Any, AsyncIterator, Dict, Iterator, List, Optional, Tuple, Union import httpx @@ -26,6 +26,7 @@ from ...openai.chat.gpt_transformation import ( class XAIChatConfig(OpenAIGPTConfig): + @property def custom_llm_provider(self) -> Optional[str]: return "xai" @@ -225,21 +226,57 @@ class XAIChatConfig(OpenAIGPTConfig): verbose_logger.debug(f"Error extracting X.AI web search usage: {e}") self._fold_reasoning_tokens_into_completion(response) + self._normalize_openai_compatible_usage_totals(getattr(response, "usage", None)) return response @staticmethod - def _fold_reasoning_tokens_into_completion(model_response: ModelResponse) -> None: + def _fold_reasoning_tokens_into_completion( + target: Union[ModelResponse, Usage, Dict[str, Any], None], + ) -> None: """Reconcile xAI Usage to the OpenAI invariant. xAI accounts ``reasoning_tokens`` separately from ``completion_tokens`` while still summing them into ``total_tokens``. OpenAI's contract (o1/o3) folds reasoning into ``completion_tokens``, so fold here to keep ``total = prompt + completion``. Idempotent. + + Accepts a ``ModelResponse`` (non-streaming), a ``Usage`` object, or a + raw usage ``dict`` (streaming chunk) so streaming and non-streaming + paths stay in sync. """ - usage = getattr(model_response, "usage", None) + if target is None: + return + + if isinstance(target, ModelResponse): + usage: Union[Usage, Dict[str, Any], None] = getattr(target, "usage", None) + else: + usage = target if usage is None: return + if isinstance(usage, dict): + details = usage.get("completion_tokens_details") or {} + if isinstance(details, dict): + reasoning_tokens = int(details.get("reasoning_tokens") or 0) + else: + reasoning_tokens = int(getattr(details, "reasoning_tokens", 0) or 0) + if reasoning_tokens <= 0: + return + + prompt_tokens = int(usage.get("prompt_tokens") or 0) + completion_tokens = int(usage.get("completion_tokens") or 0) + total_tokens = int(usage.get("total_tokens") or 0) + + if total_tokens == prompt_tokens + completion_tokens: + return + + # Guard against double-counting if xAI changes accounting. + if total_tokens != prompt_tokens + completion_tokens + reasoning_tokens: + return + + usage["completion_tokens"] = completion_tokens + reasoning_tokens + return + details = getattr(usage, "completion_tokens_details", None) reasoning_tokens = ( int(getattr(details, "reasoning_tokens", 0) or 0) if details else 0 @@ -284,6 +321,25 @@ class XAIChatConfig(OpenAIGPTConfig): setattr(usage, "num_sources_used", int(num_sources_used)) verbose_logger.debug(f"X.AI web search sources used: {num_sources_used}") + @staticmethod + def _normalize_openai_compatible_usage_totals( + usage: Union[Usage, Dict[str, Any], None], + ) -> None: + if usage is None: + return + if isinstance(usage, dict): + prompt_tokens = int(usage.get("prompt_tokens") or 0) + completion_tokens = int(usage.get("completion_tokens") or 0) + expected_total = prompt_tokens + completion_tokens + if int(usage.get("total_tokens") or 0) < expected_total: + usage["total_tokens"] = expected_total + return + prompt_tokens = int(usage.prompt_tokens or 0) + completion_tokens = int(usage.completion_tokens or 0) + expected_total = prompt_tokens + completion_tokens + if int(usage.total_tokens or 0) < expected_total: + usage.total_tokens = expected_total + class XAIChatCompletionStreamingHandler(OpenAIChatCompletionStreamingHandler): def chunk_parser(self, chunk: dict) -> ModelResponseStream: @@ -304,4 +360,8 @@ class XAIChatCompletionStreamingHandler(OpenAIChatCompletionStreamingHandler): # Add a dummy choice with empty delta to ensure proper processing chunk["choices"] = [{"index": 0, "delta": {}, "finish_reason": None}] + if "usage" in chunk and chunk["usage"] is not None: + XAIChatConfig._fold_reasoning_tokens_into_completion(chunk["usage"]) + XAIChatConfig._normalize_openai_compatible_usage_totals(chunk["usage"]) + return super().chunk_parser(chunk) diff --git a/litellm/main.py b/litellm/main.py index c3d1c2e05b0..09c70998cf7 100644 --- a/litellm/main.py +++ b/litellm/main.py @@ -1491,7 +1491,9 @@ def completion( # type: ignore # noqa: PLR0915 provider.value for provider in LlmProviders ]: provider_config = ProviderConfigManager.get_provider_chat_config( - model=model, provider=LlmProviders(custom_llm_provider) + model=model, + provider=LlmProviders(custom_llm_provider), + base_model=base_model, ) if provider_config is not None: @@ -1550,6 +1552,7 @@ def completion( # type: ignore # noqa: PLR0915 "safety_identifier": safety_identifier, "service_tier": service_tier, "allowed_openai_params": kwargs.get("allowed_openai_params"), + "base_model": base_model, } optional_params = get_optional_params( **optional_param_args, **non_default_params @@ -1670,6 +1673,10 @@ def completion( # type: ignore # noqa: PLR0915 reasoning_summary=_reasoning_summary_for_bridge, ) + # Use base_model (the true underlying model) for Azure model-type + # detection when the deployment name differs from the model name. + _azure_detection_model = base_model or model + if responses_api_model_info.get("mode") == "responses": from litellm.completion_extras import responses_api_bridge @@ -1713,7 +1720,9 @@ def completion( # type: ignore # noqa: PLR0915 and OpenAIGPT5Config.is_model_gpt_5_model(model) ) or ( custom_llm_provider == "azure" - and litellm.AzureOpenAIGPT5Config.is_model_gpt_5_model(model) + and litellm.AzureOpenAIGPT5Config.is_model_gpt_5_model( + _azure_detection_model + ) ): optional_params, _ = strip_reasoning_summary_aliases_from_optional_params( optional_params @@ -1766,7 +1775,9 @@ def completion( # type: ignore # noqa: PLR0915 if max_retries is not None: optional_params["max_retries"] = max_retries - if litellm.AzureOpenAIO1Config().is_o_series_model(model=model): + if litellm.AzureOpenAIO1Config().is_o_series_model( + model=_azure_detection_model + ): ## LOAD CONFIG - if set config = litellm.AzureOpenAIO1Config.get_config() for k, v in config.items(): @@ -5116,6 +5127,24 @@ def embedding( # noqa: PLR0915 client=client, aembedding=aembedding, ) + elif custom_llm_provider == "oci": + if headers is None: + headers = {} + response = base_llm_http_handler.embedding( + model=model, + input=input, + custom_llm_provider=custom_llm_provider, + api_base=api_base, + api_key=api_key, + logging_obj=logging, + timeout=timeout, + model_response=EmbeddingResponse(), + optional_params=optional_params, + client=client, + aembedding=aembedding, + litellm_params=litellm_params_dict, + headers=headers, + ) elif custom_llm_provider == "cohere" or custom_llm_provider == "cohere_chat": cohere_key = ( api_key @@ -5720,6 +5749,33 @@ def embedding( # noqa: PLR0915 aembedding=aembedding, headers=headers, ) + elif custom_llm_provider == "dashscope": + dashscope_key = ( + api_key or litellm.api_key or get_secret_str("DASHSCOPE_API_KEY") + ) + if dashscope_key is None: + raise ValueError( + "Missing API key for DashScope. Set DASHSCOPE_API_KEY environment variable or pass api_key parameter." + ) + if extra_headers is not None and isinstance(extra_headers, dict): + headers = extra_headers + else: + headers = {} + response = base_llm_http_handler.embedding( + model=model, + input=input, + timeout=timeout, + custom_llm_provider=custom_llm_provider, + logging_obj=logging, + api_base=api_base, + optional_params=optional_params, + litellm_params={}, + model_response=EmbeddingResponse(), + api_key=dashscope_key, + client=client, + aembedding=aembedding, + headers=headers, + ) elif custom_llm_provider == "ovhcloud": api_key = api_key or litellm.api_key or get_secret_str("OVHCLOUD_API_KEY") api_base = ( @@ -5769,22 +5825,6 @@ def embedding( # noqa: PLR0915 aembedding=aembedding, litellm_params={}, ) - elif custom_llm_provider == "oci": - response = base_llm_http_handler.embedding( - model=model, - input=input, - custom_llm_provider=custom_llm_provider, - api_base=api_base, - api_key=api_key, - logging_obj=logging, - timeout=timeout, - model_response=EmbeddingResponse(), - optional_params=optional_params, - client=client, - aembedding=aembedding, - litellm_params=litellm_params_dict, - headers=headers, - ) elif custom_llm_provider in litellm._custom_providers: custom_handler: Optional[CustomLLM] = None for item in litellm.custom_provider_map: @@ -6575,8 +6615,7 @@ def transcription( api_key=api_key, ) # type: ignore - if dynamic_api_key is not None: - api_key = dynamic_api_key + api_key = dynamic_api_key if dynamic_api_key is not None else api_key optional_params = get_optional_params_transcription( model=model, @@ -6616,7 +6655,7 @@ def transcription( provider=LlmProviders(custom_llm_provider), ) - if custom_llm_provider == "azure": + if custom_llm_provider == "azure" and provider_config is None: # azure configs api_base = api_base or litellm.api_base or get_secret_str("AZURE_API_BASE") diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 8df25d2c9b5..ce6d4ac824c 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -731,7 +731,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "anthropic.claude-haiku-4-5@20251001": { @@ -755,7 +754,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_streaming": true, "supports_native_structured_output": true }, @@ -926,8 +924,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "anthropic.claude-opus-4-20250514-v1:0": { "cache_creation_input_token_cost": 1.875e-05, @@ -952,8 +949,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "anthropic.claude-opus-4-5-20251101-v1:0": { "cache_creation_input_token_cost": 6.25e-06, @@ -977,12 +973,12 @@ "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_minimal_reasoning_effort": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159, - "supports_native_structured_output": true + "supports_native_structured_output": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "high" }, "anthropic.claude-opus-4-6-v1": { "cache_creation_input_token_cost": 6.25e-06, @@ -1009,10 +1005,10 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, + "supports_output_config": true, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "bedrock_output_config_effort_ceiling": "max" }, "global.anthropic.claude-opus-4-6-v1": { "cache_creation_input_token_cost": 6.25e-06, @@ -1039,10 +1035,10 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, + "supports_output_config": true, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "bedrock_output_config_effort_ceiling": "max" }, "us.anthropic.claude-opus-4-6-v1": { "cache_creation_input_token_cost": 6.875e-06, @@ -1069,10 +1065,10 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, + "supports_output_config": true, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "bedrock_output_config_effort_ceiling": "max" }, "eu.anthropic.claude-opus-4-6-v1": { "cache_creation_input_token_cost": 6.875e-06, @@ -1098,10 +1094,10 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, + "supports_output_config": true, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "bedrock_output_config_effort_ceiling": "max" }, "au.anthropic.claude-opus-4-6-v1": { "cache_creation_input_token_cost": 6.875e-06, @@ -1127,10 +1123,10 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, + "supports_output_config": true, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "bedrock_output_config_effort_ceiling": "max" }, "anthropic.claude-opus-4-7": { "cache_creation_input_token_cost": 6.25e-06, @@ -1158,10 +1154,10 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" }, "anthropic.claude-mythos-preview": { "input_cost_per_token": 0, @@ -1175,8 +1171,8 @@ "supports_vision": true, "supports_prompt_caching": false, "supports_reasoning": true, - "supports_minimal_reasoning_effort": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_output_config": true }, "global.anthropic.claude-opus-4-7": { "cache_creation_input_token_cost": 6.25e-06, @@ -1204,10 +1200,10 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" }, "us.anthropic.claude-opus-4-7": { "cache_creation_input_token_cost": 6.875e-06, @@ -1235,10 +1231,10 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" }, "eu.anthropic.claude-opus-4-7": { "cache_creation_input_token_cost": 6.875e-06, @@ -1265,10 +1261,10 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" }, "au.anthropic.claude-opus-4-7": { "cache_creation_input_token_cost": 6.875e-06, @@ -1295,10 +1291,165 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" + }, + "anthropic.claude-opus-4-8": { + "cache_creation_input_token_cost": 6.25e-06, + "cache_creation_input_token_cost_above_1hr": 1e-05, + "cache_read_input_token_cost": 5e-07, + "input_cost_per_token": 5e-06, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 2.5e-05, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" + }, + "global.anthropic.claude-opus-4-8": { + "cache_creation_input_token_cost": 6.25e-06, + "cache_creation_input_token_cost_above_1hr": 1e-05, + "cache_read_input_token_cost": 5e-07, + "input_cost_per_token": 5e-06, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 2.5e-05, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" + }, + "us.anthropic.claude-opus-4-8": { + "cache_creation_input_token_cost": 6.875e-06, + "cache_creation_input_token_cost_above_1hr": 1.1e-05, + "cache_read_input_token_cost": 5.5e-07, + "input_cost_per_token": 5.5e-06, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 2.75e-05, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" + }, + "eu.anthropic.claude-opus-4-8": { + "cache_creation_input_token_cost": 6.875e-06, + "cache_creation_input_token_cost_above_1hr": 1.1e-05, + "cache_read_input_token_cost": 5.5e-07, + "input_cost_per_token": 5.5e-06, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 2.75e-05, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" + }, + "au.anthropic.claude-opus-4-8": { + "cache_creation_input_token_cost": 6.875e-06, + "cache_creation_input_token_cost_above_1hr": 1.1e-05, + "cache_read_input_token_cost": 5.5e-07, + "input_cost_per_token": 5.5e-06, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 2.75e-05, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" }, "anthropic.claude-sonnet-4-6": { "cache_creation_input_token_cost": 3.75e-06, @@ -1326,9 +1477,8 @@ "supports_max_reasoning_effort": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "global.anthropic.claude-sonnet-4-6": { "cache_creation_input_token_cost": 3.75e-06, @@ -1356,9 +1506,8 @@ "supports_max_reasoning_effort": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "us.anthropic.claude-sonnet-4-6": { "cache_creation_input_token_cost": 4.125e-06, @@ -1386,9 +1535,8 @@ "supports_max_reasoning_effort": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "eu.anthropic.claude-sonnet-4-6": { "cache_creation_input_token_cost": 4.125e-06, @@ -1415,9 +1563,8 @@ "supports_max_reasoning_effort": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "au.anthropic.claude-sonnet-4-6": { "cache_creation_input_token_cost": 4.125e-06, @@ -1444,9 +1591,36 @@ "supports_max_reasoning_effort": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true + }, + "jp.anthropic.claude-sonnet-4-6": { + "cache_creation_input_token_cost": 4.125e-06, + "cache_read_input_token_cost": 3.3e-07, + "input_cost_per_token": 3.3e-06, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 64000, + "max_tokens": 64000, + "mode": "chat", + "output_cost_per_token": 1.65e-05, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": true, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_max_reasoning_effort": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_native_structured_output": true, + "supports_output_config": true }, "anthropic.claude-sonnet-4-20250514-v1:0": { "cache_creation_input_token_cost": 3.75e-06, @@ -1475,8 +1649,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "anthropic.claude-sonnet-4-5-20250929-v1:0": { "cache_creation_input_token_cost": 3.75e-06, @@ -1508,7 +1681,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159, "supports_native_structured_output": true }, "anthropic.claude-v1": { @@ -1759,7 +1931,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "apac.anthropic.claude-3-sonnet-20240229-v1:0": { @@ -1805,8 +1976,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "assemblyai/best": { "input_cost_per_second": 3.333e-05, @@ -1848,7 +2018,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "azure/ada": { @@ -1936,10 +2105,10 @@ "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_minimal_reasoning_effort": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "supports_output_config": true }, "azure_ai/claude-opus-4-6": { "input_cost_per_token": 5e-06, @@ -1966,9 +2135,8 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true, + "supports_max_reasoning_effort": true }, "azure_ai/claude-opus-4-7": { "input_cost_per_token": 5e-06, @@ -1996,9 +2164,35 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "tool_use_system_prompt_tokens": 159, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_max_reasoning_effort": true + }, + "azure_ai/claude-opus-4-8": { + "input_cost_per_token": 5e-06, + "output_cost_per_token": 2.5e-05, + "litellm_provider": "azure_ai", + "max_input_tokens": 200000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "cache_creation_input_token_cost": 6.25e-06, + "cache_creation_input_token_cost_above_1hr": 1e-05, + "cache_read_input_token_cost": 5e-07, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_max_reasoning_effort": true }, "azure_ai/claude-opus-4-1": { "cache_creation_input_token_cost": 1.875e-05, @@ -2063,8 +2257,7 @@ "supports_max_reasoning_effort": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "azure/computer-use-preview": { "input_cost_per_token": 3e-06, @@ -2112,6 +2305,380 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "azure_ai/gpt-5.4": { + "cache_read_input_token_cost": 2.5e-07, + "cache_read_input_token_cost_above_272k_tokens": 5e-07, + "cache_read_input_token_cost_priority": 5e-07, + "cache_read_input_token_cost_above_272k_tokens_priority": 1e-06, + "input_cost_per_token": 2.5e-06, + "input_cost_per_token_above_272k_tokens": 5e-06, + "input_cost_per_token_priority": 5e-06, + "input_cost_per_token_above_272k_tokens_priority": 1e-05, + "litellm_provider": "azure_ai", + "max_input_tokens": 1050000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 1.5e-05, + "output_cost_per_token_above_272k_tokens": 2.25e-05, + "output_cost_per_token_priority": 3e-05, + "output_cost_per_token_above_272k_tokens_priority": 4.5e-05, + "source": "https://ai.azure.com/catalog/models/gpt-5.4", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_service_tier": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": true + }, + "azure_ai/gpt-5.4-2026-03-05": { + "cache_read_input_token_cost": 2.5e-07, + "cache_read_input_token_cost_above_272k_tokens": 5e-07, + "cache_read_input_token_cost_priority": 5e-07, + "cache_read_input_token_cost_above_272k_tokens_priority": 1e-06, + "input_cost_per_token": 2.5e-06, + "input_cost_per_token_above_272k_tokens": 5e-06, + "input_cost_per_token_priority": 5e-06, + "input_cost_per_token_above_272k_tokens_priority": 1e-05, + "litellm_provider": "azure_ai", + "max_input_tokens": 1050000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 1.5e-05, + "output_cost_per_token_above_272k_tokens": 2.25e-05, + "output_cost_per_token_priority": 3e-05, + "output_cost_per_token_above_272k_tokens_priority": 4.5e-05, + "source": "https://ai.azure.com/catalog/models/gpt-5.4", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_service_tier": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": true + }, + "azure_ai/gpt-5.4-pro": { + "cache_read_input_token_cost": 3e-06, + "cache_read_input_token_cost_above_272k_tokens": 6e-06, + "cache_read_input_token_cost_priority": 6e-06, + "cache_read_input_token_cost_above_272k_tokens_priority": 1.2e-05, + "input_cost_per_token": 3e-05, + "input_cost_per_token_above_272k_tokens": 6e-05, + "input_cost_per_token_priority": 6e-05, + "input_cost_per_token_above_272k_tokens_priority": 0.00012, + "litellm_provider": "azure_ai", + "max_input_tokens": 1050000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "responses", + "output_cost_per_token": 0.00018, + "output_cost_per_token_above_272k_tokens": 0.00027, + "output_cost_per_token_priority": 0.00036, + "output_cost_per_token_above_272k_tokens_priority": 0.00054, + "source": "https://ai.azure.com/catalog/models/gpt-5.4-pro", + "supported_endpoints": [ + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": false, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_service_tier": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": false, + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": true + }, + "azure_ai/gpt-5.4-pro-2026-03-05": { + "cache_read_input_token_cost": 3e-06, + "cache_read_input_token_cost_above_272k_tokens": 6e-06, + "cache_read_input_token_cost_priority": 6e-06, + "cache_read_input_token_cost_above_272k_tokens_priority": 1.2e-05, + "input_cost_per_token": 3e-05, + "input_cost_per_token_above_272k_tokens": 6e-05, + "input_cost_per_token_priority": 6e-05, + "input_cost_per_token_above_272k_tokens_priority": 0.00012, + "litellm_provider": "azure_ai", + "max_input_tokens": 1050000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "responses", + "output_cost_per_token": 0.00018, + "output_cost_per_token_above_272k_tokens": 0.00027, + "output_cost_per_token_priority": 0.00036, + "output_cost_per_token_above_272k_tokens_priority": 0.00054, + "source": "https://ai.azure.com/catalog/models/gpt-5.4-pro", + "supported_endpoints": [ + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": false, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_service_tier": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": false, + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": true + }, + "azure_ai/gpt-5.4-mini": { + "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_above_272k_tokens": 1.5e-07, + "cache_read_input_token_cost_priority": 1.5e-07, + "cache_read_input_token_cost_above_272k_tokens_priority": 3e-07, + "input_cost_per_token": 7.5e-07, + "input_cost_per_token_above_272k_tokens": 1.5e-06, + "input_cost_per_token_priority": 1.5e-06, + "input_cost_per_token_above_272k_tokens_priority": 3e-06, + "litellm_provider": "azure_ai", + "max_input_tokens": 400000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 4.5e-06, + "output_cost_per_token_above_272k_tokens": 6.75e-06, + "output_cost_per_token_priority": 9e-06, + "output_cost_per_token_above_272k_tokens_priority": 1.35e-05, + "source": "https://ai.azure.com/catalog/models/gpt-5.4-mini", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_service_tier": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false + }, + "azure_ai/gpt-5.4-mini-2026-03-17": { + "cache_read_input_token_cost": 7.5e-08, + "cache_read_input_token_cost_above_272k_tokens": 1.5e-07, + "cache_read_input_token_cost_priority": 1.5e-07, + "cache_read_input_token_cost_above_272k_tokens_priority": 3e-07, + "input_cost_per_token": 7.5e-07, + "input_cost_per_token_above_272k_tokens": 1.5e-06, + "input_cost_per_token_priority": 1.5e-06, + "input_cost_per_token_above_272k_tokens_priority": 3e-06, + "litellm_provider": "azure_ai", + "max_input_tokens": 400000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 4.5e-06, + "output_cost_per_token_above_272k_tokens": 6.75e-06, + "output_cost_per_token_priority": 9e-06, + "output_cost_per_token_above_272k_tokens_priority": 1.35e-05, + "source": "https://ai.azure.com/catalog/models/gpt-5.4-mini", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_service_tier": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false + }, + "azure_ai/gpt-5.4-nano": { + "cache_read_input_token_cost": 2e-08, + "cache_read_input_token_cost_above_272k_tokens": 4e-08, + "cache_read_input_token_cost_priority": 4e-08, + "cache_read_input_token_cost_above_272k_tokens_priority": 8e-08, + "input_cost_per_token": 2e-07, + "input_cost_per_token_above_272k_tokens": 4e-07, + "input_cost_per_token_priority": 4e-07, + "input_cost_per_token_above_272k_tokens_priority": 8e-07, + "litellm_provider": "azure_ai", + "max_input_tokens": 400000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 1.25e-06, + "output_cost_per_token_above_272k_tokens": 1.875e-06, + "output_cost_per_token_priority": 2.5e-06, + "output_cost_per_token_above_272k_tokens_priority": 3.75e-06, + "source": "https://ai.azure.com/catalog/models/gpt-5.4-nano", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_service_tier": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false + }, + "azure_ai/gpt-5.4-nano-2026-03-17": { + "cache_read_input_token_cost": 2e-08, + "cache_read_input_token_cost_above_272k_tokens": 4e-08, + "cache_read_input_token_cost_priority": 4e-08, + "cache_read_input_token_cost_above_272k_tokens_priority": 8e-08, + "input_cost_per_token": 2e-07, + "input_cost_per_token_above_272k_tokens": 4e-07, + "input_cost_per_token_priority": 4e-07, + "input_cost_per_token_above_272k_tokens_priority": 8e-07, + "litellm_provider": "azure_ai", + "max_input_tokens": 400000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 1.25e-06, + "output_cost_per_token_above_272k_tokens": 1.875e-06, + "output_cost_per_token_priority": 2.5e-06, + "output_cost_per_token_above_272k_tokens_priority": 3.75e-06, + "source": "https://ai.azure.com/catalog/models/gpt-5.4-nano", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_service_tier": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false + }, "azure_ai/model_router": { "input_cost_per_token": 1.4e-07, "output_cost_per_token": 0, @@ -3521,7 +4088,7 @@ "supports_tool_choice": true }, "azure/gpt-4o-mini-transcribe": { - "input_cost_per_audio_token": 3e-06, + "input_cost_per_audio_token": 1.25e-06, "input_cost_per_token": 1.25e-06, "litellm_provider": "azure", "max_input_tokens": 16000, @@ -3596,7 +4163,7 @@ "supports_tool_choice": true }, "azure/gpt-4o-transcribe": { - "input_cost_per_audio_token": 6e-06, + "input_cost_per_audio_token": 2.5e-06, "input_cost_per_token": 2.5e-06, "litellm_provider": "azure", "max_input_tokens": 16000, @@ -3608,7 +4175,7 @@ ] }, "azure/gpt-4o-transcribe-diarize": { - "input_cost_per_audio_token": 6e-06, + "input_cost_per_audio_token": 2.5e-06, "input_cost_per_token": 2.5e-06, "litellm_provider": "azure", "max_input_tokens": 16000, @@ -5651,6 +6218,17 @@ "mode": "audio_speech", "source": "https://azure.microsoft.com/en-us/pricing/calculator/" }, + "azure/speech/azure-stt": { + "audio_transcription_config": "azure_speech", + "input_cost_per_second": 0.0002777778, + "litellm_provider": "azure", + "mode": "audio_transcription", + "output_cost_per_second": 0.0, + "source": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/speech-services/", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ] + }, "azure/tts-1": { "input_cost_per_character": 1.5e-05, "litellm_provider": "azure", @@ -8974,7 +9552,7 @@ "supports_vision": true }, "gpt-4o-transcribe-diarize": { - "input_cost_per_audio_token": 6e-06, + "input_cost_per_audio_token": 2.5e-06, "input_cost_per_token": 2.5e-06, "litellm_provider": "openai", "max_input_tokens": 16000, @@ -9053,8 +9631,7 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true, - "tool_use_system_prompt_tokens": 159 + "supports_web_search": true }, "claude-3-haiku-20240307": { "cache_creation_input_token_cost": 3e-07, @@ -9072,8 +9649,7 @@ "supports_prompt_caching": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 264 + "supports_vision": true }, "claude-3-opus-20240229": { "cache_creation_input_token_cost": 1.875e-05, @@ -9092,8 +9668,7 @@ "supports_prompt_caching": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 395 + "supports_vision": true }, "claude-4-opus-20250514": { "cache_creation_input_token_cost": 1.875e-05, @@ -9118,8 +9693,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "claude-4-sonnet-20250514": { "cache_creation_input_token_cost": 3.75e-06, @@ -9149,8 +9723,7 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true, - "tool_use_system_prompt_tokens": 159 + "supports_web_search": true }, "claude-sonnet-4-5": { "cache_creation_input_token_cost": 3.75e-06, @@ -9179,8 +9752,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 346 + "supports_vision": true }, "claude-sonnet-4-5-20250929": { "cache_creation_input_token_cost": 3.75e-06, @@ -9210,8 +9782,7 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true, - "tool_use_system_prompt_tokens": 346 + "supports_web_search": true }, "claude-sonnet-4-6": { "cache_creation_input_token_cost": 3.75e-06, @@ -9228,6 +9799,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, @@ -9238,8 +9810,7 @@ "supports_max_reasoning_effort": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "claude-sonnet-4-5-20250929-v1:0": { "cache_creation_input_token_cost": 3.75e-06, @@ -9263,8 +9834,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "claude-opus-4-1": { "cache_creation_input_token_cost": 1.875e-05, @@ -9290,8 +9860,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "claude-opus-4-1-20250805": { "cache_creation_input_token_cost": 1.875e-05, @@ -9318,8 +9887,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "claude-opus-4-20250514": { "cache_creation_input_token_cost": 1.875e-05, @@ -9346,8 +9914,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "claude-opus-4-5-20251101": { "cache_creation_input_token_cost": 6.25e-06, @@ -9371,11 +9938,10 @@ "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_minimal_reasoning_effort": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_output_config": true }, "claude-opus-4-5": { "cache_creation_input_token_cost": 6.25e-06, @@ -9399,11 +9965,10 @@ "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_minimal_reasoning_effort": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_output_config": true }, "claude-opus-4-6": { "cache_creation_input_token_cost": 6.25e-06, @@ -9421,6 +9986,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -9430,13 +9996,12 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "provider_specific_entry": { "us": 1.1, "fast": 6.0 }, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true, + "supports_max_reasoning_effort": true }, "claude-opus-4-6-20260205": { "cache_creation_input_token_cost": 6.25e-06, @@ -9454,6 +10019,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -9463,13 +10029,12 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "provider_specific_entry": { "us": 1.1, "fast": 6.0 }, "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "claude-opus-4-7": { "cache_creation_input_token_cost": 6.25e-06, @@ -9487,6 +10052,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -9498,12 +10064,11 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_max_reasoning_effort": true, - "tool_use_system_prompt_tokens": 346, "provider_specific_entry": { "us": 1.1, "fast": 6.0 }, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "claude-opus-4-7-20260416": { "cache_creation_input_token_cost": 6.25e-06, @@ -9521,6 +10086,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -9532,12 +10098,45 @@ "supports_vision": true, "supports_xhigh_reasoning_effort": true, "supports_max_reasoning_effort": true, - "tool_use_system_prompt_tokens": 346, "provider_specific_entry": { "us": 1.1, "fast": 6.0 }, - "supports_minimal_reasoning_effort": true + "supports_output_config": true + }, + "claude-opus-4-8": { + "cache_creation_input_token_cost": 6.25e-06, + "cache_creation_input_token_cost_above_1hr": 1e-05, + "cache_read_input_token_cost": 5e-07, + "input_cost_per_token": 5e-06, + "litellm_provider": "anthropic", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 2.5e-05, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_max_reasoning_effort": true, + "provider_specific_entry": { + "us": 1.1, + "fast": 2.0 + }, + "supports_output_config": true }, "claude-sonnet-4-20250514": { "deprecation_date": "2026-05-14", @@ -9568,8 +10167,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "cloudflare/@cf/meta/llama-2-7b-chat-fp16": { "input_cost_per_token": 1.923e-06, @@ -10814,8 +11412,8 @@ "supports_assistant_prefill": true, "supports_function_calling": true, "supports_reasoning": true, - "supports_minimal_reasoning_effort": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_output_config": true }, "databricks/databricks-claude-sonnet-4": { "input_cost_per_token": 2.9999900000000002e-06, @@ -12981,7 +13579,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "eu.anthropic.claude-3-5-sonnet-20240620-v1:0": { @@ -13109,8 +13706,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "eu.anthropic.claude-opus-4-20250514-v1:0": { "cache_creation_input_token_cost": 1.875e-05, @@ -13135,8 +13731,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "eu.anthropic.claude-sonnet-4-20250514-v1:0": { "cache_creation_input_token_cost": 3.75e-06, @@ -13165,8 +13760,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "eu.anthropic.claude-sonnet-4-5-20250929-v1:0": { "cache_creation_input_token_cost": 4.125e-06, @@ -13196,7 +13790,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "eu.meta.llama3-2-1b-instruct-v1:0": { @@ -13574,6 +14167,21 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "fireworks_ai/accounts/fireworks/models/glm-5p1": { + "cache_read_input_token_cost": 2.6e-07, + "input_cost_per_token": 1.4e-06, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 202800, + "max_output_tokens": 202800, + "max_tokens": 202800, + "mode": "chat", + "output_cost_per_token": 4.4e-06, + "source": "https://fireworks.ai/models/fireworks/glm-5p1", + "supports_function_calling": false, + "supports_reasoning": true, + "supports_response_schema": false, + "supports_tool_choice": false + }, "fireworks_ai/accounts/fireworks/models/gpt-oss-120b": { "input_cost_per_token": 1.5e-07, "litellm_provider": "fireworks_ai", @@ -13840,6 +14448,21 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "fireworks_ai/glm-5p1": { + "cache_read_input_token_cost": 2.6e-07, + "input_cost_per_token": 1.4e-06, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 202800, + "max_output_tokens": 202800, + "max_tokens": 202800, + "mode": "chat", + "output_cost_per_token": 4.4e-06, + "source": "https://fireworks.ai/models/fireworks/glm-5p1", + "supports_function_calling": false, + "supports_reasoning": true, + "supports_response_schema": false, + "supports_tool_choice": false + }, "fireworks_ai/kimi-k2p5": { "cache_read_input_token_cost": 1e-07, "input_cost_per_token": 6e-07, @@ -14549,6 +15172,64 @@ "web_search_billing_unit": "per_query", "supports_service_tier": true }, + "gemini-3.1-flash-lite": { + "cache_read_input_token_cost": 4.5e-08, + "cache_read_input_token_cost_per_audio_token": 9e-08, + "input_cost_per_audio_token": 9e-07, + "input_cost_per_token": 4.5e-07, + "litellm_provider": "vertex_ai-language-models", + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_images_per_prompt": 3000, + "max_input_tokens": 1048576, + "max_output_tokens": 65536, + "max_pdf_size_mb": 30, + "max_tokens": 65536, + "max_video_length": 1, + "max_videos_per_prompt": 10, + "mode": "chat", + "output_cost_per_reasoning_token": 2.7e-06, + "output_cost_per_token": 2.7e-06, + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#gemini-models", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_input": true, + "supports_audio_output": false, + "supports_code_execution": true, + "supports_file_search": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "supports_native_streaming": true, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, + "web_search_billing_unit": "per_query", + "supports_service_tier": true + }, "deep-research-pro-preview-12-2025": { "input_cost_per_image": 0.0011, "input_cost_per_token": 2e-06, @@ -15237,6 +15918,64 @@ }, "web_search_billing_unit": "per_query" }, + "vertex_ai/gemini-3.5-flash": { + "cache_read_input_token_cost": 1.5e-07, + "input_cost_per_token": 1.5e-06, + "input_cost_per_audio_token": 1e-06, + "litellm_provider": "vertex_ai", + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_images_per_prompt": 3000, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_pdf_size_mb": 30, + "max_tokens": 65535, + "max_video_length": 1, + "max_videos_per_prompt": 10, + "mode": "chat", + "output_cost_per_reasoning_token": 9e-06, + "output_cost_per_token": 9e-06, + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_input": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "supports_native_streaming": true, + "input_cost_per_token_priority": 2.7e-06, + "input_cost_per_audio_token_priority": 1.8e-06, + "output_cost_per_token_priority": 1.62e-05, + "cache_read_input_token_cost_priority": 2.7e-07, + "supports_service_tier": true, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, + "web_search_billing_unit": "per_query" + }, "vertex_ai/gemini-3.1-pro-preview": { "cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, @@ -15551,14 +16290,17 @@ "uses_embed_content": true }, "vertex_ai/gemini-embedding-2-preview": { - "input_cost_per_token": 1.5e-07, + "input_cost_per_audio_per_second": 0.00016, + "input_cost_per_image": 0.00012, + "input_cost_per_token": 2e-07, + "input_cost_per_video_per_second": 0.00079, "litellm_provider": "vertex_ai", "max_input_tokens": 8192, "max_tokens": 8192, "mode": "embedding", "output_cost_per_token": 0, "output_vector_size": 3072, - "source": "https://ai.google.dev/gemini-api/docs/embeddings#multimodal", + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", "supports_multimodal": true, "uses_embed_content": true }, @@ -15573,7 +16315,7 @@ "mode": "embedding", "output_cost_per_token": 0, "output_vector_size": 3072, - "source": "https://ai.google.dev/gemini-api/docs/embeddings#multimodal", + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", "supports_multimodal": true, "uses_embed_content": true }, @@ -16552,6 +17294,66 @@ "web_search_billing_unit": "per_query", "supports_service_tier": true }, + "gemini/gemini-3.1-flash-lite": { + "cache_read_input_token_cost": 4.5e-08, + "cache_read_input_token_cost_per_audio_token": 9e-08, + "input_cost_per_audio_token": 9e-07, + "input_cost_per_token": 4.5e-07, + "litellm_provider": "gemini", + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_images_per_prompt": 3000, + "max_input_tokens": 1048576, + "max_output_tokens": 65536, + "max_pdf_size_mb": 30, + "max_tokens": 65536, + "max_video_length": 1, + "max_videos_per_prompt": 10, + "mode": "chat", + "output_cost_per_reasoning_token": 2.7e-06, + "output_cost_per_token": 2.7e-06, + "rpm": 15, + "source": "https://ai.google.dev/gemini-api/docs/pricing", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_input": true, + "supports_audio_output": false, + "supports_code_execution": true, + "supports_file_search": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "supports_native_streaming": true, + "tpm": 250000, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, + "web_search_billing_unit": "per_query", + "supports_service_tier": true + }, "gemini/gemini-3-flash-preview": { "cache_read_input_token_cost": 5e-08, "input_cost_per_audio_token": 1e-06, @@ -16611,6 +17413,67 @@ }, "web_search_billing_unit": "per_query" }, + "gemini/gemini-3.5-flash": { + "cache_read_input_token_cost": 1.5e-07, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-06, + "litellm_provider": "gemini", + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_images_per_prompt": 3000, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_pdf_size_mb": 30, + "max_tokens": 65535, + "max_video_length": 1, + "max_videos_per_prompt": 10, + "mode": "chat", + "output_cost_per_reasoning_token": 9e-06, + "output_cost_per_token": 9e-06, + "rpm": 2000, + "source": "https://ai.google.dev/pricing/gemini-3", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_output": false, + "supports_audio_input": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "supports_native_streaming": true, + "tpm": 800000, + "input_cost_per_token_priority": 2.7e-06, + "input_cost_per_audio_token_priority": 1.8e-06, + "output_cost_per_token_priority": 1.62e-05, + "cache_read_input_token_cost_priority": 2.7e-07, + "supports_service_tier": true, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, + "web_search_billing_unit": "per_query" + }, "gemini/gemini-3.1-pro-preview": { "cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, @@ -16796,6 +17659,65 @@ }, "web_search_billing_unit": "per_query" }, + "gemini-3.5-flash": { + "cache_read_input_token_cost": 1.5e-07, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-06, + "litellm_provider": "vertex_ai-language-models", + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_images_per_prompt": 3000, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_pdf_size_mb": 30, + "max_tokens": 65535, + "max_video_length": 1, + "max_videos_per_prompt": 10, + "mode": "chat", + "output_cost_per_reasoning_token": 9e-06, + "output_cost_per_token": 9e-06, + "source": "https://ai.google.dev/pricing/gemini-3", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_output": false, + "supports_audio_input": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "supports_native_streaming": true, + "input_cost_per_token_priority": 2.7e-06, + "input_cost_per_audio_token_priority": 1.8e-06, + "output_cost_per_token_priority": 1.62e-05, + "cache_read_input_token_cost_priority": 2.7e-07, + "supports_service_tier": true, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, + "web_search_billing_unit": "per_query" + }, "gemini/gemini-2.5-pro-preview-tts": { "cache_read_input_token_cost": 1.25e-07, "cache_read_input_token_cost_above_200k_tokens": 2.5e-07, @@ -17153,22 +18075,9 @@ }, "github_copilot/claude-haiku-4.5": { "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, - "mode": "chat", - "supported_endpoints": [ - "/v1/chat/completions" - ], - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_vision": true - }, - "github_copilot/claude-opus-4.5": { - "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "max_tokens": 32000, "mode": "chat", "supported_endpoints": [ "/v1/chat/completions" @@ -17176,7 +18085,22 @@ "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": true, - "supports_minimal_reasoning_effort": true + "supports_reasoning": true + }, + "github_copilot/claude-opus-4.5": { + "litellm_provider": "github_copilot", + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "max_tokens": 32000, + "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_vision": true, + "supports_reasoning": true, + "supports_output_config": true }, "github_copilot/claude-opus-4.6-fast": { "litellm_provider": "github_copilot", @@ -17191,6 +18115,22 @@ "supports_parallel_function_calling": true, "supports_vision": true }, + "github_copilot/claude-opus-4.7": { + "litellm_provider": "github_copilot", + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "max_tokens": 64000, + "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/messages" + ], + "supports_vision": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true, + "supports_reasoning": true + }, "github_copilot/claude-opus-41": { "litellm_provider": "github_copilot", "max_input_tokens": 80000, @@ -17217,16 +18157,33 @@ }, "github_copilot/claude-sonnet-4.5": { "litellm_provider": "github_copilot", - "max_input_tokens": 128000, - "max_output_tokens": 16000, - "max_tokens": 16000, + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "max_tokens": 32000, "mode": "chat", "supported_endpoints": [ "/v1/chat/completions" ], "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_vision": true + "supports_vision": true, + "supports_reasoning": true + }, + "github_copilot/claude-sonnet-4.6": { + "litellm_provider": "github_copilot", + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "max_tokens": 32000, + "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/messages" + ], + "supports_vision": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true, + "supports_reasoning": true }, "github_copilot/gemini-2.5-pro": { "litellm_provider": "github_copilot", @@ -17236,7 +18193,25 @@ "mode": "chat", "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_vision": true + "supports_vision": true, + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_reasoning": true + }, + "github_copilot/gemini-3-flash-preview": { + "litellm_provider": "github_copilot", + "max_input_tokens": 128000, + "max_output_tokens": 64000, + "max_tokens": 64000, + "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_vision": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_reasoning": true }, "github_copilot/gemini-3-pro-preview": { "litellm_provider": "github_copilot", @@ -17248,13 +18223,30 @@ "supports_parallel_function_calling": true, "supports_vision": true }, + "github_copilot/gemini-3.1-pro-preview": { + "litellm_provider": "github_copilot", + "max_input_tokens": 128000, + "max_output_tokens": 64000, + "max_tokens": 64000, + "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_vision": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_reasoning": true + }, "github_copilot/gpt-3.5-turbo": { "litellm_provider": "github_copilot", "max_input_tokens": 16384, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", - "supports_function_calling": true + "supports_function_calling": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-3.5-turbo-0613": { "litellm_provider": "github_copilot", @@ -17262,7 +18254,10 @@ "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", - "supports_function_calling": true + "supports_function_calling": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-4": { "litellm_provider": "github_copilot", @@ -17270,7 +18265,22 @@ "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", - "supports_function_calling": true + "supports_function_calling": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] + }, + "github_copilot/gpt-4-0125-preview": { + "litellm_provider": "github_copilot", + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "max_tokens": 4096, + "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_parallel_function_calling": true }, "github_copilot/gpt-4-0613": { "litellm_provider": "github_copilot", @@ -17278,16 +18288,22 @@ "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", - "supports_function_calling": true + "supports_function_calling": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-4-o-preview": { "litellm_provider": "github_copilot", - "max_input_tokens": 64000, + "max_input_tokens": 128000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", "supports_function_calling": true, - "supports_parallel_function_calling": true + "supports_parallel_function_calling": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-4.1": { "litellm_provider": "github_copilot", @@ -17298,7 +18314,10 @@ "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, - "supports_vision": true + "supports_vision": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-4.1-2025-04-14": { "litellm_provider": "github_copilot", @@ -17309,68 +18328,89 @@ "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, - "supports_vision": true + "supports_vision": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-41-copilot": { "litellm_provider": "github_copilot", - "mode": "completion" + "mode": "chat" }, "github_copilot/gpt-4o": { "litellm_provider": "github_copilot", - "max_input_tokens": 64000, + "max_input_tokens": 128000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_vision": true + "supports_vision": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-4o-2024-05-13": { "litellm_provider": "github_copilot", - "max_input_tokens": 64000, + "max_input_tokens": 128000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_vision": true + "supports_vision": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-4o-2024-08-06": { "litellm_provider": "github_copilot", - "max_input_tokens": 64000, - "max_output_tokens": 16384, - "max_tokens": 16384, - "mode": "chat", - "supports_function_calling": true, - "supports_parallel_function_calling": true - }, - "github_copilot/gpt-4o-2024-11-20": { - "litellm_provider": "github_copilot", - "max_input_tokens": 64000, + "max_input_tokens": 128000, "max_output_tokens": 16384, "max_tokens": 16384, "mode": "chat", "supports_function_calling": true, "supports_parallel_function_calling": true, - "supports_vision": true + "supported_endpoints": [ + "/v1/chat/completions" + ] + }, + "github_copilot/gpt-4o-2024-11-20": { + "litellm_provider": "github_copilot", + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_vision": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-4o-mini": { "litellm_provider": "github_copilot", - "max_input_tokens": 64000, + "max_input_tokens": 128000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", "supports_function_calling": true, - "supports_parallel_function_calling": true + "supports_parallel_function_calling": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-4o-mini-2024-07-18": { "litellm_provider": "github_copilot", - "max_input_tokens": 64000, + "max_input_tokens": 128000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", "supports_function_calling": true, - "supports_parallel_function_calling": true + "supports_parallel_function_calling": true, + "supported_endpoints": [ + "/v1/chat/completions" + ] }, "github_copilot/gpt-5": { "litellm_provider": "github_copilot", @@ -17389,14 +18429,19 @@ }, "github_copilot/gpt-5-mini": { "litellm_provider": "github_copilot", - "max_input_tokens": 128000, + "max_input_tokens": 264000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, - "supports_vision": true + "supports_vision": true, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses" + ], + "supports_reasoning": true }, "github_copilot/gpt-5.1": { "litellm_provider": "github_copilot", @@ -17429,7 +18474,7 @@ }, "github_copilot/gpt-5.2": { "litellm_provider": "github_copilot", - "max_input_tokens": 128000, + "max_input_tokens": 264000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", @@ -17440,11 +18485,27 @@ "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, - "supports_vision": true + "supports_vision": true, + "supports_reasoning": true + }, + "github_copilot/gpt-5.2-codex": { + "litellm_provider": "github_copilot", + "max_input_tokens": 400000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "responses", + "supported_endpoints": [ + "/v1/responses" + ], + "supports_vision": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true, + "supports_reasoning": true }, "github_copilot/gpt-5.3-codex": { "litellm_provider": "github_copilot", - "max_input_tokens": 128000, + "max_input_tokens": 400000, "max_output_tokens": 128000, "max_tokens": 128000, "mode": "responses", @@ -17454,25 +18515,96 @@ "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, - "supports_vision": true + "supports_vision": true, + "supports_reasoning": true + }, + "github_copilot/gpt-5.4": { + "litellm_provider": "github_copilot", + "max_input_tokens": 400000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses" + ], + "supports_vision": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true, + "supports_reasoning": true + }, + "github_copilot/gpt-5.4-mini": { + "litellm_provider": "github_copilot", + "max_input_tokens": 400000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "responses", + "supported_endpoints": [ + "/v1/responses" + ], + "supports_vision": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true, + "supports_reasoning": true + }, + "github_copilot/gpt-5.5": { + "litellm_provider": "github_copilot", + "max_input_tokens": 400000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "responses", + "supported_endpoints": [ + "/v1/responses" + ], + "supports_vision": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true, + "supports_reasoning": true + }, + "github_copilot/oswe-vscode-prime": { + "litellm_provider": "github_copilot", + "max_input_tokens": 264000, + "max_output_tokens": 64000, + "max_tokens": 64000, + "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses" + ], + "supports_vision": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true }, "github_copilot/text-embedding-3-small": { "litellm_provider": "github_copilot", "max_input_tokens": 8191, "max_tokens": 8191, - "mode": "embedding" + "mode": "embedding", + "supported_endpoints": [ + "/v1/embeddings" + ] }, "github_copilot/text-embedding-3-small-inference": { "litellm_provider": "github_copilot", "max_input_tokens": 8191, "max_tokens": 8191, - "mode": "embedding" + "mode": "embedding", + "supported_endpoints": [ + "/v1/embeddings" + ] }, "github_copilot/text-embedding-ada-002": { "litellm_provider": "github_copilot", "max_input_tokens": 8191, "max_tokens": 8191, - "mode": "embedding" + "mode": "embedding", + "supported_endpoints": [ + "/v1/embeddings" + ] }, "chatgpt/gpt-5.4": { "litellm_provider": "chatgpt", @@ -17690,7 +18822,7 @@ "output_cost_per_token": 2.5e-05, "supports_function_calling": true, "supports_vision": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "gmi/anthropic/claude-sonnet-4.5": { "input_cost_per_token": 3e-06, @@ -17990,7 +19122,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "global.anthropic.claude-sonnet-4-20250514-v1:0": { @@ -18020,8 +19151,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "global.anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 1.25e-06, @@ -18044,7 +19174,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "global.amazon.nova-2-lite-v1:0": { @@ -18266,6 +19395,8 @@ "output_cost_per_token": 8e-06, "output_cost_per_token_batches": 4e-06, "output_cost_per_token_priority": 1.4e-05, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", @@ -18339,6 +19470,8 @@ "output_cost_per_token": 1.6e-06, "output_cost_per_token_batches": 8e-07, "output_cost_per_token_priority": 2.8e-06, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", @@ -18412,6 +19545,8 @@ "output_cost_per_token": 4e-07, "output_cost_per_token_batches": 2e-07, "output_cost_per_token_priority": 8e-07, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", @@ -18483,6 +19618,8 @@ "output_cost_per_token": 1e-05, "output_cost_per_token_batches": 5e-06, "output_cost_per_token_priority": 1.7e-05, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -18524,6 +19661,8 @@ "mode": "chat", "output_cost_per_token": 1e-05, "output_cost_per_token_batches": 5e-06, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -18545,6 +19684,8 @@ "mode": "chat", "output_cost_per_token": 1e-05, "output_cost_per_token_batches": 5e-06, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -18833,6 +19974,8 @@ "output_cost_per_token": 6e-07, "output_cost_per_token_batches": 3e-07, "output_cost_per_token_priority": 1e-06, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_pdf_input": true, @@ -18988,7 +20131,7 @@ "supports_vision": true }, "gpt-4o-mini-transcribe": { - "input_cost_per_audio_token": 3e-06, + "input_cost_per_audio_token": 1.25e-06, "input_cost_per_token": 1.25e-06, "litellm_provider": "openai", "max_input_tokens": 16000, @@ -19118,7 +20261,7 @@ "supports_vision": true }, "gpt-4o-transcribe": { - "input_cost_per_audio_token": 6e-06, + "input_cost_per_audio_token": 2.5e-06, "input_cost_per_token": 2.5e-06, "litellm_provider": "openai", "max_input_tokens": 16000, @@ -19536,6 +20679,8 @@ "output_cost_per_token": 1e-05, "output_cost_per_token_flex": 5e-06, "output_cost_per_token_priority": 2e-05, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", @@ -20458,6 +21603,8 @@ "mode": "responses", "output_cost_per_token": 0.00012, "output_cost_per_token_batches": 6e-05, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "supported_endpoints": [ "/v1/batch", "/v1/responses" @@ -20864,6 +22011,8 @@ "output_cost_per_token": 2e-06, "output_cost_per_token_flex": 1e-06, "output_cost_per_token_priority": 3.6e-06, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "supported_endpoints": [ "/v1/chat/completions", "/v1/batch", @@ -20945,6 +22094,8 @@ "max_input_tokens": 272000, "max_output_tokens": 128000, "max_tokens": 128000, + "regional_processing_uplift_multiplier_eu": 1.10, + "regional_processing_uplift_multiplier_us": 1.10, "mode": "chat", "output_cost_per_token": 4e-07, "output_cost_per_token_flex": 2e-07, @@ -22130,7 +23281,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "jp.anthropic.claude-haiku-4-5-20251001-v1:0": { @@ -22153,7 +23303,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "crusoe/deepseek-ai/DeepSeek-R1-0528": { @@ -23730,6 +24879,21 @@ "supports_tool_choice": true, "supports_vision": true }, + "mistral/ministral-8b-2512": { + "input_cost_per_token": 1.5e-07, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 1.5e-07, + "source": "https://mistral.ai/pricing", + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "mistral/mistral-tiny": { "input_cost_per_token": 2.5e-07, "litellm_provider": "mistral", @@ -25545,6 +26709,51 @@ "supports_function_calling": true, "supports_response_schema": false }, + "oci/openai.gpt-5": { + "input_cost_per_token": 1.25e-06, + "litellm_provider": "oci", + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 1e-05, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true + }, + "oci/openai.gpt-5-mini": { + "input_cost_per_token": 2.5e-07, + "litellm_provider": "oci", + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 2e-06, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true + }, + "oci/openai.gpt-5-nano": { + "input_cost_per_token": 5e-08, + "litellm_provider": "oci", + "max_input_tokens": 272000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 4e-07, + "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing", + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_vision": true + }, "oci/google.gemini-2.5-pro": { "input_cost_per_token": 1.25e-06, "litellm_provider": "oci", @@ -26035,8 +27244,7 @@ "supports_computer_use": true, "supports_function_calling": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "openrouter/anthropic/claude-3.7-sonnet": { "input_cost_per_image": 0.0048, @@ -26052,8 +27260,7 @@ "supports_function_calling": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "openrouter/anthropic/claude-opus-4": { "input_cost_per_image": 0.0048, @@ -26072,8 +27279,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "openrouter/anthropic/claude-opus-4.1": { "input_cost_per_image": 0.0048, @@ -26093,8 +27299,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "openrouter/anthropic/claude-sonnet-4": { "input_cost_per_image": 0.0048, @@ -26117,8 +27322,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "openrouter/anthropic/claude-sonnet-4.6": { "cache_creation_input_token_cost": 3.75e-06, @@ -26142,9 +27346,7 @@ "supports_reasoning": true, "supports_max_reasoning_effort": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159, - "supports_minimal_reasoning_effort": true + "supports_vision": true }, "openrouter/anthropic/claude-opus-4.5": { "cache_creation_input_token_cost": 6.25e-06, @@ -26159,12 +27361,11 @@ "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_output_config": true }, "openrouter/anthropic/claude-opus-4.6": { "cache_creation_input_token_cost": 6.25e-06, @@ -26183,9 +27384,7 @@ "supports_reasoning": true, "supports_max_reasoning_effort": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 346, - "supports_minimal_reasoning_effort": true + "supports_vision": true }, "openrouter/anthropic/claude-sonnet-4.5": { "input_cost_per_image": 0.0048, @@ -26208,8 +27407,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "openrouter/anthropic/claude-haiku-4.5": { "cache_creation_input_token_cost": 1.25e-06, @@ -26227,8 +27425,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 346 + "supports_vision": true }, "openrouter/anthropic/claude-opus-4.7": { "cache_creation_input_token_cost": 6.25e-06, @@ -26250,8 +27447,7 @@ "supports_max_reasoning_effort": true, "supports_tool_choice": true, "supports_vision": true, - "supports_xhigh_reasoning_effort": true, - "tool_use_system_prompt_tokens": 346 + "supports_xhigh_reasoning_effort": true }, "openrouter/bytedance/ui-tars-1.5-7b": { "input_cost_per_token": 1e-07, @@ -26574,6 +27770,58 @@ "supports_web_search": true, "tpm": 800000 }, + "openrouter/google/gemini-3.1-flash-lite": { + "cache_read_input_token_cost": 2.5e-08, + "cache_read_input_token_cost_per_audio_token": 5e-08, + "input_cost_per_audio_token": 5e-07, + "input_cost_per_token": 2.5e-07, + "litellm_provider": "openrouter", + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_images_per_prompt": 3000, + "max_input_tokens": 1048576, + "max_output_tokens": 65536, + "max_pdf_size_mb": 30, + "max_tokens": 65536, + "max_video_length": 1, + "max_videos_per_prompt": 10, + "mode": "chat", + "output_cost_per_reasoning_token": 1.5e-06, + "output_cost_per_token": 1.5e-06, + "rpm": 2000, + "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-3.1-flash-lite", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_input": true, + "supports_audio_output": false, + "supports_code_execution": true, + "supports_file_search": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "tpm": 800000 + }, "openrouter/google/gemini-3.1-pro-preview": { "cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, @@ -27383,10 +28631,10 @@ "supports_tool_choice": true }, "openrouter/xiaomi/mimo-v2-flash": { - "input_cost_per_token": 9e-08, - "output_cost_per_token": 2.9e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 3e-07, "cache_creation_input_token_cost": 0.0, - "cache_read_input_token_cost": 0.0, + "cache_read_input_token_cost": 1e-08, "litellm_provider": "openrouter", "max_input_tokens": 262144, "max_output_tokens": 16384, @@ -27396,7 +28644,43 @@ "supports_tool_choice": true, "supports_reasoning": true, "supports_vision": false, - "supports_prompt_caching": false + "supports_prompt_caching": true + }, + "openrouter/xiaomi/mimo-v2.5-pro": { + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, + "cache_creation_input_token_cost": 0.0, + "cache_read_input_token_cost": 2e-07, + "litellm_provider": "openrouter", + "max_input_tokens": 1048576, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_vision": false, + "supports_response_schema": true, + "supports_prompt_caching": true + }, + "openrouter/xiaomi/mimo-v2.5": { + "input_cost_per_token": 4e-07, + "output_cost_per_token": 2e-06, + "cache_creation_input_token_cost": 0.0, + "cache_read_input_token_cost": 8e-08, + "litellm_provider": "openrouter", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_vision": true, + "supports_audio_input": true, + "supports_video_input": true, + "supports_response_schema": true, + "supports_prompt_caching": true }, "openrouter/z-ai/glm-4.7": { "input_cost_per_token": 4e-07, @@ -28127,14 +29411,16 @@ "mode": "responses", "supports_web_search": true, "supports_reasoning": false, - "supports_function_calling": true + "supports_function_calling": true, + "supports_output_config": true }, "perplexity/anthropic/claude-opus-4-7": { "litellm_provider": "perplexity", "mode": "responses", "supports_web_search": true, "supports_reasoning": false, - "supports_function_calling": true + "supports_function_calling": true, + "supports_output_config": true }, "perplexity/anthropic/claude-opus-4-5": { "litellm_provider": "perplexity", @@ -28142,7 +29428,7 @@ "supports_web_search": true, "supports_reasoning": false, "supports_function_calling": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "perplexity/anthropic/claude-sonnet-4-5": { "litellm_provider": "perplexity", @@ -28348,6 +29634,24 @@ "supports_tool_choice": true, "source": "https://aws.amazon.com/bedrock/pricing/" }, + "reducto/parse-legacy": { + "litellm_provider": "reducto", + "mode": "ocr", + "ocr_cost_per_credit": 0.015, + "source": "https://reducto.ai/pricing", + "supported_endpoints": [ + "/v1/ocr" + ] + }, + "reducto/parse-v3": { + "litellm_provider": "reducto", + "mode": "ocr", + "ocr_cost_per_credit": 0.015, + "source": "https://reducto.ai/pricing", + "supported_endpoints": [ + "/v1/ocr" + ] + }, "recraft/recraftv2": { "litellm_provider": "recraft", "mode": "image_generation", @@ -30377,7 +31681,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "us.anthropic.claude-3-5-sonnet-20240620-v1:0": { @@ -30505,8 +31808,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "us.anthropic.claude-sonnet-4-5-20250929-v1:0": { "cache_creation_input_token_cost": 4.125e-06, @@ -30538,7 +31840,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "us-gov.anthropic.claude-sonnet-4-5-20250929-v1:0": { @@ -30564,7 +31865,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "au.anthropic.claude-haiku-4-5-20251001-v1:0": { @@ -30586,7 +31886,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true }, "us.anthropic.claude-opus-4-20250514-v1:0": { @@ -30612,8 +31911,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "us.anthropic.claude-opus-4-5-20251101-v1:0": { "cache_creation_input_token_cost": 6.875e-06, @@ -30634,15 +31932,15 @@ "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159, - "supports_native_structured_output": true + "supports_native_structured_output": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "high" }, "global.anthropic.claude-opus-4-5-20251101-v1:0": { "cache_creation_input_token_cost": 6.25e-06, @@ -30663,15 +31961,15 @@ "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159, - "supports_native_structured_output": true + "supports_native_structured_output": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "high" }, "eu.anthropic.claude-opus-4-5-20251101-v1:0": { "cache_creation_input_token_cost": 6.25e-06, @@ -30691,15 +31989,15 @@ "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159, - "supports_native_structured_output": true + "supports_native_structured_output": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "high" }, "us.anthropic.claude-sonnet-4-20250514-v1:0": { "cache_creation_input_token_cost": 3.75e-06, @@ -30728,8 +32026,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "us.deepseek.r1-v1:0": { "input_cost_per_token": 1.35e-06, @@ -31272,13 +32569,13 @@ "output_cost_per_token": 2.5e-05, "supports_assistant_prefill": true, "supports_computer_use": true, - "supports_minimal_reasoning_effort": true, "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "supports_output_config": true }, "vercel_ai_gateway/anthropic/claude-opus-4.6": { "cache_creation_input_token_cost": 6.25e-06, @@ -31298,7 +32595,7 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "vercel_ai_gateway/anthropic/claude-sonnet-4": { "cache_creation_input_token_cost": 3.75e-06, @@ -32306,8 +33603,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "vertex_ai/claude-3-haiku": { "input_cost_per_token": 2.5e-07, @@ -32410,8 +33706,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "vertex_ai/claude-opus-4-1": { "cache_creation_input_token_cost": 1.875e-05, @@ -32465,14 +33760,13 @@ "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_output_config": true }, "vertex_ai/claude-opus-4-5@20251101": { "cache_creation_input_token_cost": 6.25e-06, @@ -32492,15 +33786,14 @@ "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, - "supports_minimal_reasoning_effort": true, "supports_pdf_input": true, "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 159, - "supports_native_streaming": true + "supports_native_streaming": true, + "supports_output_config": true }, "vertex_ai/claude-opus-4-6": { "cache_creation_input_token_cost": 6.25e-06, @@ -32526,9 +33819,8 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true, + "supports_max_reasoning_effort": true }, "vertex_ai/claude-opus-4-6@default": { "cache_creation_input_token_cost": 6.25e-06, @@ -32554,9 +33846,8 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_output_config": true, + "supports_max_reasoning_effort": true }, "vertex_ai/claude-opus-4-7": { "cache_creation_input_token_cost": 6.25e-06, @@ -32583,9 +33874,7 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "tool_use_system_prompt_tokens": 346, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_max_reasoning_effort": true }, "vertex_ai/claude-opus-4-7@default": { "cache_creation_input_token_cost": 6.25e-06, @@ -32612,9 +33901,63 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "tool_use_system_prompt_tokens": 346, - "supports_max_reasoning_effort": true, - "supports_minimal_reasoning_effort": true + "supports_max_reasoning_effort": true + }, + "vertex_ai/claude-opus-4-8": { + "cache_creation_input_token_cost": 6.25e-06, + "cache_creation_input_token_cost_above_1hr": 1e-05, + "cache_read_input_token_cost": 5e-07, + "input_cost_per_token": 5e-06, + "litellm_provider": "vertex_ai-anthropic_models", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 2.5e-05, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_max_reasoning_effort": true + }, + "vertex_ai/claude-opus-4-8@default": { + "cache_creation_input_token_cost": 6.25e-06, + "cache_creation_input_token_cost_above_1hr": 1e-05, + "cache_read_input_token_cost": 5e-07, + "input_cost_per_token": 5e-06, + "litellm_provider": "vertex_ai-anthropic_models", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 2.5e-05, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_max_reasoning_effort": true }, "vertex_ai/claude-sonnet-4-5": { "cache_creation_input_token_cost": 3.75e-06, @@ -32662,13 +34005,12 @@ "supports_max_reasoning_effort": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "search_context_cost_per_query": { "search_context_size_high": 0.01, "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "vertex_ai/claude-sonnet-4-5@20250929": { "cache_creation_input_token_cost": 3.75e-06, @@ -32720,8 +34062,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "vertex_ai/claude-sonnet-4": { "cache_creation_input_token_cost": 3.75e-06, @@ -32750,8 +34091,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "vertex_ai/claude-sonnet-4@20250514": { "cache_creation_input_token_cost": 3.75e-06, @@ -32780,8 +34120,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": true, - "tool_use_system_prompt_tokens": 159 + "supports_vision": true }, "vertex_ai/mistralai/codestral-2@001": { "input_cost_per_token": 3e-07, @@ -33050,6 +34389,64 @@ }, "web_search_billing_unit": "per_query" }, + "vertex_ai/gemini-3.1-flash-lite": { + "cache_read_input_token_cost": 4.5e-08, + "cache_read_input_token_cost_per_audio_token": 9e-08, + "input_cost_per_audio_token": 9e-07, + "input_cost_per_token": 4.5e-07, + "litellm_provider": "vertex_ai-language-models", + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_images_per_prompt": 3000, + "max_input_tokens": 1048576, + "max_output_tokens": 65536, + "max_pdf_size_mb": 30, + "max_tokens": 65536, + "max_video_length": 1, + "max_videos_per_prompt": 10, + "mode": "chat", + "output_cost_per_reasoning_token": 2.7e-06, + "output_cost_per_token": 2.7e-06, + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#gemini-models", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "supports_audio_input": true, + "supports_audio_output": false, + "supports_code_execution": true, + "supports_file_search": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_url_context": true, + "supports_video_input": true, + "supports_vision": true, + "supports_web_search": true, + "supports_native_streaming": true, + "search_context_cost_per_query": { + "search_context_size_low": 0.014, + "search_context_size_medium": 0.014, + "search_context_size_high": 0.014 + }, + "web_search_billing_unit": "per_query", + "supports_service_tier": true + }, "vertex_ai/deep-research-pro-preview-12-2025": { "input_cost_per_image": 0.0011, "input_cost_per_token": 2e-06, @@ -38930,7 +40327,7 @@ ] }, "gpt-4o-mini-transcribe-2025-03-20": { - "input_cost_per_audio_token": 3e-06, + "input_cost_per_audio_token": 1.25e-06, "input_cost_per_token": 1.25e-06, "litellm_provider": "openai", "max_input_tokens": 16000, @@ -38942,7 +40339,7 @@ ] }, "gpt-4o-mini-transcribe-2025-12-15": { - "input_cost_per_audio_token": 3e-06, + "input_cost_per_audio_token": 1.25e-06, "input_cost_per_token": 1.25e-06, "litellm_provider": "openai", "max_input_tokens": 16000, @@ -39716,13 +41113,12 @@ "supports_max_reasoning_effort": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "search_context_cost_per_query": { "search_context_size_high": 0.01, "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, - "supports_minimal_reasoning_effort": true + "supports_output_config": true }, "duckduckgo/search": { "litellm_provider": "duckduckgo", @@ -40038,7 +41434,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, "supports_pdf_input": true }, @@ -40061,7 +41456,6 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "tool_use_system_prompt_tokens": 346, "supports_native_structured_output": true, "supports_pdf_input": true } diff --git a/litellm/passthrough/utils.py b/litellm/passthrough/utils.py index d39a0dda152..9484922833a 100644 --- a/litellm/passthrough/utils.py +++ b/litellm/passthrough/utils.py @@ -71,6 +71,11 @@ class BasePassthroughUtils: request_headers.pop("content-length", None) request_headers.pop("host", None) + custom_header_names = {header_name.lower() for header_name in headers} + for header_name in list(request_headers.keys()): + if header_name.lower() in custom_header_names: + request_headers.pop(header_name, None) + # Combine request headers with custom headers headers = {**request_headers, **headers} diff --git a/litellm/proxy/_experimental/mcp_server/CLAUDE.md b/litellm/proxy/_experimental/mcp_server/CLAUDE.md new file mode 100644 index 00000000000..0ba8f73315f --- /dev/null +++ b/litellm/proxy/_experimental/mcp_server/CLAUDE.md @@ -0,0 +1 @@ +MCP note: **`available_on_public_internet: false` with `delegate_auth_to_upstream: true` (oauth2, interactive - not `client_credentials`)** - LiteLLM still allows the anonymous upstream PKCE path (no proxy API key for `/authorize` and matching MCP routes). The internal-only flag mainly affects other surfaces (e.g. IP-based discovery). Rely on the upstream IdP and network policy; the dashboard shows a warning when both are set, and the proxy logs a warning when the server is loaded from config or the database diff --git a/litellm/proxy/_experimental/mcp_server/auth/litellm_auth_handler.py b/litellm/proxy/_experimental/mcp_server/auth/litellm_auth_handler.py index 75b75d3ba44..7122c64ec64 100644 --- a/litellm/proxy/_experimental/mcp_server/auth/litellm_auth_handler.py +++ b/litellm/proxy/_experimental/mcp_server/auth/litellm_auth_handler.py @@ -20,7 +20,7 @@ class MCPAuthenticatedUser(AuthenticatedUser): def __init__( self, - user_api_key_auth: UserAPIKeyAuth, + user_api_key_auth: Optional[UserAPIKeyAuth], mcp_auth_header: Optional[str] = None, mcp_servers: Optional[List[str]] = None, mcp_server_auth_headers: Optional[Dict[str, Dict[str, str]]] = None, diff --git a/litellm/proxy/_experimental/mcp_server/auth/user_api_key_auth_mcp.py b/litellm/proxy/_experimental/mcp_server/auth/user_api_key_auth_mcp.py index a05af66118c..2aacab80f57 100644 --- a/litellm/proxy/_experimental/mcp_server/auth/user_api_key_auth_mcp.py +++ b/litellm/proxy/_experimental/mcp_server/auth/user_api_key_auth_mcp.py @@ -1,3 +1,4 @@ +import re from typing import Dict, List, Optional, Set, Tuple, cast from fastapi import HTTPException @@ -117,10 +118,32 @@ class MCPRequestHandler: return b"{}" request.body = mock_body # type: ignore + # Inline import — auth_utils participates in a proxy import cycle. + from litellm.proxy.auth.auth_utils import ( # noqa: PLC0415 + get_request_route, + ) + + request_route = get_request_route(request) # Only OAuth metadata routes registered under /.well-known/ are public. - # Match on request.url.path (path-only, exact prefix) so the substring - # cannot be smuggled via query string, hostname, or a deeper URL segment. - if request.url.path.startswith("/.well-known/"): + if request_route.startswith("/.well-known/"): + validated_user_api_key_auth = UserAPIKeyAuth() + elif ( + not litellm_api_key + and MCPRequestHandler._target_servers_delegate_auth_to_upstream( # noqa: E501 + path=request_route, mcp_servers=mcp_servers + ) + ): + # Operator opted this oauth2 server into upstream-delegated auth + # (PKCE passthrough): skip LiteLLM API-key/SSO entirely so the + # client authenticates directly with the upstream MCP server. + # Fires ONLY when neither x-litellm-api-key nor Authorization is + # present. If any LiteLLM key is supplied (primary or secondary + # header), we fall through so user_id is resolved, spend/rate + # limiting apply, and any stored OAuth token can be retrieved + # and forwarded upstream. Gated by + # _target_servers_delegate_auth_to_upstream, which only returns + # True when EVERY target is auth_type=oauth2 AND has the + # delegate_auth_to_upstream flag set — fails closed otherwise. validated_user_api_key_auth = UserAPIKeyAuth() elif has_explicit_litellm_key: # Explicit x-litellm-api-key provided - always validate normally @@ -155,7 +178,7 @@ class MCPRequestHandler: "401", "403", ) and MCPRequestHandler._target_servers_use_oauth2( - path=request.url.path, mcp_servers=mcp_servers + path=request_route, mcp_servers=mcp_servers ): verbose_logger.debug( "MCP OAuth2: target server is OAuth2-mode, treating " @@ -181,23 +204,62 @@ class MCPRequestHandler: @staticmethod def _extract_target_server_names_from_path(path: str) -> List[str]: """ - Extract the target MCP server name from the standard MCP transport - URL patterns: ``/mcp/{server_name}[/...]`` and + Extract the target MCP server name(s) from the standard MCP transport + URL patterns: ``/mcp/{server_name_or_csv}[/...]`` and ``/{server_name}/mcp[/...]``. Returns ``[]`` for any other path so callers fail closed when the target cannot be resolved. + Mirrors the regex-based parser in ``server.py::_get_mcp_servers_in_path`` + so the names used for auth gating match the names used for downstream + filtering. Without this alignment, an attacker could craft + ``/mcp//`` so that auth treats the request + as targeting the delegate server (bypassing LiteLLM auth) while + downstream filtering sees a different (non-existent) target and falls + back to the caller's full allowed-server set. + REST/admin endpoints, OAuth2 server endpoints (``/{server_name}/authorize``, ``/token`` etc.), and ``.well-known`` discovery routes intentionally fall through — those flows do not need OAuth2 token passthrough. Clients aggregating multiple servers should - use ``x-mcp-servers``, which takes precedence over path parsing. + use ``x-mcp-servers`` on a path that does not encode a target. """ + # ``/{server_name}/mcp[/...]`` form — single server. The literal + # ``mcp`` must be the second segment (not the first, which would be + # the ``/mcp/...`` form handled below). This branch must stay in sync + # with ``server.py::_get_mcp_servers_in_path``, which also accepts the + # un-rewritten form (some entry points may skip the + # ``dynamic_mcp_route`` rewrite). segments = [s for s in path.split("/") if s] - if len(segments) >= 2 and segments[0] == "mcp": - return [segments[1]] - if len(segments) >= 2 and segments[1] == "mcp": + if len(segments) >= 2 and segments[1] == "mcp" and segments[0] != "mcp": return [segments[0]] - return [] + + # ``/mcp/...`` form — server name(s) may contain a slash (e.g. + # ``custom_solutions/user_123``) and may be a comma-separated list. + # Use the same parsing logic as ``_get_mcp_servers_in_path`` so the + # parsed names match downstream routing. + mcp_path_match = re.match(r"^/mcp/([^?#]+)(?:\?.*)?(?:#.*)?$", path) + if not mcp_path_match: + return [] + servers_and_path = mcp_path_match.group(1) + if not servers_and_path: + return [] + + if "," in servers_and_path: + # Comma-separated servers, possibly followed by a trailing path. + path_match = re.search(r"/([^/,]+(?:/[^/,]+)*)$", servers_and_path) + if path_match: + servers_part = servers_and_path[: -(len(path_match.group(1)) + 1)] + else: + servers_part = servers_and_path + return [s.strip() for s in servers_part.split(",") if s.strip()] + + # Single-server case — server name may contain at most one slash. + single_server_match = re.match( + r"^([^/]+(?:/[^/]+)?)(?:/.*)?$", servers_and_path + ) + if single_server_match: + return [single_server_match.group(1)] + return [servers_and_path] @staticmethod def _target_servers_use_oauth2(path: str, mcp_servers: Optional[List[str]]) -> bool: @@ -217,13 +279,13 @@ class MCPRequestHandler: ) from litellm.types.mcp import MCPAuth - # Use the x-mcp-servers header verbatim when present (including the - # explicitly-empty list, which means "no targets" → fail closed). - # Only fall back to path parsing when the header was absent entirely. - target_names = ( - mcp_servers - if mcp_servers is not None - else MCPRequestHandler._extract_target_server_names_from_path(path) + # Resolve the same target list downstream routing will use. For + # ``/mcp/...`` routes, ``extract_mcp_auth_context`` overrides the + # ``x-mcp-servers`` header with path-derived names, so we must mirror + # that here — otherwise a caller could set the header to a permissive + # server while the path targets a stricter one (header/path TOCTOU). + target_names = MCPRequestHandler._resolve_target_server_names( + path=path, mcp_servers_header=mcp_servers ) if not target_names: return False @@ -234,6 +296,76 @@ class MCPRequestHandler: return False return True + @staticmethod + def _target_servers_delegate_auth_to_upstream( + path: str, mcp_servers: Optional[List[str]] + ) -> bool: + """ + True only when EVERY MCP server the request targets is configured for + ``auth_type == oauth2`` AND has ``delegate_auth_to_upstream=True``. + Fails closed when any target does not opt in or cannot be resolved. + + Used by :meth:`process_mcp_request` to skip LiteLLM API-key/SSO auth + entirely (PKCE passthrough) so the client authenticates directly with + the upstream MCP server. Mixed-target requests (e.g. one delegated + + one non-delegated server) fall back to normal LiteLLM auth. + """ + # Inline imports avoid a circular dependency: mcp_server_manager imports + # from this module. + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) + from litellm.types.mcp import MCPAuth + + # See _target_servers_use_oauth2: must mirror the downstream + # header-vs-path override or an attacker could set + # ``x-mcp-servers`` to a delegate-enabled server while the URL path + # targets a non-delegate server, skipping LiteLLM auth for it. + target_names = MCPRequestHandler._resolve_target_server_names( + path=path, mcp_servers_header=mcp_servers + ) + if not target_names: + return False + + for name in target_names: + server = global_mcp_server_manager.get_mcp_server_by_name(name) + if server is None or server.auth_type != MCPAuth.oauth2: + return False + # `is True` is intentional: opt-in must be an explicit boolean + # True. A MagicMock attribute (in tests) or any other truthy + # non-bool must not silently enable the bypass. + if getattr(server, "delegate_auth_to_upstream", False) is not True: + return False + # Never delegate for M2M (client_credentials) servers: LiteLLM + # fetches the upstream token automatically using stored credentials, + # so allowing anonymous bypass would let any external caller invoke + # tools authenticated as LiteLLM's service account. + if server.has_client_credentials: + return False + return True + + @staticmethod + def _resolve_target_server_names( + path: str, mcp_servers_header: Optional[List[str]] + ) -> List[str]: + """ + Resolve the target MCP server names exactly as downstream routing + does (``server.py::extract_mcp_auth_context``). + + For ``/mcp/...`` paths, downstream routing **overrides** any + ``x-mcp-servers`` header value with the path-derived names. Mirror + that here so an attacker cannot use a permissive header value to + flip an auth gate while the path targets a stricter server + (header/path TOCTOU). For non-``/mcp/...`` paths (where the path + does not encode targets), fall back to the header. + """ + path_targets = MCPRequestHandler._extract_target_server_names_from_path(path) + if path_targets: + return path_targets + # Path did not resolve to /mcp/... targets — trust the header + # (including an explicitly empty list, which means "no targets"). + return mcp_servers_header if mcp_servers_header is not None else [] + @staticmethod def _get_mcp_auth_header_from_headers(headers: Headers) -> Optional[str]: """ @@ -434,25 +566,33 @@ class MCPRequestHandler: ) ) + key_access_group_grants = ( + await MCPRequestHandler._get_key_access_group_mcp_server_extras( + user_api_key_auth + ) + ) + ######################################################### # Calculate key/team allowed servers using inheritance and intersection logic ######################################################### - allowed_mcp_servers: List[str] = [] - has_lower_level_mcp_restrictions = ( - len(allowed_mcp_servers_for_key) > 0 - or len(allowed_mcp_servers_for_team) > 0 - ) - if len(allowed_mcp_servers_for_team) > 0: - if len(allowed_mcp_servers_for_key) > 0: - # Key has its own MCP permissions - use intersection with team permissions - for _mcp_server in allowed_mcp_servers_for_key: - if _mcp_server in allowed_mcp_servers_for_team: - allowed_mcp_servers.append(_mcp_server) - else: - # Key has no MCP permissions - inherit from team - allowed_mcp_servers = allowed_mcp_servers_for_team + key_set = set(allowed_mcp_servers_for_key) + team_set = set(allowed_mcp_servers_for_team) + grants_set = set(key_access_group_grants) + + has_lower_level_mcp_restrictions = bool(key_set or team_set or grants_set) + + # 1. Key/team ceiling. An empty set means "this level does not restrict". + if not team_set: + base = key_set # no team restriction + elif not key_set: + base = team_set # key has no own perms → inherits team else: - allowed_mcp_servers = allowed_mcp_servers_for_key + base = key_set & team_set # both restrict → intersect + + # 2. Add the key's access-group grants on top. These are additive: + # attaching a group to the key grants its servers regardless of the + # team ceiling. + allowed_mcp_servers: List[str] = list(base | grants_set) ######################################################### # Check end_user permissions if end_user_id is set @@ -745,43 +885,98 @@ class MCPRequestHandler: return True return False + @staticmethod + async def _get_key_access_group_mcp_server_extras( + user_api_key_auth: Optional[UserAPIKeyAuth] = None, + ) -> List[str]: + """ + Resolve the key's unified `access_group_ids` (LiteLLM_AccessGroupTable) to + MCP server IDs as additive grants: a group attached to the key extends the + key's allowed servers on top of the key/team ceiling rather than being + capped by the team. Attaching the group to the key is itself the grant — + no `assigned_key_ids` / `assigned_team_ids` re-check. Tag-style + `mcp_access_groups` (per-server tags) live in the key's object_permission + scope, not here. + """ + if user_api_key_auth is None: + return [] + try: + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) + from litellm.proxy.auth.auth_checks import ( + _get_mcp_server_ids_from_access_groups, + ) + from litellm.proxy.proxy_server import ( + prisma_client, + proxy_logging_obj, + user_api_key_cache, + ) + + raw_server_ids = await _get_mcp_server_ids_from_access_groups( + access_group_ids=user_api_key_auth.access_group_ids or [], + prisma_client=prisma_client, + user_api_key_cache=user_api_key_cache, + proxy_logging_obj=proxy_logging_obj, + ) + if not raw_server_ids: + return [] + # Permission entries may be server_ids OR names/aliases — expand to ids. + return global_mcp_server_manager.expand_permission_list(raw_server_ids) + except Exception as e: + verbose_logger.warning( + f"Failed to get key access group MCP server grants: {str(e)}" + ) + return [] + @staticmethod async def _get_allowed_mcp_servers_for_key( user_api_key_auth: Optional[UserAPIKeyAuth] = None, ) -> List[str]: + """ + Get the key's own MCP ceiling from its object_permission + (mcp_servers, tag-style mcp_access_groups, mcp_tool_permissions). + + Unified key.access_group_ids are NOT resolved here — they are additive + grants handled by _get_key_access_group_mcp_server_extras and unioned on + top of the key/team ceiling, so they must not enter this scope (which is + intersected against the team). + """ + if user_api_key_auth is None: + return [] try: + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) + from litellm.proxy.auth.auth_checks import ( + get_object_permission, + ) + from litellm.proxy.proxy_server import ( + prisma_client, + proxy_logging_obj, + user_api_key_cache, + ) + # Get key object permission (already loaded in main auth flow, or fetch from DB) key_object_permission = MCPRequestHandler._get_key_object_permission( user_api_key_auth ) if ( key_object_permission is None - and user_api_key_auth and user_api_key_auth.object_permission_id + and prisma_client is not None ): - from litellm.proxy.auth.auth_checks import get_object_permission - from litellm.proxy.proxy_server import ( - prisma_client, - proxy_logging_obj, - user_api_key_cache, + key_object_permission = await get_object_permission( + object_permission_id=user_api_key_auth.object_permission_id, + prisma_client=prisma_client, + user_api_key_cache=user_api_key_cache, + parent_otel_span=user_api_key_auth.parent_otel_span, + proxy_logging_obj=proxy_logging_obj, ) - - if prisma_client is not None: - key_object_permission = await get_object_permission( - object_permission_id=user_api_key_auth.object_permission_id, - prisma_client=prisma_client, - user_api_key_cache=user_api_key_cache, - parent_otel_span=user_api_key_auth.parent_otel_span, - proxy_logging_obj=proxy_logging_obj, - ) if key_object_permission is None: return [] # Permission entries may be server_ids OR names/aliases — expand to ids. - from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( - global_mcp_server_manager, - ) - direct_mcp_servers = global_mcp_server_manager.expand_permission_list( key_object_permission.mcp_servers or [] ) @@ -816,42 +1011,78 @@ class MCPRequestHandler: """ Get allowed MCP servers for a team. - Note: object_permission is automatically loaded by get_team_object() in main auth flow. + Unions two sources: + - Legacy team.object_permission (mcp_servers, mcp_access_groups, + mcp_tool_permissions). + - Unified team.access_group_ids → access_group.access_mcp_server_ids. + Mirrors the model-side pattern in can_team_access_model — the group + is already attached to the team, so the team relationship is itself + the gate (no assigned_team_ids check needed here). """ try: - # Get team object permission (already loaded in main auth flow) - object_permissions = await MCPRequestHandler._get_team_object_permission( - user_api_key_auth - ) - - if object_permissions is None: - return [] - - # Permission entries may be server_ids OR names/aliases — expand to ids. from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( global_mcp_server_manager, ) + from litellm.proxy.auth.auth_checks import ( + _get_mcp_server_ids_from_access_groups, + get_team_object, + ) + from litellm.proxy.proxy_server import ( + prisma_client, + proxy_logging_obj, + user_api_key_cache, + ) + + if ( + user_api_key_auth is None + or not user_api_key_auth.team_id + or prisma_client is None + ): + return [] + + team_obj: Optional[LiteLLM_TeamTable] = await get_team_object( + team_id=user_api_key_auth.team_id, + prisma_client=prisma_client, + user_api_key_cache=user_api_key_cache, + parent_otel_span=user_api_key_auth.parent_otel_span, + proxy_logging_obj=proxy_logging_obj, + ) + if team_obj is None: + return [] + + team_access_group_servers = await _get_mcp_server_ids_from_access_groups( + access_group_ids=team_obj.access_group_ids or [], + prisma_client=prisma_client, + user_api_key_cache=user_api_key_cache, + proxy_logging_obj=proxy_logging_obj, + ) + + object_permissions = team_obj.object_permission + if object_permissions is None: + return list(set(team_access_group_servers)) direct_mcp_servers = global_mcp_server_manager.expand_permission_list( object_permissions.mcp_servers or [] ) - # Get MCP servers from access groups - access_group_servers = ( + legacy_access_group_servers = ( await MCPRequestHandler._get_mcp_servers_from_access_groups( object_permissions.mcp_access_groups or [] ) ) - # servers referenced in tool permissions should also be accessible tool_perm_servers = list( global_mcp_server_manager.expand_tool_permissions( object_permissions.mcp_tool_permissions ).keys() ) - # Combine all lists - all_servers = direct_mcp_servers + access_group_servers + tool_perm_servers + all_servers = ( + direct_mcp_servers + + legacy_access_group_servers + + tool_perm_servers + + team_access_group_servers + ) return list(set(all_servers)) except Exception as e: verbose_logger.warning( diff --git a/litellm/proxy/_experimental/mcp_server/db.py b/litellm/proxy/_experimental/mcp_server/db.py index a6f0d145e9b..e30667776c1 100644 --- a/litellm/proxy/_experimental/mcp_server/db.py +++ b/litellm/proxy/_experimental/mcp_server/db.py @@ -30,6 +30,8 @@ from litellm.types.mcp import MCPCredentials def _prepare_mcp_server_data( data: Union[NewMCPServerRequest, UpdateMCPServerRequest], + exclude_unset: bool = False, + fields_set: Optional[Set[str]] = None, ) -> Dict[str, Any]: """ Helper function to prepare MCP server data for database operations. @@ -37,17 +39,39 @@ def _prepare_mcp_server_data( Args: data: NewMCPServerRequest or UpdateMCPServerRequest object + exclude_unset: When True, only fields the caller explicitly provided are + included. Used for partial updates (PUT /v1/mcp/server) so omitted + fields keep their existing DB value instead of being silently reset + to a Pydantic schema default. ``exclude_none`` is not enough here: + non-Optional fields (e.g. ``transport=MCPTransport.sse``, + ``mcp_access_groups=[]``, ``allow_all_keys=False``) are backfilled + with their default when omitted, and a non-None default survives the + ``exclude_none`` filter and overwrites the row. Returns: Dict with properly serialized JSON fields """ from litellm.litellm_core_utils.safe_json_dumps import safe_dumps - # Convert model to dict - data_dict = data.model_dump(exclude_none=True) - # Ensure alias is always present in the dict (even if None) - if "alias" not in data_dict: - data_dict["alias"] = getattr(data, "alias", None) + # Convert model to dict. + # - Partial update (exclude_unset): only caller-provided keys are emitted, so + # omitted fields are never written and keep their existing DB value. + # - Create (exclude_none): drop None-valued fields and let DB defaults apply. + if exclude_unset: + if fields_set is None: + fields_set = data.fields_set() + data_dict = data.model_dump(exclude_unset=True) + # ``validate_and_normalize_mcp_server_payload`` always assigns ``alias`` + # on the payload, which marks it as set even when the caller omitted it. + # Drop it only when the original request omitted alias; an explicit + # ``alias=None`` is a valid request to clear the stored alias. + if data_dict.get("alias") is None and "alias" not in fields_set: + data_dict.pop("alias", None) + else: + data_dict = data.model_dump(exclude_none=True) + # Ensure alias is always present in the dict (even if None) + if "alias" not in data_dict: + data_dict["alias"] = getattr(data, "alias", None) # Handle credentials serialization credentials = data_dict.get("credentials") @@ -57,33 +81,33 @@ def _prepare_mcp_server_data( ) data_dict["credentials"] = safe_dumps(data_dict["credentials"]) - # Handle static_headers serialization - if data.static_headers is not None: - data_dict["static_headers"] = safe_dumps(data.static_headers) + # Serialize JSON fields from ``data_dict`` (not ``data``) so the + # exclude_unset filter is respected. Reading back from ``data`` would + # reintroduce defaults (e.g. ``env={}``) for fields the caller never set. + if data_dict.get("static_headers") is not None: + data_dict["static_headers"] = safe_dumps(data_dict["static_headers"]) - # Handle mcp_info serialization - if data.mcp_info is not None: - data_dict["mcp_info"] = safe_dumps(data.mcp_info) + if data_dict.get("mcp_info") is not None: + data_dict["mcp_info"] = safe_dumps(data_dict["mcp_info"]) - # Handle env serialization - if data.env is not None: - data_dict["env"] = safe_dumps(data.env) + if data_dict.get("env") is not None: + data_dict["env"] = safe_dumps(data_dict["env"]) - # Handle tool name override serialization - if data.tool_name_to_display_name is not None: + if data_dict.get("tool_name_to_display_name") is not None: data_dict["tool_name_to_display_name"] = safe_dumps( - data.tool_name_to_display_name + data_dict["tool_name_to_display_name"] ) - if data.tool_name_to_description is not None: + if data_dict.get("tool_name_to_description") is not None: data_dict["tool_name_to_description"] = safe_dumps( - data.tool_name_to_description + data_dict["tool_name_to_description"] ) # mcp_access_groups is already List[str], no serialization needed - # Force include is_byok even when False (exclude_none=True would not drop it, - # but be explicit to ensure a False value is always written to the DB). - data_dict["is_byok"] = getattr(data, "is_byok", False) + # On create, force is_byok so a False value is always written to the DB. On + # partial update, only write it when the caller explicitly provided it. + if not exclude_unset: + data_dict["is_byok"] = getattr(data, "is_byok", False) return data_dict @@ -398,7 +422,10 @@ async def create_mcp_server( async def update_mcp_server( - prisma_client: PrismaClient, data: UpdateMCPServerRequest, touched_by: str + prisma_client: PrismaClient, + data: UpdateMCPServerRequest, + touched_by: str, + fields_set: Optional[Set[str]] = None, ) -> LiteLLM_MCPServerTable: """ Update a new mcp server record in the db @@ -407,8 +434,13 @@ async def update_mcp_server( from litellm.litellm_core_utils.safe_json_dumps import safe_dumps - # Use helper to prepare data with proper JSON serialization - data_dict = _prepare_mcp_server_data(data) + # Use helper to prepare data with proper JSON serialization. + # exclude_unset=True makes this a true partial update: fields the caller did + # not provide are not written, so they keep their existing DB value instead + # of being reset to a schema default (transport=sse, allow_all_keys=False...). + data_dict = _prepare_mcp_server_data( + data, exclude_unset=True, fields_set=fields_set + ) # Pre-fetch existing record once if we need it for auth_type or credential logic existing = None diff --git a/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py b/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py index 62691641234..8324ba641a4 100644 --- a/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py +++ b/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py @@ -1,3 +1,4 @@ +import html as _html import json from typing import Any, Dict, Optional from urllib.parse import parse_qsl, urlencode, urlparse, urlunparse @@ -86,7 +87,9 @@ def decode_state_hash(encrypted_state: str) -> dict: def _get_validated_client_redirect_uri( request: Request, state_data: Dict[str, Any] ) -> str: - """Return a trusted (same-origin or loopback) client redirect URI from OAuth state.""" + """Return a trusted (same-origin, loopback, or ops-allowlisted) + client redirect URI from OAuth state. + """ redirect_uri = state_data.get("client_redirect_uri") or state_data.get("base_url") if not redirect_uri or not isinstance(redirect_uri, str): raise HTTPException(status_code=400, detail="Invalid redirect URI") @@ -296,11 +299,10 @@ async def authorize_with_server( status_code=400, detail="MCP server authorization url is not set" ) - # Loopback OR same-origin redirect_uri. The URI is encrypted into the - # OAuth state and decoded on /callback to redirect the user back; - # restricting to trusted origins blocks the open-redirect + - # code-theft primitive (VERIA-57 root cause B). Loopback supports - # native MCP clients; same-origin supports the proxy's own UI callback. + # Trusted redirect_uri: same-origin, loopback, or ops-allowlisted. + # The URI is encrypted into the OAuth state and decoded on + # /callback to redirect the user back; a non-trusted URI would be + # an open-redirect + code-theft primitive (VERIA-57 root cause B). validate_trusted_redirect_uri(request, redirect_uri) parsed = urlparse(redirect_uri) base_url = urlunparse(parsed._replace(query="")) @@ -617,18 +619,115 @@ async def token_endpoint( ) +# Per RFC 6749 §4.1.2.1, an IdP that rejects an OAuth authorization request +# redirects back to the configured redirect URI with ``error`` / +# ``error_description`` / ``error_uri`` query params and no ``code``. The MCP +# loopback flow funnels that response through this /callback endpoint, so +# the endpoint must accept either a successful (``code``+``state``) or an +# error response. Declaring ``code``/``state`` as required would cause +# FastAPI to reject the error response with a 422 before the handler runs, +# which strands the MCP client waiting on the loopback (see LIT-2750). + + +def _render_oauth_error_html(error: str, description: Optional[str]) -> HTMLResponse: + """Render an actionable HTML page for an IdP-reported OAuth error. + + Used when we cannot propagate the error back to the registered + ``redirect_uri`` (state missing or undecryptable). Returned with a 400 + status so the failure is observable to operators while still being a + human-readable page for the end user. + """ + safe_error = _html.escape(error or "unknown_error") + safe_description = _html.escape(description) if description else "" + description_html = f"

{safe_description}

" if safe_description else "" + body = ( + "" + "

Authentication failed

" + f"

Error: {safe_error}

" + f"{description_html}" + "

You can close this window and try again.

" + "" + ) + return HTMLResponse(body, status_code=400) + + @router.get("/callback") -async def callback(request: Request, code: str, state: str): +async def callback( + request: Request, + code: Optional[str] = None, + state: Optional[str] = None, + error: Optional[str] = None, + error_description: Optional[str] = None, + error_uri: Optional[str] = None, +): + """OAuth 2.0 authorization response handler for MCP loopback clients. + + Accepts either: + + - A successful authorization response (``code`` + ``state``), which is + forwarded back to the validated client ``redirect_uri`` with the + original (un-wrapped) ``state``. + - An error response (``error``[+``error_description``/``error_uri``]), per + RFC 6749 §4.1.2.1. When ``state`` is present and decodes to a trusted + ``redirect_uri``, the error params are propagated back to the client so + its OAuth library can surface them. Otherwise we render an HTML error + page so the user is not left on an opaque 422 / blank screen. + """ + # 1. IdP-reported error path (e.g. ``?error=access_denied``). + if error: + verbose_logger.info( + "MCP /callback received IdP error: error=%s, error_description=%s", + error, + error_description, + ) + if state: + try: + state_data = decode_state_hash(state) + original_state = state_data.get("original_state") + redirect_uri = _get_validated_client_redirect_uri(request, state_data) + except HTTPException: + # Untrusted/invalid client redirect_uri — surface inline rather + # than blindly forwarding the error to an attacker-controlled URL. + return _render_oauth_error_html(error, error_description) + except Exception: + # State could not be decrypted (expired key, tampered, etc.). + return _render_oauth_error_html(error, error_description) + + params: Dict[str, str] = {"error": error} + if error_description: + params["error_description"] = error_description + if error_uri: + params["error_uri"] = error_uri + if original_state is not None: + params["state"] = original_state + complete_returned_url = _append_query_params(redirect_uri, params) + return RedirectResponse(url=complete_returned_url, status_code=302) + + # No state — nothing to round-trip to. Show the user the error. + return _render_oauth_error_html(error, error_description) + + # 2. Neither success nor error parameters present — most likely a stray + # GET / dropped SSO redirect chain. Surface a 400 instead of 422. + if not code or not state: + missing = [ + name for name, value in (("code", code), ("state", state)) if not value + ] + return _render_oauth_error_html( + "invalid_request", + f"Missing authorization {' and '.join(repr(m) for m in missing)} parameter(s).", + ) + + # 3. Successful authorization response. try: state_data = decode_state_hash(state) original_state = state_data["original_state"] - # Re-validate at the sink. /authorize rejects untrusted - # redirect_uri before encoding into state, but encrypted states - # minted before that check was added have no expiry and remain - # valid indefinitely. Validating here (same-origin OR loopback) - # blocks the open-redirect + code-theft primitive even for pre-fix - # states while allowing the UI's same-origin callback to work. + # Re-validate the client redirect URI at the sink. /authorize + # rejects untrusted URIs before encoding them into state, but + # encrypted states minted before that check was added have no + # expiry and remain valid indefinitely. Validating here blocks + # the open-redirect + code-theft primitive even for pre-fix + # states while permitting same-origin / allowlisted clients. redirect_uri = _get_validated_client_redirect_uri(request, state_data) params = {"code": code, "state": original_state} diff --git a/litellm/proxy/_experimental/mcp_server/mcp_server_manager.py b/litellm/proxy/_experimental/mcp_server/mcp_server_manager.py index 4901bc76d2f..f35aa30a7c9 100644 --- a/litellm/proxy/_experimental/mcp_server/mcp_server_manager.py +++ b/litellm/proxy/_experimental/mcp_server/mcp_server_manager.py @@ -12,6 +12,7 @@ import hashlib import json import os import re +import time from typing import Any, Callable, Dict, List, Literal, Optional, Set, Tuple, Union, cast from urllib.parse import urlparse @@ -145,6 +146,30 @@ def _warn_on_server_name_fields( _warn("server_name", server_name) +def _warn_internal_delegate_pkce_if_applicable( + server: MCPServer, *, source: str +) -> None: + """Surface internal + upstream PKCE delegate in logs for operators.""" + if server.auth_type != MCPAuth.oauth2: + return + if getattr(server, "delegate_auth_to_upstream", False) is not True: + return + if getattr(server, "available_on_public_internet", True): + return + if server.has_client_credentials: + return + label = get_server_prefix(server) + verbose_logger.warning( + "MCP server %r (id=%s, source=%s): internal-only (available_on_public_internet=false) " + "with delegate_auth_to_upstream=true. Anonymous callers can reach the upstream OAuth2 " + "/authorize flow and complete PKCE without a LiteLLM API key session; ensure the " + "upstream IdP and network enforce your access policy.", + label, + server.server_id, + source, + ) + + def _deserialize_json_dict(data: Any) -> Optional[Dict[str, str]]: """ Deserialize optional JSON mappings stored in the database. @@ -226,6 +251,10 @@ class MCPServerManager: } """ self._upstream_initialize_instructions_by_server_id: Dict[str, str] = {} + # Per-server monotonic timestamp of last upstream prefetch attempt (success, + # empty result, or failure). Used to throttle re-probes for servers that do + # not return instructions, and to apply a short cooldown after failures. + self._upstream_initialize_instructions_probed_at: Dict[str, float] = {} def _remember_upstream_initialize_instructions( self, server: MCPServer, client: MCPClient @@ -236,6 +265,80 @@ class MCPServerManager: raw ).strip() + async def _ensure_upstream_initialize_instructions_cached( + self, server: MCPServer + ) -> None: + """ + Open one upstream session and cache InitializeResult.instructions if missing. + + No-op when: + - YAML/DB instructions are set on the server record, + - server is OpenAPI (spec_path), + - non-empty upstream instructions are already cached, + - auth preconditions match health_check_server's skip rules + (per-user auth / missing static auth token), + - a prior probe attempt for this server is within + MCP_HEALTH_CHECK_TIMEOUT seconds (the probe is a health-check-shaped + op and already uses this knob for its inner call timeout; reusing it + as the cooldown avoids reconnecting on every gateway initialize when + upstream returns empty or fails). + """ + if server.spec_path: + return + if server.instructions and server.instructions.strip(): + return + if self._upstream_initialize_instructions_by_server_id.get(server.server_id): + return + if server.requires_per_user_auth: + return + if ( + server.auth_type + and server.auth_type != MCPAuth.none + and server.auth_type != MCPAuth.aws_sigv4 + and not server.authentication_token + ): + return + + last_probed_at = self._upstream_initialize_instructions_probed_at.get( + server.server_id + ) + if ( + last_probed_at is not None + and (time.monotonic() - last_probed_at) < MCP_HEALTH_CHECK_TIMEOUT + ): + return + + # Record the attempt up-front so that a failure / empty response does not + # cause every subsequent initialize request to re-open the upstream session. + self._upstream_initialize_instructions_probed_at[server.server_id] = ( + time.monotonic() + ) + + try: + extra_headers: Optional[Dict[str, str]] = ( + dict(server.static_headers) if server.static_headers else None + ) + client = await self._create_mcp_client( + server=server, + mcp_auth_header=None, + extra_headers=extra_headers, + stdio_env=None, + ) + + async def _noop(_session): + return "ok" + + await asyncio.wait_for( + client.run_with_session(_noop), timeout=MCP_HEALTH_CHECK_TIMEOUT + ) + self._remember_upstream_initialize_instructions(server, client) + except Exception as e: + verbose_logger.debug( + "Upstream initialize instructions prefetch failed for %s: %s", + server.name, + e, + ) + def get_registry(self) -> Dict[str, MCPServer]: """ Get the registered MCP Servers from the registry and union with the config MCP Servers @@ -256,6 +359,7 @@ class MCPServerManager: """ verbose_logger.debug("Loading MCP Servers from config-----") self._upstream_initialize_instructions_by_server_id.clear() + self._upstream_initialize_instructions_probed_at.clear() # Track which aliases have been used to ensure only first occurrence is used used_aliases = set() @@ -297,32 +401,6 @@ class MCPServerManager: )() name_for_prefix = get_server_prefix(temp_server) - # Use alias for name if present, else server_name - alias = server_config.get("alias", None) - - # Apply mcp_aliases mapping if provided - if mcp_aliases and alias is None: - # Check if this server_name has an alias in mcp_aliases - for alias_name, target_server_name in mcp_aliases.items(): - if ( - target_server_name == server_name - and alias_name not in used_aliases - ): - alias = alias_name - used_aliases.add(alias_name) - verbose_logger.debug( - f"Mapped alias '{alias_name}' to server '{server_name}'" - ) - break - - # Create a temporary server object to use with get_server_prefix utility - temp_server = type( - "TempServer", - (), - {"alias": alias, "server_name": server_name, "server_id": None}, - )() - name_for_prefix = get_server_prefix(temp_server) - server_url = server_config.get("url", None) or "" # Generate stable server ID based on parameters server_id = self._generate_stable_server_id( @@ -402,6 +480,9 @@ class MCPServerManager: available_on_public_internet=bool( server_config.get("available_on_public_internet", True) ), + delegate_auth_to_upstream=bool( + server_config.get("delegate_auth_to_upstream", False) + ), # AWS SigV4 fields aws_access_key_id=server_config.get("aws_access_key_id", None), aws_secret_access_key=server_config.get("aws_secret_access_key", None), @@ -422,6 +503,7 @@ class MCPServerManager: ), ) self._assign_unique_short_prefix(new_server) + _warn_internal_delegate_pkce_if_applicable(new_server, source="config") self.config_mcp_servers[server_id] = new_server # Check if this is an OpenAPI-based server @@ -796,6 +878,9 @@ class MCPServerManager: available_on_public_internet=bool( getattr(mcp_server, "available_on_public_internet", True) ), + delegate_auth_to_upstream=bool( + getattr(mcp_server, "delegate_auth_to_upstream", False) + ), created_at=getattr(mcp_server, "created_at", None), updated_at=getattr(mcp_server, "updated_at", None), tool_name_to_display_name=_deserialize_json_dict( @@ -807,6 +892,7 @@ class MCPServerManager: is_byok=bool(getattr(mcp_server, "is_byok", False)), byok_description=getattr(mcp_server, "byok_description", None) or [], byok_api_key_help_url=getattr(mcp_server, "byok_api_key_help_url", None), + source_url=getattr(mcp_server, "source_url", None), # AWS SigV4 fields aws_access_key_id=aws_creds.get("aws_access_key_id"), aws_secret_access_key=aws_creds.get("aws_secret_access_key"), @@ -828,6 +914,7 @@ class MCPServerManager: ) or "urn:ietf:params:oauth:token-type:access_token", ) + _warn_internal_delegate_pkce_if_applicable(new_server, source="database") return new_server async def _maybe_register_openapi_tools( @@ -967,6 +1054,31 @@ class MCPServerManager: if not in_toolset_scope: combined_servers.update(allow_all_server_ids) + # For anonymous callers (no user_id, no role), also surface any + # servers the operator has opted into upstream-delegated auth. + # These servers handle their own auth at the upstream level, so + # LiteLLM granting access here does not bypass any security gate. + is_anonymous = not ( + user_api_key_auth + and ( + getattr(user_api_key_auth, "user_id", None) + or getattr(user_api_key_auth, "user_role", None) + or getattr(user_api_key_auth, "api_key", None) + ) + ) + if is_anonymous: + delegate_server_ids = [ + server.server_id + for server in self.get_registry().values() + if getattr(server, "auth_type", None) == MCPAuth.oauth2 + and getattr(server, "delegate_auth_to_upstream", False) is True + # M2M servers must not be exposed anonymously: an + # unauthenticated caller would get LiteLLM to proxy tool + # calls using its stored client_credentials. + and not server.has_client_credentials + ] + combined_servers.update(delegate_server_ids) + if len(combined_servers) == 0: verbose_logger.debug( "No allowed MCP Servers found for user api key auth." @@ -1181,11 +1293,17 @@ class MCPServerManager: return [] # Get server-specific auth header if available - server_auth_header = None - if mcp_server_auth_headers and server.alias: - server_auth_header = mcp_server_auth_headers.get(server.alias) - elif mcp_server_auth_headers and server.server_name: - server_auth_header = mcp_server_auth_headers.get(server.server_name) + server_auth_header: Optional[Union[str, Dict[str, str]]] = None + if mcp_server_auth_headers: + from litellm.proxy._experimental.mcp_server.utils import ( + lookup_mcp_server_auth_in_headers, + ) + + server_auth_header = lookup_mcp_server_auth_in_headers( + mcp_server_auth_headers, + alias=server.alias, + server_name=server.server_name, + ) # Fall back to deprecated mcp_auth_header if no server-specific header found if server_auth_header is None: @@ -1195,6 +1313,7 @@ class MCPServerManager: tools = await self._get_tools_from_server( server=server, mcp_auth_header=server_auth_header, + user_api_key_auth=user_api_key_auth, ) return tools except Exception as e: @@ -1375,6 +1494,7 @@ class MCPServerManager: extra_headers: Optional[Dict[str, str]] = None, add_prefix: bool = True, raw_headers: Optional[Dict[str, str]] = None, + user_api_key_auth: Optional[UserAPIKeyAuth] = None, ) -> List[MCPTool]: """ Helper method to get tools from a single MCP server with prefixed names. @@ -1401,6 +1521,46 @@ class MCPServerManager: extra_headers = {} extra_headers.update(server.static_headers) + # MCPJWTSigner: inject signed JWT for tools/list (list path skips pre_call_hook). + # Skip entirely when the signer is not configured (avoid an unnecessary + # dict copy on every list call), when the server has its own static + # Authorization header, when a per-user mcp_auth_header has already + # been resolved, or when the caller already supplied an Authorization + # entry in extra_headers (e.g. a per-user OAuth token resolved + # upstream) — admin-configured static auth and per-user OAuth must + # take precedence so the signer doesn't silently overwrite e.g. an + # upstream API key or a user's OAuth token (MCPClient._get_auth_headers + # applies extra_headers after writing Authorization from auth_value, so + # an injected JWT would otherwise clobber the per-user token). + if user_api_key_auth is not None and not server.spec_path: + from litellm.proxy.guardrails.guardrail_hooks.mcp_jwt_signer.mcp_jwt_signer import ( + get_mcp_jwt_signer, + inject_mcp_jwt_headers_for_upstream, + ) + + static_headers = server.static_headers or {} + has_static_authorization = any( + isinstance(k, str) and k.lower() == "authorization" + for k in static_headers.keys() + ) + has_extra_authorization = bool(extra_headers) and any( + isinstance(k, str) and k.lower() == "authorization" + for k in (extra_headers or {}).keys() + ) + + if ( + get_mcp_jwt_signer() is not None + and not has_static_authorization + and not mcp_auth_header + and not has_extra_authorization + ): + extra_headers = await inject_mcp_jwt_headers_for_upstream( + user_api_key_dict=user_api_key_auth, + extra_headers=extra_headers, + raw_headers=raw_headers, + for_list_tools=True, + ) + stdio_env = self._build_stdio_env(server, raw_headers) client = await self._create_mcp_client( @@ -2634,16 +2794,15 @@ class MCPServerManager: server_auth_header: Optional[Union[Dict[str, str], str]] = None if mcp_server_auth_headers: # Normalize keys for case-insensitive lookup - normalized_headers = { - k.lower(): v for k, v in mcp_server_auth_headers.items() - } + from litellm.proxy._experimental.mcp_server.utils import ( + lookup_mcp_server_auth_in_headers, + ) - if mcp_server.alias: - server_auth_header = normalized_headers.get(mcp_server.alias.lower()) - if server_auth_header is None and mcp_server.server_name: - server_auth_header = normalized_headers.get( - mcp_server.server_name.lower() - ) + server_auth_header = lookup_mcp_server_auth_in_headers( + mcp_server_auth_headers, + alias=mcp_server.alias, + server_name=mcp_server.server_name, + ) # Fall back to deprecated mcp_auth_header if no server-specific header found if server_auth_header is None: @@ -2760,6 +2919,112 @@ class MCPServerManager: return cast(CallToolResult, result) + def _resolve_mcp_server_for_tool_call( + self, + server_name: str, + name: str, + ) -> MCPServer: + """Resolve MCP server for call_tool (prefixed name, registry, fallback).""" + prefixed_tool_name = add_server_prefix_to_name(name, server_name) + mcp_server = self._get_mcp_server_from_tool_name(prefixed_tool_name) + resolved_by_server_name_only = False + normalized_server_name = normalize_server_name(server_name) + + def _candidate_matches_server_name(candidate: MCPServer) -> bool: + for identifier in ( + candidate.alias, + candidate.server_name, + candidate.name, + ): + if identifier and normalize_server_name(identifier) == ( + normalized_server_name + ): + return True + return False + + if mcp_server is None: + for candidate in self.get_registry().values(): + if _candidate_matches_server_name(candidate): + mcp_server = candidate + resolved_by_server_name_only = True + break + if mcp_server is None: + fallback = self._get_mcp_server_from_tool_name(name) + if fallback is not None and ( + not server_name or _candidate_matches_server_name(fallback) + ): + mcp_server = fallback + if mcp_server is None: + raise ValueError(f"Tool {name} not found") + + if resolved_by_server_name_only: + tool_known = ( + name in self.tool_name_to_mcp_server_name_mapping + or prefixed_tool_name in self.tool_name_to_mcp_server_name_mapping + ) + if not tool_known: + raise ValueError(f"Tool {name} not found") + + return mcp_server + + async def _resolve_oauth2_headers_for_tool_call( + self, + mcp_server: MCPServer, + oauth2_headers: Optional[Dict[str, str]], + user_api_key_auth: Optional[UserAPIKeyAuth], + ) -> Optional[Dict[str, str]]: + """Look up per-user OAuth headers when the client did not supply a token.""" + if ( + not mcp_server.needs_user_oauth_token + or oauth2_headers + or user_api_key_auth is None + ): + return oauth2_headers + + user_id = getattr(user_api_key_auth, "user_id", None) + if not user_id: + return oauth2_headers + + try: + from litellm.proxy._experimental.mcp_server.server import ( # noqa: PLC0415 + _get_user_oauth_extra_headers_from_db, + ) + + stored_headers = await _get_user_oauth_extra_headers_from_db( + server=mcp_server, + user_api_key_auth=user_api_key_auth, + ) + if stored_headers: + return stored_headers + except Exception as _lookup_exc: + verbose_logger.debug( + "call_tool: per-user token lookup failed for " "user=%s server=%s: %s", + user_id, + mcp_server.server_id, + _lookup_exc, + ) + return oauth2_headers + + async def _gather_openapi_tool_tasks( + self, + tasks: List[Any], + proxy_logging_obj: Optional[ProxyLogging], + ) -> CallToolResult: + """Await OpenAPI tool tasks and return the tool call result.""" + try: + mcp_responses = await asyncio.gather(*tasks) + result_index = 1 if proxy_logging_obj else 0 + return cast(CallToolResult, mcp_responses[result_index]) + except ( + BlockedPiiEntityError, + GuardrailRaisedException, + HTTPException, + ) as e: + verbose_logger.error( + f"Guardrail blocked MCP tool call during result check: {str(e)}" + ) + raise e + async def call_tool( self, server_name: str, @@ -2790,12 +3055,7 @@ class MCPServerManager: CallToolResult from the MCP server """ start_time = datetime.datetime.now() - - # Get the MCP server - prefixed_tool_name = add_server_prefix_to_name(name, server_name) - mcp_server = self._get_mcp_server_from_tool_name(prefixed_tool_name) - if mcp_server is None: - raise ValueError(f"Tool {name} not found") + mcp_server = self._resolve_mcp_server_for_tool_call(server_name, name) ######################################################### # Pre MCP Tool Call Hook @@ -2829,36 +3089,9 @@ class MCPServerManager: ) tasks.append(during_hook_task) - # For per-user OAuth servers: if the client didn't supply a token in - # oauth2_headers, look up the stored token from Redis / DB. This is the - # call_tool equivalent of _get_user_oauth_extra_headers_from_db used in - # list_tools. - if ( - mcp_server.needs_user_oauth_token - and not oauth2_headers - and user_api_key_auth is not None - ): - user_id = getattr(user_api_key_auth, "user_id", None) - if user_id: - try: - from litellm.proxy._experimental.mcp_server.server import ( # noqa: PLC0415 - _get_user_oauth_extra_headers_from_db, - ) - - stored_headers = await _get_user_oauth_extra_headers_from_db( - server=mcp_server, - user_api_key_auth=user_api_key_auth, - ) - if stored_headers: - oauth2_headers = stored_headers - except Exception as _lookup_exc: - verbose_logger.debug( - "call_tool: per-user token lookup failed for " - "user=%s server=%s: %s", - user_id, - mcp_server.server_id, - _lookup_exc, - ) + oauth2_headers = await self._resolve_oauth2_headers_for_tool_call( + mcp_server, oauth2_headers, user_api_key_auth + ) # For OpenAPI servers, call the tool handler directly instead of via MCP client if mcp_server.spec_path: @@ -2894,26 +3127,7 @@ class MCPServerManager: hook_extra_headers=hook_result.get("extra_headers"), ) - # For OpenAPI tools, await outside the client context - try: - mcp_responses = await asyncio.gather(*tasks) - - # If proxy_logging_obj is None, the tool call result is at index 0 - # If proxy_logging_obj is not None, the tool call result is at index 1 (after the during hook task) - result_index = 1 if proxy_logging_obj else 0 - result = mcp_responses[result_index] - - return cast(CallToolResult, result) - except ( - BlockedPiiEntityError, - GuardrailRaisedException, - HTTPException, - ) as e: - # Re-raise guardrail exceptions to properly fail the MCP call - verbose_logger.error( - f"Guardrail blocked MCP tool call during result check: {str(e)}" - ) - raise e + return await self._gather_openapi_tool_tasks(tasks, proxy_logging_obj) ######################################################### # End of Methods that call the upstream MCP servers @@ -3008,6 +3222,7 @@ class MCPServerManager: verbose_logger.debug("Loading MCP servers from database into registry...") self._upstream_initialize_instructions_by_server_id.clear() + self._upstream_initialize_instructions_probed_at.clear() # perform authz check to filter the mcp servers user has access to prisma_client = get_prisma_client_or_throw( @@ -3532,9 +3747,11 @@ class MCPServerManager: registration_url=server.registration_url, allow_all_keys=server.allow_all_keys, available_on_public_internet=server.available_on_public_internet, + delegate_auth_to_upstream=server.delegate_auth_to_upstream, is_byok=server.is_byok, byok_description=server.byok_description, byok_api_key_help_url=server.byok_api_key_help_url, + source_url=server.source_url, instructions=server.instructions, ) diff --git a/litellm/proxy/_experimental/mcp_server/oauth_utils.py b/litellm/proxy/_experimental/mcp_server/oauth_utils.py index 343d1bee613..e8b591c39cf 100644 --- a/litellm/proxy/_experimental/mcp_server/oauth_utils.py +++ b/litellm/proxy/_experimental/mcp_server/oauth_utils.py @@ -1,8 +1,10 @@ """Shared helpers for the MCP OAuth authorization endpoints (BYOK + discoverable / pass-through OAuth proxy).""" +import os from ipaddress import ip_address -from urllib.parse import urlparse, urlunparse +from typing import Any, Dict, List, NoReturn, Optional +from urllib.parse import ParseResult, urlparse, urlunparse from fastapi import HTTPException, Request @@ -13,24 +15,95 @@ from litellm.proxy.auth.ip_address_utils import IPAddressUtils # must not be cached — both success and error bodies may reveal secrets. TOKEN_NO_CACHE_HEADERS = {"Cache-Control": "no-store", "Pragma": "no-cache"} +# Stripped from netloc before same-origin comparison so +# ``llm.example.com`` matches ``llm.example.com:443`` (load balancers +# routinely set X-Forwarded-Port: 443 even when the client URL has no +# explicit port, which would otherwise break a literal netloc compare). +_DEFAULT_PORTS = {"http": 80, "https": 443} + +# Env var for ops to allowlist additional redirect_uri origins beyond +# same-origin + loopback — needed for first-party OAuth clients hosted +# on sister domains (e.g. a web app on app.example.com registering as +# an OAuth client of the MCP proxy on llm.example.com). Comma-separated; +# each entry is ``host`` or ``host:port``; a ``*.`` prefix matches any +# subdomain. HTTPS only. +_TRUSTED_REDIRECT_ORIGINS_ENV = "MCP_TRUSTED_REDIRECT_ORIGINS" + +# Comma-separated private-use URI allowlist for native MCP clients. +# A trailing ``*`` is a prefix match; end the prefix with ``/`` (e.g. +# ``myapp://host/oauth/*``) so ``.../oauth/callback*`` does not also +# match ``.../oauth/callback-2``. +_TRUSTED_NATIVE_REDIRECT_URIS_ENV = "MCP_TRUSTED_NATIVE_REDIRECT_URIS" + +# Default allowlist for trusted native redirect URIs. +_DEFAULT_NATIVE_REDIRECT_URIS: List[str] = [ + "cursor://anysphere.cursor-mcp/oauth/callback", +] + +_warned_invalid_proxy_base_url: Optional[str] = None + + +def _oauth_invalid_request( + error_description: str, + *, + hint: Optional[str] = None, + **extra: Any, +) -> NoReturn: + """Raise ``invalid_request`` (RFC 6749) with a debuggable description. + + FastAPI serializes ``detail`` as JSON. Callers still see ``error``: + ``invalid_request``; ``error_description`` and ``hint`` explain what + failed and how to fix it (e.g. reverse-proxy / PROXY_BASE_URL issues). + """ + detail: Dict[str, Any] = { + "error": "invalid_request", + "error_description": error_description, + } + if hint: + detail["hint"] = hint + detail.update(extra) + raise HTTPException(status_code=400, detail=detail) + + +def _origin_label(scheme: str, netloc: str) -> str: + """Human-readable origin for error messages (scheme + host[:port]).""" + return f"{scheme}://{netloc}" if netloc else f"{scheme}://" + + +def _resolve_proxy_base_url_env() -> Optional[str]: + global _warned_invalid_proxy_base_url + configured = os.environ.get("PROXY_BASE_URL", "").strip() + if not configured: + return None + parsed = urlparse(configured) + if parsed.scheme in ("http", "https") and parsed.netloc: + normalized = urlunparse((parsed.scheme, parsed.netloc, parsed.path, "", "", "")) + return normalized.rstrip("/") + if _warned_invalid_proxy_base_url != configured: + verbose_logger.warning( + "PROXY_BASE_URL=%r is not a valid http(s) URL (missing scheme " + "or host) and will be ignored for MCP OAuth origin resolution. " + "Set it to a full URL like https://litellm.example.com.", + configured, + ) + _warned_invalid_proxy_base_url = configured + return None + def get_request_base_url(request: Request) -> str: """ Get the base URL for the request, considering X-Forwarded-* headers. - X-Forwarded-Proto / X-Forwarded-Host / X-Forwarded-Port are only honoured - when the request comes from a configured trusted proxy - (``use_x_forwarded_for`` enabled AND caller in ``mcp_trusted_proxy_ranges``). - Otherwise the request's literal ``base_url`` is returned, so an - untrusted caller cannot poison OAuth-discovery / redirect_uri values - by injecting headers. - - Args: - request: FastAPI Request object - - Returns: - The reconstructed base URL (e.g., "https://proxy.example.com") + Resolution order: ``PROXY_BASE_URL`` env var, then X-Forwarded-* when + the caller is a trusted proxy (``use_x_forwarded_for`` enabled AND + caller in ``mcp_trusted_proxy_ranges``), otherwise the request's + literal ``base_url``. Untrusted callers cannot poison OAuth-discovery + / redirect_uri values by injecting headers. """ + configured = _resolve_proxy_base_url_env() + if configured: + return configured + base_url = str(request.base_url).rstrip("/") parsed = urlparse(base_url) @@ -72,17 +145,15 @@ def validate_loopback_redirect_uri(redirect_uri: str) -> None: ``"127.0.0.1"`` alone would miss ``127.0.0.2`` and the full-form IPv6 loopback ``0:0:0:0:0:0:0:1``. """ - try: - parsed = urlparse(redirect_uri) - except ValueError: - raise HTTPException(status_code=400, detail="invalid_request") + parsed = _parse_redirect_uri_for_validation(redirect_uri) if parsed.scheme not in ("http", "https"): - raise HTTPException(status_code=400, detail="invalid_request") - # Fragments are not allowed in OAuth redirect URIs (RFC 6749 §3.1.2) - # — rejecting them prevents a ``http://127.0.0.1/cb#frag?code=...`` - # from silently eating the authorization code. + _oauth_invalid_request( + f"redirect_uri scheme {parsed.scheme!r} is not allowed; use http or https.", + ) if parsed.fragment: - raise HTTPException(status_code=400, detail="invalid_request") + _oauth_invalid_request( + "redirect_uri must not contain a URL fragment (#...).", + ) host = (parsed.hostname or "").lower() if host == "localhost": return @@ -93,61 +164,367 @@ def validate_loopback_redirect_uri(redirect_uri: str) -> None: # Unparseable host (malformed IPv6, etc.) — treat as invalid, # don't let it bubble up as a 500. pass - raise HTTPException(status_code=400, detail="invalid_request") + _oauth_invalid_request( + "redirect_uri must use a loopback host (localhost or 127.0.0.0/8).", + hint="Native MCP clients should register a callback on http://127.0.0.1:/...", + ) -def validate_trusted_redirect_uri(request: Request, redirect_uri: str) -> None: - """Accept same-origin (proxy's own origin) OR loopback ``redirect_uri``. - - Same-origin is required for the LiteLLM UI's OAuth flow: the UI - redirects to ``/ui/mcp/oauth/callback`` which is not loopback - but is on the proxy's own trusted HTTPS origin. An attacker cannot - host content on the proxy's own origin without already owning the - proxy, so the open-redirect / code-theft primitive that motivated - :func:`validate_loopback_redirect_uri` does not apply here. - - Loopback continues to be accepted for native MCP clients (per - OAuth 2.1 §4.1.2.1 + RFC 8252 §7.3). - - Use this in the discoverable OAuth proxy endpoints that serve both - native clients and the proxy's own UI. BYOK endpoints that only - support native clients should keep - :func:`validate_loopback_redirect_uri`. +def _strip_default_port(scheme: str, netloc: str) -> str: + """Return ``netloc`` lowercased with the scheme's default port + stripped. ``Llm.Example.com:443`` with scheme ``https`` becomes + ``llm.example.com``. Used so a literal netloc comparison between + the proxy's origin and the client redirect_uri survives a load- + balancer that sets ``X-Forwarded-Port: 443``. """ - try: - parsed = urlparse(redirect_uri) - except ValueError: - raise HTTPException(status_code=400, detail="invalid_request") - if parsed.scheme not in ("http", "https"): - raise HTTPException(status_code=400, detail="invalid_request") - if parsed.fragment: - raise HTTPException(status_code=400, detail="invalid_request") + if not netloc: + return netloc + lowered = netloc.lower() + if lowered.startswith("["): + # IPv6 literal: port (if any) appears after the "]". + close = lowered.rfind("]") + if close != -1 and lowered[close + 1 :].startswith(":"): + try: + port = int(lowered[close + 2 :]) + except ValueError: + return lowered + if _DEFAULT_PORTS.get(scheme) == port: + return lowered[: close + 1] + return lowered + if ":" in lowered: + host, _, port_str = lowered.rpartition(":") + try: + port = int(port_str) + except ValueError: + return lowered + if _DEFAULT_PORTS.get(scheme) == port: + return host + return lowered - # Same-origin: scheme + netloc (host[:port]) must match the proxy's - # own base URL at this request (honouring trusted X-Forwarded-*). + +def _parse_trusted_redirect_origins() -> List[str]: + """Parse ``MCP_TRUSTED_REDIRECT_ORIGINS`` into normalized entries. + Empty / unset env var → empty list. Entries are lowercased and any + scheme / path component the operator included is stripped. Default + ``:443`` is also stripped from non-wildcard entries so + ``app.example.com:443`` matches a redirect_netloc whose own ``:443`` + has already been normalized away — the allowlist path is https-only, + so ``:443`` is the only default port that can legitimately appear. + """ + raw = os.environ.get(_TRUSTED_REDIRECT_ORIGINS_ENV, "").strip() + if not raw: + return [] + entries: List[str] = [] + for token in raw.split(","): + entry = token.strip().lower() + if not entry: + continue + if "://" in entry: + entry = entry.split("://", 1)[1] + entry = entry.split("/", 1)[0] + if not entry: + continue + # Wildcards don't express port constraints; leave them alone. + if not entry.startswith("*."): + entry = _strip_default_port("https", entry) + if entry: + entries.append(entry) + return entries + + +def _matches_trusted_origin_entry(netloc: str, entry: str) -> bool: + """``entry`` is either ``host[:port]`` (exact match after port + normalization) or ``*.suffix`` (subdomain wildcard; matches any + strictly-deeper subdomain of ``suffix`` but not ``suffix`` itself). + ``netloc`` is the already-port-normalized, lowercased netloc of + the redirect_uri being validated. + """ + if entry.startswith("*."): + suffix = entry[2:] + if not suffix or suffix.startswith("."): + return False + # Strip port from netloc for wildcard host comparison; + # wildcards don't express port constraints. + host = netloc.split(":", 1)[0] if ":" in netloc else netloc + return host != suffix and host.endswith("." + suffix) + return netloc == entry + + +def _normalize_native_redirect_uri( + parsed, +) -> str: + """Lowercase scheme, netloc, and path for allowlist comparison.""" + return urlunparse( + ( + (parsed.scheme or "").lower(), + (parsed.netloc or "").lower(), + (parsed.path or "").lower(), + "", + "", + "", + ) + ) + + +def _parse_trusted_native_redirect_uris() -> List[str]: + """Built-in native MCP callbacks plus ``MCP_TRUSTED_NATIVE_REDIRECT_URIS``.""" + entries: List[str] = [uri.lower() for uri in _DEFAULT_NATIVE_REDIRECT_URIS] + raw = os.environ.get(_TRUSTED_NATIVE_REDIRECT_URIS_ENV, "").strip() + if not raw: + return entries + for token in raw.split(","): + entry = token.strip().lower() + if entry and entry not in entries: + entries.append(entry) + return entries + + +def _native_wildcard_prefix_matches(normalized: str, prefix: str) -> bool: + """Prefix match for ``entry*`` allowlist rows. + + When the prefix does not end with ``/``, only exact matches or + deeper path segments (``prefix/...``) are accepted — not siblings + like ``prefix-2``. + """ + if not normalized.startswith(prefix): + return False + suffix = normalized[len(prefix) :] + if not suffix: + return True + if prefix.endswith("/"): + return True + return suffix[0] == "/" + + +def _matches_trusted_native_redirect_uri(parsed) -> bool: + """Allowlisted private-use / custom-scheme OAuth callbacks for native MCP clients.""" + if parsed.fragment: + return False + # Query strings are not part of registered redirect_uris (RFC 6749 §3.1.2). + # Rejecting them prevents allowlist bypass via ``.../callback?injected=...``. + if parsed.query: + return False + if not parsed.netloc: + return False + if parsed.username is not None or parsed.password is not None: + return False + if "\\" in parsed.netloc: + return False + + normalized = _normalize_native_redirect_uri(parsed) + for entry in _parse_trusted_native_redirect_uris(): + if entry.endswith("*"): + if _native_wildcard_prefix_matches(normalized, entry[:-1]): + return True + elif normalized == entry: + return True + return False + + +def _parse_redirect_uri_for_validation(redirect_uri: str) -> ParseResult: try: - proxy_base = urlparse(get_request_base_url(request)) - if ( - parsed.netloc - and parsed.scheme == proxy_base.scheme - and parsed.netloc.lower() == proxy_base.netloc.lower() - ): - return + return urlparse(redirect_uri) + except ValueError: + _oauth_invalid_request( + "redirect_uri is not a valid URL.", + hint="Use a full absolute URL for redirect_uri (e.g. https://your-host/ui/mcp/oauth/callback).", + ) + + +def _validate_trusted_http_redirect_shape(parsed: ParseResult) -> bool: + """Return True when ``parsed`` is an allowlisted native callback (caller may return).""" + if parsed.scheme not in ("http", "https"): + if _matches_trusted_native_redirect_uri(parsed): + return True + _oauth_invalid_request( + f"redirect_uri scheme {parsed.scheme!r} is not allowed; use http/https " + "or a registered native callback (e.g. cursor://).", + hint="Add the full URI to MCP_TRUSTED_NATIVE_REDIRECT_URIS for custom native clients.", + ) + if parsed.fragment: + _oauth_invalid_request( + "redirect_uri must not contain a URL fragment (#...).", + ) + if not parsed.netloc: + _oauth_invalid_request( + "redirect_uri must include a host (e.g. https://your-host/path).", + ) + if parsed.username is not None or parsed.password is not None: + _oauth_invalid_request( + "redirect_uri must not contain userinfo (user:pass@host).", + ) + if "\\" in parsed.netloc: + _oauth_invalid_request( + "redirect_uri host must not contain backslashes.", + ) + return False + + +def _resolve_proxy_base_for_redirect(request: Request) -> Optional[str]: + try: + return get_request_base_url(request) except Exception as exc: - # If we can't determine the proxy's origin, fall through to - # loopback. Log so the failure is diagnosable in production. verbose_logger.warning( "validate_trusted_redirect_uri: could not determine proxy origin, " - "falling back to loopback-only check. error=%s", + "falling back to loopback + allowlist. error=%s", exc, ) + return None + + +def _trusted_redirect_uri_is_allowed( + parsed: ParseResult, + redirect_netloc: str, + proxy_base: Optional[str], +) -> bool: + if proxy_base: + proxy_parsed = urlparse(proxy_base) + if ( + parsed.scheme == proxy_parsed.scheme + and redirect_netloc + == _strip_default_port(proxy_parsed.scheme, proxy_parsed.netloc) + ): + return True host = (parsed.hostname or "").lower() if host == "localhost": - return + return True try: if ip_address(host).is_loopback: - return + return True except ValueError: pass - raise HTTPException(status_code=400, detail="invalid_request") + + if parsed.scheme == "https": + for entry in _parse_trusted_redirect_origins(): + if _matches_trusted_origin_entry(redirect_netloc, entry): + return True + return False + + +def _build_trusted_redirect_rejection_message( + redirect_uri: str, + parsed: ParseResult, + redirect_netloc: str, + proxy_base: Optional[str], +) -> str: + """Build a client-facing rejection message. + + Intentionally omits the proxy's resolved scheme / host / port to avoid + leaking internal network topology (e.g. ``http://litellm-internal:4000``) + through an unauthenticated endpoint. Full diagnostic detail — including + the computed proxy base — is logged server-side by the caller. + """ + redirect_origin = _origin_label(parsed.scheme, redirect_netloc) + proxy_parsed = urlparse(proxy_base) if proxy_base else None + proxy_netloc_norm = ( + _strip_default_port(proxy_parsed.scheme, proxy_parsed.netloc) + if proxy_parsed and proxy_parsed.netloc + else "" + ) + + mismatch_parts: List[str] = [] + if proxy_parsed and proxy_parsed.netloc: + if parsed.scheme != proxy_parsed.scheme: + mismatch_parts.append( + f"scheme: redirect_uri uses {parsed.scheme!r}, but the proxy " + "resolved a different scheme " + "(TLS often terminates at ingress — set PROXY_BASE_URL to https://… " + "or trust X-Forwarded-Proto from your ingress)" + ) + if redirect_netloc != proxy_netloc_norm: + mismatch_parts.append( + f"host/port: redirect_uri {redirect_netloc!r} does not match " + "the proxy origin" + ) + + if mismatch_parts: + return ( + f"redirect_uri origin ({redirect_origin}) does not match the proxy " + "origin. " + "; ".join(mismatch_parts) + ) + return ( + f"redirect_uri ({redirect_uri!r}) is not allowed: not same-origin with " + f"the proxy origin, not loopback, and not listed in " + f"{_TRUSTED_REDIRECT_ORIGINS_ENV}." + ) + + +def _raise_trusted_redirect_uri_rejected( + request: Request, + redirect_uri: str, + parsed: ParseResult, + redirect_netloc: str, + proxy_base: Optional[str], +) -> NoReturn: + description = _build_trusted_redirect_rejection_message( + redirect_uri, parsed, redirect_netloc, proxy_base + ) + + hint = ( + "Align the proxy public URL with the browser URL. Set PROXY_BASE_URL to your " + "HTTPS origin (e.g. https://litellm.example.com), or enable " + "general_settings.use_x_forwarded_for with mcp_trusted_proxy_ranges for your " + "ingress. Verify: curl https:///.well-known/oauth-authorization-server " + "| jq .issuer — issuer must match window.location.origin in the UI." + ) + + verbose_logger.warning( + "MCP OAuth: rejecting redirect_uri %r. %s " + "Computed proxy base=%r (PROXY_BASE_URL=%r). " + "Inbound headers: X-Forwarded-Proto=%r X-Forwarded-Host=%r " + "X-Forwarded-Port=%r Host=%r. " + "Trusted-redirect-origins env=%r. " + "Trusted-native-redirect-uris env=%r.", + redirect_uri, + description, + proxy_base, + os.environ.get("PROXY_BASE_URL"), + request.headers.get("X-Forwarded-Proto"), + request.headers.get("X-Forwarded-Host"), + request.headers.get("X-Forwarded-Port"), + request.headers.get("Host"), + os.environ.get(_TRUSTED_REDIRECT_ORIGINS_ENV), + os.environ.get(_TRUSTED_NATIVE_REDIRECT_URIS_ENV), + ) + + _oauth_invalid_request( + description, + hint=hint, + redirect_uri=redirect_uri, + ) + + +def validate_trusted_redirect_uri(request: Request, redirect_uri: str) -> None: + """Accept ``redirect_uri`` when it is (a) same-origin with the + proxy's own request origin, (b) loopback, (c) listed in the + ``MCP_TRUSTED_REDIRECT_ORIGINS`` ops allowlist, or (d) a built-in / + env-configured native MCP client callback (e.g. ``cursor://``). + + Same-origin is VERIA-57's threat-model-safe equivalent of loopback: + an attacker who can host content on the proxy's own HTTPS origin + has already compromised the proxy, so the open-redirect + code- + theft primitive that motivated the loopback-only rule does not + apply. The same reasoning extends to ops-trusted first-party + hosts (e.g. an internal web app registering as an OAuth client of + the proxy on a sister domain). + + Allowlisted non-loopback hosts are accepted only when the + redirect_uri scheme is ``https`` — an attacker on the network + cannot elevate to https without controlling the host's TLS key. + + Use this in the discoverable OAuth proxy endpoints that serve both + native clients and the proxy's UI / cross-origin web clients. The + BYOK endpoints, which only serve native MCP clients, retain + :func:`validate_loopback_redirect_uri`. + """ + parsed = _parse_redirect_uri_for_validation(redirect_uri) + if _validate_trusted_http_redirect_shape(parsed): + return + redirect_netloc = _strip_default_port(parsed.scheme, parsed.netloc) + proxy_base = _resolve_proxy_base_for_redirect(request) + if _trusted_redirect_uri_is_allowed(parsed, redirect_netloc, proxy_base): + return + _raise_trusted_redirect_uri_rejected( + request, redirect_uri, parsed, redirect_netloc, proxy_base + ) diff --git a/litellm/proxy/_experimental/mcp_server/openapi_to_mcp_generator.py b/litellm/proxy/_experimental/mcp_server/openapi_to_mcp_generator.py index 271517bb1e6..de70fe1331e 100644 --- a/litellm/proxy/_experimental/mcp_server/openapi_to_mcp_generator.py +++ b/litellm/proxy/_experimental/mcp_server/openapi_to_mcp_generator.py @@ -305,7 +305,7 @@ def build_input_schema(operation: Dict[str, Any]) -> Dict[str, Any]: def _merge_openapi_tool_request_headers( - static_headers: Dict[str, str] + static_headers: Dict[str, str], ) -> Dict[str, str]: """Merge static closure headers with per-request ContextVar overrides. diff --git a/litellm/proxy/_experimental/mcp_server/rest_endpoints.py b/litellm/proxy/_experimental/mcp_server/rest_endpoints.py index 829863d2dbb..cec5224e183 100644 --- a/litellm/proxy/_experimental/mcp_server/rest_endpoints.py +++ b/litellm/proxy/_experimental/mcp_server/rest_endpoints.py @@ -1,6 +1,17 @@ import importlib from datetime import datetime -from typing import Any, Awaitable, Callable, Dict, List, Literal, Optional, Set, Union +from typing import ( + Any, + Awaitable, + Callable, + Dict, + List, + Literal, + Optional, + Set, + Tuple, + Union, +) from fastapi import APIRouter, Depends, HTTPException, Query, Request, status @@ -51,20 +62,16 @@ if MCP_AVAILABLE: mcp_auth_header: Optional[str], ) -> Optional[Union[Dict[str, str], str]]: """Helper function to get server-specific auth header with case-insensitive matching.""" - if mcp_server_auth_headers and server.alias: - normalized_server_alias = server.alias.lower() - normalized_headers = { - k.lower(): v for k, v in mcp_server_auth_headers.items() - } - server_auth = normalized_headers.get(normalized_server_alias) - if server_auth is not None: - return server_auth - elif mcp_server_auth_headers and server.server_name: - normalized_server_name = server.server_name.lower() - normalized_headers = { - k.lower(): v for k, v in mcp_server_auth_headers.items() - } - server_auth = normalized_headers.get(normalized_server_name) + from litellm.proxy._experimental.mcp_server.utils import ( + lookup_mcp_server_auth_in_headers, + ) + + if mcp_server_auth_headers: + server_auth = lookup_mcp_server_auth_in_headers( + mcp_server_auth_headers, + alias=getattr(server, "alias", None), + server_name=getattr(server, "server_name", None), + ) if server_auth is not None: return server_auth return mcp_auth_header @@ -231,11 +238,32 @@ if MCP_AVAILABLE: ) return mcp_auth_header, mcp_server_auth_headers, raw_headers + def _resolve_mcp_server_id_for_rest( + server_id: str, + allowed_server_ids: Union[Set[str], List[str]], + client_ip: Optional[str] = None, + ) -> str: + """ + Map REST ``server_id`` (UUID, server_name, or alias) to canonical server_id. + + tools/list already did this; tools/call must match so clients can pass + server names like ``order_status_mcp`` instead of only UUIDs. + """ + allowed = set(allowed_server_ids) + if server_id in allowed: + return server_id + by_name = global_mcp_server_manager.get_mcp_server_by_name( + server_id, client_ip=client_ip + ) + if by_name is not None and by_name.server_id in allowed: + return by_name.server_id + return server_id + async def _resolve_allowed_mcp_servers_with_ip_filter( request: Request, user_api_key_dict: UserAPIKeyAuth, server_id: str, - ) -> List[MCPServer]: + ) -> Tuple[List[MCPServer], str]: """ Resolve allowed MCP servers for a tool call with IP filtering. @@ -245,10 +273,10 @@ if MCP_AVAILABLE: server_id: The server ID to validate access for Returns: - List of allowed MCPServer objects + Tuple of (allowed MCPServer objects, canonical server_id) Raises: - HTTPException: If the server_id is not allowed + HTTPException: If the server_id is not allowed or not found """ # Get all auth contexts auth_contexts = await build_effective_auth_contexts(user_api_key_dict) @@ -268,8 +296,41 @@ if MCP_AVAILABLE: ) ) - # Check if the specified server_id is allowed - if server_id not in allowed_server_ids_set: + canonical_server_id = _resolve_mcp_server_id_for_rest( + server_id, allowed_server_ids_set, _rest_client_ip + ) + + if canonical_server_id not in allowed_server_ids_set: + _server = global_mcp_server_manager.get_mcp_server_by_id( + server_id + ) or global_mcp_server_manager.get_mcp_server_by_name(server_id) + if ( + _server is not None + and _rest_client_ip is not None + and not global_mcp_server_manager._is_server_accessible_from_ip( + _server, _rest_client_ip + ) + ): + raise HTTPException( + status_code=403, + detail={ + "error": "ip_filtering", + "message": ( + f"MCP server '{server_id}' is not accessible from your IP address " + f"({_rest_client_ip}). This server is restricted to internal " + "networks only. To make it externally accessible, set " + "'available_on_public_internet: true' in the server configuration." + ), + }, + ) + if _server is None: + raise HTTPException( + status_code=404, + detail={ + "error": "server_not_found", + "message": f"MCP server '{server_id}' was not found", + }, + ) raise HTTPException( status_code=403, detail={ @@ -285,7 +346,7 @@ if MCP_AVAILABLE: if server is not None: allowed_mcp_servers.append(server) - return allowed_mcp_servers + return allowed_mcp_servers, canonical_server_id async def _get_tools_for_single_server( server, @@ -301,6 +362,7 @@ if MCP_AVAILABLE: extra_headers=extra_headers, add_prefix=False, raw_headers=raw_headers, + user_api_key_auth=user_api_key_auth, ) # Filter tools based on allowed_tools configuration @@ -753,7 +815,7 @@ if MCP_AVAILABLE: }, ) - tool_arguments = data.get("arguments") + tool_arguments = data.get("arguments") or {} proxy_base_llm_response_processor = ProxyBaseLLMRequestProcessing(data=data) ( @@ -786,14 +848,18 @@ if MCP_AVAILABLE: data["user_api_key_auth"] = data["metadata"]["user_api_key_auth"] # Resolve allowed MCP servers with IP filtering - allowed_mcp_servers = await _resolve_allowed_mcp_servers_with_ip_filter( + ( + allowed_mcp_servers, + canonical_server_id, + ) = await _resolve_allowed_mcp_servers_with_ip_filter( request, user_api_key_dict, server_id ) # Look up per-user OAuth headers for this server (mirrors list_tool_rest_api). user_oauth_extra_headers: Optional[Dict[str, str]] = None target_server = next( - (s for s in allowed_mcp_servers if s.server_id == server_id), None + (s for s in allowed_mcp_servers if s.server_id == canonical_server_id), + None, ) if target_server is not None: user_oauth_extra_headers = await _get_user_oauth_extra_headers( @@ -812,6 +878,7 @@ if MCP_AVAILABLE: oauth2_headers=user_oauth_extra_headers or data.get("oauth2_headers"), raw_headers=data.get("raw_headers"), litellm_logging_obj=data.get("litellm_logging_obj"), + requested_server_id=canonical_server_id, ) return result except BlockedPiiEntityError as e: diff --git a/litellm/proxy/_experimental/mcp_server/server.py b/litellm/proxy/_experimental/mcp_server/server.py index 276a6e8a3bb..a05ce3f7417 100644 --- a/litellm/proxy/_experimental/mcp_server/server.py +++ b/litellm/proxy/_experimental/mcp_server/server.py @@ -6,6 +6,8 @@ LiteLLM MCP Server Routes import asyncio import contextlib +import hashlib +import json import time import types import traceback @@ -23,11 +25,12 @@ from typing import ( cast, ) +import httpx from fastapi import FastAPI, HTTPException from pydantic import AnyUrl, ConfigDict from starlette.requests import Request as StarletteRequest from starlette.responses import JSONResponse -from starlette.types import Receive, Scope, Send +from starlette.types import Message, Receive, Scope, Send from litellm._logging import verbose_logger from litellm.constants import MAXIMUM_TRACEBACK_LINES_TO_LOG @@ -51,13 +54,17 @@ from litellm.proxy._experimental.mcp_server.utils import ( get_server_prefix, iter_known_server_prefixes, ) +from litellm.llms.custom_httpx.http_handler import ( + get_async_httpx_client, + httpxSpecialProvider, +) from litellm.proxy._types import UserAPIKeyAuth from litellm.proxy.auth.ip_address_utils import IPAddressUtils from litellm.proxy.litellm_pre_call_utils import ( LiteLLMProxyRequestSetup, get_chain_id_from_headers, ) -from litellm.types.mcp import MCPAuth +from litellm.types.mcp import MCPAuth, MCPSpecVersion from litellm.types.mcp_server.mcp_server_manager import MCPInfo, MCPServer from litellm.types.utils import CallTypes, StandardLoggingMCPToolCall from litellm.utils import Rules, client, function_setup @@ -69,6 +76,19 @@ from litellm.utils import Rules, client, function_setup _byok_cred_cache: Dict[Tuple[str, str], Tuple[Optional[str], float]] = {} _BYOK_CRED_CACHE_TTL = 60 # seconds _BYOK_CRED_CACHE_MAX_SIZE = 4096 # cap to prevent unbounded growth +_STATEFUL_SESSION_IDLE_TIMEOUT_SECONDS = 30 * 60 +# Upper bound on concurrent stateful sessions a single caller may hold. Each +# `initialize` creates a session that survives until the idle timeout, so +# without a cap an authenticated client could spam `initialize` and exhaust +# memory. The caller's own oldest idle sessions are evicted to make room; if +# the cap is still hit (every session in flight), the new `initialize` is +# rejected with 429. +_MAX_STATEFUL_SESSIONS_PER_OWNER = 100 +# Maximum bytes to peek when sniffing the JSON-RPC method on a POST. +# An `initialize` envelope is a few hundred bytes; capping the peek +# prevents an authenticated client from forcing the proxy to buffer an +# arbitrarily large body just to make a routing decision. +_MCP_ROUTING_PEEK_MAX_BYTES = 4096 def _invalidate_byok_cred_cache(user_id: str, server_id: str) -> None: @@ -237,13 +257,45 @@ if MCP_AVAILABLE: sse: SseServerTransport = SseServerTransport("/mcp/sse/messages") # Create session managers - session_manager = StreamableHTTPSessionManager( + session_manager_stateless = StreamableHTTPSessionManager( app=server, event_store=None, json_response=False, # enables SSE streaming stateless=True, ) + session_manager_stateful = StreamableHTTPSessionManager( + app=server, + event_store=None, # TODO: Add EventStore for reconnection/event replay if needed + json_response=False, # enables SSE streaming + stateless=False, + ) + _stateful_session_auth_contexts: Dict[str, MCPAuthenticatedUser] = {} + _stateful_session_auth_context_last_seen: Dict[str, float] = {} + # Maps session_id -> owner identifier (hashed API key/token) so we can + # reject requests that supply a session_id created by a different caller. + # Without this, a leaked mcp-session-id could be driven (or terminated) + # by any other authenticated proxy user. + _stateful_session_owners: Dict[str, str] = {} + # Per-session lock that serializes ``handle_request`` for the same + # mcp-session-id. The stored ``MCPAuthenticatedUser`` is mutated in place + # by ``_update_auth_context`` each request; without this lock, two + # concurrent requests on the same session would clobber each other's + # auth headers / mcp_servers / oauth state while in-flight callbacks are + # still reading the shared object. + _stateful_session_locks: Dict[str, asyncio.Lock] = {} + _stateful_session_active_request_counts: Dict[str, int] = {} + + def _remove_stateful_session_tracking(session_id: str) -> None: + _stateful_session_auth_contexts.pop(session_id, None) + _stateful_session_auth_context_last_seen.pop(session_id, None) + _stateful_session_owners.pop(session_id, None) + _stateful_session_locks.pop(session_id, None) + _stateful_session_active_request_counts.pop(session_id, None) + + # Keep this alias so existing references to session_manager still work + session_manager = session_manager_stateless + # Create SSE session manager sse_session_manager = StreamableHTTPSessionManager( app=server, @@ -254,11 +306,100 @@ if MCP_AVAILABLE: # Context managers for proper lifecycle management _session_manager_cm = None + _session_manager_stateful_cm = None _sse_session_manager_cm = None + _stateful_auth_context_cleanup_task: Optional[asyncio.Task] = None + + async def _purge_expired_stateful_session_auth_contexts( + now: Optional[float] = None, + ) -> None: + """Terminate expired stateful sessions and drop their auth contexts.""" + now = time.monotonic() if now is None else now + server_instances = getattr(session_manager_stateful, "_server_instances", {}) + expired_session_ids = [] + for session_id, last_seen in _stateful_session_auth_context_last_seen.items(): + if _stateful_session_active_request_counts.get(session_id, 0) > 0: + continue + if ( + now - last_seen >= _STATEFUL_SESSION_IDLE_TIMEOUT_SECONDS + or session_id not in server_instances + ): + expired_session_ids.append(session_id) + + for session_id in expired_session_ids: + # Re-check the active-request count immediately before tearing + # the session down. ``await transport.terminate()`` yields to + # the event loop, so a request that started after the first + # collection pass could otherwise observe its transport being + # ripped out from under it mid-flight. + if _stateful_session_active_request_counts.get(session_id, 0) > 0: + continue + # Pop transport + terminate BEFORE removing owner/auth tracking. + # Reversing the order avoids a window where ``_stateful_session_owners`` + # is empty but ``server_instances`` still serves the session — a + # concurrent request in that window would observe ``expected_owner + # is None`` and bypass the owner-binding check. + transport = server_instances.pop(session_id, None) + if transport is not None: + await transport.terminate() + _remove_stateful_session_tracking(session_id) + + for session_id in list(_stateful_session_auth_context_last_seen): + if session_id not in _stateful_session_auth_contexts: + _remove_stateful_session_tracking(session_id) + + async def _enforce_stateful_session_cap_for_owner(owner: str) -> bool: + """ + Bound the number of concurrent stateful sessions a single caller holds + before routing a new ``initialize`` to the stateful manager. + + Evicts the caller's *own* oldest idle sessions (no in-flight requests) + to make room, so a busy-but-legitimate client keeps its newest sessions + and other callers are never affected. Returns ``True`` if the new + session may proceed, or ``False`` when the caller is already at the cap + with every session in flight (the new ``initialize`` should be rejected). + """ + server_instances = getattr(session_manager_stateful, "_server_instances", {}) + + def _owned_live_session_ids() -> List[str]: + return [ + session_id + for session_id, session_owner in _stateful_session_owners.items() + if session_owner == owner and session_id in server_instances + ] + + owned = _owned_live_session_ids() + if len(owned) < _MAX_STATEFUL_SESSIONS_PER_OWNER: + return True + + for session_id in sorted( + owned, + key=lambda sid: _stateful_session_auth_context_last_seen.get(sid, 0.0), + ): + if len(_owned_live_session_ids()) < _MAX_STATEFUL_SESSIONS_PER_OWNER: + break + if _stateful_session_active_request_counts.get(session_id, 0) > 0: + continue + transport = server_instances.pop(session_id, None) + if transport is not None: + await transport.terminate() + _remove_stateful_session_tracking(session_id) + + return len(_owned_live_session_ids()) < _MAX_STATEFUL_SESSIONS_PER_OWNER + + async def _cleanup_expired_stateful_session_auth_contexts() -> None: + while True: + await asyncio.sleep(_STATEFUL_SESSION_IDLE_TIMEOUT_SECONDS) + try: + await _purge_expired_stateful_session_auth_contexts() + except Exception as e: + verbose_logger.exception( + f"Error cleaning up expired MCP stateful sessions: {e}" + ) async def initialize_session_managers(): """Initialize the session managers. Can be called from main app lifespan.""" - global _SESSION_MANAGERS_INITIALIZED, _session_manager_cm, _sse_session_manager_cm + global _SESSION_MANAGERS_INITIALIZED, _session_manager_cm, _session_manager_stateful_cm, _sse_session_manager_cm, _stateful_auth_context_cleanup_task # Use async lock to prevent concurrent initialization async with _INITIALIZATION_LOCK: @@ -268,12 +409,17 @@ if MCP_AVAILABLE: verbose_logger.info("Initializing MCP session managers...") # Start the session managers with context managers - _session_manager_cm = session_manager.run() + _session_manager_cm = session_manager_stateless.run() + _session_manager_stateful_cm = session_manager_stateful.run() _sse_session_manager_cm = sse_session_manager.run() # Enter the context managers await _session_manager_cm.__aenter__() + await _session_manager_stateful_cm.__aenter__() await _sse_session_manager_cm.__aenter__() + _stateful_auth_context_cleanup_task = asyncio.create_task( + _cleanup_expired_stateful_session_auth_contexts() + ) _SESSION_MANAGERS_INITIALIZED = True verbose_logger.info( @@ -282,21 +428,29 @@ if MCP_AVAILABLE: async def shutdown_session_managers(): """Shutdown the session managers.""" - global _SESSION_MANAGERS_INITIALIZED, _session_manager_cm, _sse_session_manager_cm + global _SESSION_MANAGERS_INITIALIZED, _session_manager_cm, _session_manager_stateful_cm, _sse_session_manager_cm, _stateful_auth_context_cleanup_task if _SESSION_MANAGERS_INITIALIZED: verbose_logger.info("Shutting down MCP session managers...") try: + if _stateful_auth_context_cleanup_task: + _stateful_auth_context_cleanup_task.cancel() + with contextlib.suppress(asyncio.CancelledError): + await _stateful_auth_context_cleanup_task if _session_manager_cm: await _session_manager_cm.__aexit__(None, None, None) + if _session_manager_stateful_cm: + await _session_manager_stateful_cm.__aexit__(None, None, None) if _sse_session_manager_cm: await _sse_session_manager_cm.__aexit__(None, None, None) except Exception as e: verbose_logger.exception(f"Error during session manager shutdown: {e}") _session_manager_cm = None + _session_manager_stateful_cm = None _sse_session_manager_cm = None + _stateful_auth_context_cleanup_task = None _SESSION_MANAGERS_INITIALIZED = False @contextlib.asynccontextmanager @@ -361,7 +515,7 @@ if MCP_AVAILABLE: @server.call_tool() async def mcp_server_tool_call( - name: str, arguments: Dict[str, Any] | None + name: str, arguments: Optional[Dict[str, Any]] ) -> CallToolResult: """ Call a specific tool with the provided arguments @@ -404,7 +558,7 @@ if MCP_AVAILABLE: if host_token and hasattr(host_ctx, "session") and host_ctx.session: host_session = host_ctx.session - async def forward_progress(progress: float, total: float | None): + async def forward_progress(progress: float, total: Optional[float]): """Forward progress notifications from external MCP to Host""" try: await host_session.send_progress_notification( @@ -546,7 +700,7 @@ if MCP_AVAILABLE: @server.get_prompt() async def get_prompt( - name: str, arguments: dict[str, str] | None + name: str, arguments: Optional[Dict[str, str]] ) -> GetPromptResult: """ Get a specific prompt with the provided arguments @@ -1109,10 +1263,16 @@ if MCP_AVAILABLE: ) -> Tuple[Optional[Union[Dict[str, str], str]], Optional[Dict[str, str]]]: """Build auth and extra headers for a server.""" server_auth_header: Optional[Union[Dict[str, str], str]] = None - if mcp_server_auth_headers and server.alias is not None: - server_auth_header = mcp_server_auth_headers.get(server.alias) - elif mcp_server_auth_headers and server.server_name is not None: - server_auth_header = mcp_server_auth_headers.get(server.server_name) + if mcp_server_auth_headers: + from litellm.proxy._experimental.mcp_server.utils import ( + lookup_mcp_server_auth_in_headers, + ) + + server_auth_header = lookup_mcp_server_auth_in_headers( + mcp_server_auth_headers, + alias=server.alias, + server_name=server.server_name, + ) extra_headers: Optional[Dict[str, str]] = None if server.auth_type == MCPAuth.oauth2: @@ -1154,7 +1314,7 @@ if MCP_AVAILABLE: def _merge_gateway_initialize_instructions( allowed_mcp_servers: List[MCPServer], ) -> Optional[str]: - """YAML/DB override, else in-memory upstream text from list_tools / health_check / call_tool.""" + """YAML/DB override, else upstream text (prefetch on init, or list_tools / health_check / call_tool cache).""" if not allowed_mcp_servers: return None @@ -1195,6 +1355,20 @@ if MCP_AVAILABLE: mcp_servers=mcp_servers, client_ip=client_ip, ) + if allowed: + # return_exceptions=True: a per-server probe failure (incl. CancelledError + # bubbled from anyio task group teardown on connection refused) must not + # cancel sibling probes or 500 the gateway initialize request. + await asyncio.gather( + *[ + global_mcp_server_manager._ensure_upstream_initialize_instructions_cached( + s + ) + for s in allowed + if s is not None + ], + return_exceptions=True, + ) merged = _merge_gateway_initialize_instructions(allowed_mcp_servers=allowed) tok = _mcp_gateway_initialize_instructions.set(merged) try: @@ -1332,8 +1506,24 @@ if MCP_AVAILABLE: raw_headers=raw_headers, ) - # If no OAuth2 token came from request headers, fall back to pre-fetched creds - if extra_headers is None and server.auth_type == MCPAuth.oauth2: + # Prefer server-stored per-user OAuth when configured, so a stale + # Authorization header from the MCP client cannot override Redis/DB + # (same issue as call_tool in mcp_server_manager: VS Code caches tokens). + if ( + server.auth_type == MCPAuth.oauth2 + and getattr(server, "needs_user_oauth_token", False) + and user_api_key_auth is not None + ): + db_headers = await _get_user_oauth_extra_headers_from_db( + server, + user_api_key_auth, + prefetched_creds=_prefetched_oauth_creds, + ) + if db_headers: + extra_headers = db_headers + + # If still no OAuth2 token, fall back to pre-fetched creds (non-stale-client path) + elif extra_headers is None and server.auth_type == MCPAuth.oauth2: extra_headers = await _get_user_oauth_extra_headers_from_db( server, user_api_key_auth, @@ -1347,6 +1537,7 @@ if MCP_AVAILABLE: extra_headers=extra_headers, add_prefix=True, # Always add server prefix raw_headers=raw_headers, + user_api_key_auth=user_api_key_auth, ) filtered_tools = filter_tools_by_allowed_tools(tools, server) @@ -2053,6 +2244,7 @@ if MCP_AVAILABLE: """ # Track resolved MCP server for both permission checks and dispatch mcp_server: Optional[MCPServer] = None + requested_server_id: Optional[str] = kwargs.get("requested_server_id") # If the client called with a display-name override (e.g. "Get Pet"), # translate it back to the original prefixed name before any routing. @@ -2061,14 +2253,55 @@ if MCP_AVAILABLE: # Remove prefix from tool name for logging and processing original_tool_name, server_name = split_server_prefix_from_name(name) + requested_server: Optional[MCPServer] = None + if requested_server_id: + requested_server = next( + (s for s in allowed_mcp_servers if s.server_id == requested_server_id), + None, + ) + # Resolve the actual MCP server up-front so the permission check uses # the canonical server.name even when the tool name is prefixed with a # short ID (LITELLM_USE_SHORT_MCP_TOOL_PREFIX) that doesn't match the # server's display name directly. mcp_server = global_mcp_server_manager._get_mcp_server_from_tool_name(name) + if mcp_server is None and requested_server is not None: + # REST callers may pass the raw tool name (no prefix) plus a + # ``requested_server_id``. The mapping might only contain the + # prefixed form, so retry the lookup with every known prefix of + # the requested server before treating the tool as unresolved — + # otherwise the tool_server_mismatch guard below is silently + # bypassed. + for known_prefix in iter_known_server_prefixes(requested_server): + candidate = global_mcp_server_manager._get_mcp_server_from_tool_name( + add_server_prefix_to_name(name, known_prefix) + ) + if candidate is not None: + mcp_server = candidate + break if mcp_server is not None: server_name = mcp_server.name + # REST /mcp-rest/tools/call passes server_id — tool must belong to that server + if requested_server is not None: + if ( + mcp_server is not None + and mcp_server.server_id != requested_server.server_id + ): + raise HTTPException( + status_code=403, + detail={ + "error": "tool_server_mismatch", + "message": ( + f"Tool '{name}' belongs to MCP server '{mcp_server.name}' " + f"but request specified server_id for '{requested_server.name}'." + ), + }, + ) + if mcp_server is None: + mcp_server = requested_server + server_name = requested_server.name + # Only enforce server-level permissions when we can resolve a server if server_name: if not MCPRequestHandler.is_tool_allowed( @@ -2536,6 +2769,10 @@ if MCP_AVAILABLE: import re mcp_servers_from_path: Optional[List[str]] = None + segments = [s for s in path.split("/") if s] + if len(segments) >= 2 and segments[1] == "mcp" and segments[0] != "mcp": + return [segments[0]] + # Match /mcp/ # Where servers can be comma-separated list of server names # Server names can contain slashes (e.g., "custom_solutions/user_123") @@ -2609,6 +2846,144 @@ if MCP_AVAILABLE: raw_headers, ) + def _get_session_id_from_scope(scope: Scope) -> Optional[str]: + """ + Extract mcp-session-id from ASGI scope headers. + Returns None if not present. + """ + for header_name, header_value in scope.get("headers", []): + name = ( + header_name if isinstance(header_name, bytes) else header_name.encode() + ) + if name.lower() == b"mcp-session-id": + return ( + header_value.decode() + if isinstance(header_value, bytes) + else str(header_value) + ) + return None + + def _owner_fingerprint_for( + user_api_key_auth: Optional[UserAPIKeyAuth], + oauth2_headers: Optional[Dict[str, str]] = None, + client_ip: Optional[str] = None, + ) -> str: + """ + Stable, non-reversible identifier for the caller used to bind an + mcp-session-id to its creator. Hash the resolved credential before + using it so custom key formats are never stored in cleartext. + + For OAuth2 passthrough (``UserAPIKeyAuth()`` with no key/user_id), + the caller's identity is the upstream OAuth bearer; hash it so two + OAuth callers with different tokens don't both fingerprint to + ``anonymous`` and end up sharing a session. + + When no caller-identifying credentials are available at all + (e.g. proxy running without master key, or an unauthenticated + passthrough path), fall back to the client IP so two unrelated + anonymous callers from different sources do not collapse to a + single ``anonymous`` owner and end up able to drive each other's + stateful sessions. Note: when even client IP is unavailable + (exotic deployments without trusted X-Forwarded-For and direct + socket info), the fingerprint degrades to the ``anonymous`` + sentinel and cannot meaningfully protect against another + unauthenticated caller who learns the session id — owner-binding + is best-effort in that mode. + """ + + def _bytes_for_hash(value: Any) -> Optional[bytes]: + """Only hash str/bytes secrets; skip mocks and other unexpected types.""" + if value is None: + return None + if isinstance(value, (bytes, bytearray)): + return bytes(value) + if isinstance(value, str): + return value.encode("utf-8") + return None + + if user_api_key_auth is not None: + key_material = _bytes_for_hash(getattr(user_api_key_auth, "api_key", None)) + if key_material: + api_key_hash = hashlib.sha256(key_material).hexdigest() + return f"key:{api_key_hash}" + uid_material = _bytes_for_hash(getattr(user_api_key_auth, "user_id", None)) + if uid_material: + user_id_hash = hashlib.sha256(uid_material).hexdigest() + return f"user:{user_id_hash}" + if oauth2_headers: + authz = oauth2_headers.get("Authorization") or oauth2_headers.get( + "authorization" + ) + authz_bytes = _bytes_for_hash(authz) + if authz_bytes: + return f"oauth:{hashlib.sha256(authz_bytes).hexdigest()}" + if client_ip and isinstance(client_ip, str): + return f"ip:{hashlib.sha256(client_ip.encode('utf-8')).hexdigest()}" + return "anonymous" + + def _is_initialize_request(body: bytes) -> bool: + """ + Check if the request body is a JSON-RPC initialize method. + Returns True if method is "initialize", False otherwise or on parse error. + """ + if not body: + return False + try: + data = json.loads(body) + return isinstance(data, dict) and data.get("method") == "initialize" + except (json.JSONDecodeError, TypeError): + return False + + async def _read_request_body_for_routing( + receive: Receive, + ) -> Tuple[List[Message], bytes]: + """ + Read just enough of the request body to decide whether this is a + JSON-RPC ``initialize`` call. Returns the consumed ASGI messages so + the caller can replay them faithfully to the downstream handler, and + the peeked body bytes (capped at ``_MCP_ROUTING_PEEK_MAX_BYTES``). + + Stops reading from the wire as soon as either (a) we have peeked + ``_MCP_ROUTING_PEEK_MAX_BYTES`` of body, or (b) the body is complete. + The remainder of an oversized body is streamed lazily through + ``wrapped_receive`` in the caller — so an authenticated client cannot + force the proxy to buffer an arbitrarily large payload just to make a + routing decision. + """ + consumed_messages: List[Message] = [] + body_chunks: List[bytes] = [] + peeked_bytes = 0 + + while True: + message = await receive() + consumed_messages.append(message) + + if message.get("type") != "http.request": + break + + body = message.get("body", b"") or b"" + if body: + # Only retain up to the remaining peek budget for sniffing. + # The full ``message`` is already in memory (delivered by + # the ASGI server) and must round-trip to the downstream + # handler via ``consumed_messages``, but ``body_chunks`` is + # purely for the JSON-RPC method check — there is no reason + # to copy a large body frame into a second buffer. + remaining = _MCP_ROUTING_PEEK_MAX_BYTES - peeked_bytes + if remaining > 0: + body_chunks.append(body[:remaining]) + peeked_bytes += min(len(body), remaining) + + if not message.get("more_body", False): + break + + if peeked_bytes >= _MCP_ROUTING_PEEK_MAX_BYTES: + # Stop draining; downstream replay will pull remaining chunks + # directly from the original `receive` via wrapped_receive. + break + + return consumed_messages, b"".join(body_chunks) + async def _handle_stale_mcp_session( scope: Scope, receive: Receive, @@ -2672,6 +3047,7 @@ if MCP_AVAILABLE: method = scope.get("method", "").upper() if method == "DELETE": + _remove_stateful_session_tracking(_session_id) verbose_logger.info( "DELETE request for non-existent MCP session '%s'. " "Returning success (idempotent DELETE).", @@ -2754,7 +3130,158 @@ if MCP_AVAILABLE: ) return user_api_key_auth.model_copy(update={"object_permission": updated_op}) - async def handle_streamable_http_mcp( + def _get_forwarded_auth_from_scope(scope: Scope) -> Optional[str]: + """Return the upstream-bound ``Authorization`` header value, or None. + + Only returns the ``Authorization`` header when ``x-litellm-api-key`` is + also present. In that case ``Authorization`` is unambiguously the + upstream token the caller wants forwarded to the MCP server. When + ``x-litellm-api-key`` is absent the ``Authorization`` header may itself + be the LiteLLM proxy API key (backward-compat path in + ``MCPRequestHandler.process_mcp_request``), and forwarding it upstream + would leak the proxy key to a third-party MCP server. + """ + authorization = None + has_litellm_key_header = False + for key, value in scope.get("headers", []): + key_lower = key.lower() + if key_lower == b"authorization": + authorization = value.decode("latin-1") + elif key_lower == b"x-litellm-api-key": + has_litellm_key_header = True + if not has_litellm_key_header: + return None + return authorization + + async def _probe_upstream_auth( + url: str, + auth_header: str, + timeout: float = 5.0, + ) -> tuple: + """JSON-RPC initialize-probe the upstream URL to check whether the token is accepted. + + Uses POST so StreamableHTTP MCP servers run the same auth path as a + real client request. Returns (status_code, www_authenticate). + Fails-open with (200, None) on network errors so a transient hiccup + does not block valid requests. + + Uses the public ``AsyncHTTPHandler.post()`` interface and catches + ``httpx.HTTPStatusError`` separately so the 401/403 we want to surface + is not swallowed by the broad fail-open ``except Exception`` below. + """ + client = get_async_httpx_client( + llm_provider=httpxSpecialProvider.MCP, + params={"timeout": timeout}, + ) + probe_payload = { + "jsonrpc": "2.0", + "id": "litellm-mcp-auth-probe", + "method": "initialize", + "params": { + "protocolVersion": MCPSpecVersion.jun_2025.value, + "capabilities": {}, + "clientInfo": { + "name": "litellm-mcp-auth-probe", + "version": "1.0.0", + }, + }, + } + probe_headers = { + "Authorization": auth_header, + "Accept": "application/json, text/event-stream", + } + try: + resp = await client.post( + url=url, + headers=probe_headers, + json=probe_payload, + timeout=timeout, + ) + return resp.status_code, resp.headers.get("www-authenticate") + except httpx.HTTPStatusError as exc: + # AsyncHTTPHandler.post() calls raise_for_status(); a 401/403 from + # upstream lands here. Return its status so the caller can map it + # to the appropriate response. + return exc.response.status_code, exc.response.headers.get( + "www-authenticate" + ) + except Exception as exc: + verbose_logger.debug( + f"_probe_upstream_auth: probe to {url} failed ({exc}), allowing request through" + ) + return 200, None + + async def _check_passthrough_upstream_auth( + scope: Scope, + user_api_key_auth: Optional[UserAPIKeyAuth], + mcp_servers: Optional[List[str]], + client_ip: Optional[str], + ) -> None: + """Probe pass-through upstream servers in parallel before the MCP session starts. + + Only servers the caller's key is already authorized to reach are probed — + the list is derived from _get_allowed_mcp_servers so that a user cannot + trigger an upstream probe against a server their key is not permitted for. + + The MCP SDK commits HTTP 200 headers before invoking handlers, so a 401 + can only be returned before that point. This function raises HTTPException(401) + with a WWW-Authenticate header if any upstream rejects the client token. + Fails-open: network errors are logged and the request is allowed through. + """ + forwarded_auth = _get_forwarded_auth_from_scope(scope) + if not forwarded_auth: + return + + # Use the authorized server set, not the raw user-supplied names, so that + # a caller cannot force a probe to a server their key is not allowed to use. + allowed_servers = await _get_allowed_mcp_servers( + user_api_key_auth=user_api_key_auth, + mcp_servers=mcp_servers, + client_ip=client_ip, + ) + passthrough_servers = [ + srv + for srv in allowed_servers + if srv.extra_headers + and any(h.lower() == "authorization" for h in srv.extra_headers) + # Exclude M2M servers: _prepare_mcp_server_headers skips caller + # Authorization when has_client_credentials is set, so probing + # those with the caller's token would send the wrong credential. + and not srv.has_client_credentials + ] + if not passthrough_servers: + return + + probe_results = await asyncio.gather( + *[ + _probe_upstream_auth(srv.url or "", forwarded_auth) + for srv in passthrough_servers + ] + ) + request = StarletteRequest(scope) + base_url = get_request_base_url(request) + for srv, (probe_status, _) in zip(passthrough_servers, probe_results): + if probe_status == 401: + # Token is missing or expired — direct the client to re-authorize. + authorization_uri = ( + f"Bearer authorization_uri=" + f"{base_url}/.well-known/oauth-authorization-server/{srv.name}" + ) + raise HTTPException( + status_code=401, + detail="Unauthorized", + headers={"WWW-Authenticate": authorization_uri}, + ) + if probe_status == 403: + # Token is valid but the caller lacks permission — do not hint + # at re-authorization (RFC 9110: a fresh token with the same + # scopes would just hit 403 again and loop indefinitely). + raise HTTPException( + status_code=403, + detail="Forbidden", + ) + + async def handle_streamable_http_mcp( # noqa: PLR0915 scope: Scope, receive: Receive, send: Send ) -> None: """Handle MCP requests through StreamableHTTP.""" @@ -2827,6 +3354,13 @@ if MCP_AVAILABLE: user_api_key_auth, active_toolset_id ) + # Pre-flight auth check for pass-through servers. Must run after + # toolset scoping so the probe list is derived from the fully-authorized + # server set, not the raw user-supplied names. + await _check_passthrough_upstream_auth( + scope, user_api_key_auth, mcp_servers, _client_ip + ) + # Inject masked debug headers when client sends x-litellm-mcp-debug: true _debug_headers = MCPDebug.maybe_build_debug_headers( raw_headers=raw_headers, @@ -2840,38 +3374,215 @@ if MCP_AVAILABLE: if _debug_headers: send = MCPDebug.wrap_send_with_debug_headers(send, _debug_headers) - # Set the auth context variable for easy access in MCP functions - set_auth_context( - user_api_key_auth=user_api_key_auth, - mcp_auth_header=mcp_auth_header, - mcp_servers=mcp_servers, - mcp_server_auth_headers=mcp_server_auth_headers, - oauth2_headers=oauth2_headers, - raw_headers=raw_headers, - client_ip=_client_ip, - ) - # Ensure session managers are initialized if not _SESSION_MANAGERS_INITIALIZED: await initialize_session_managers() # Give it a moment to start up await asyncio.sleep(0.1) - # Handle stale session IDs - either strip them for reconnection - # or return success for idempotent DELETE operations - handled = await _handle_stale_mcp_session( - scope, receive, send, session_manager - ) - if handled: - # Request was fully handled (e.g., DELETE on non-existent session) - return + # Route based on mcp-session-id and request method: + # - Has session ID → stateful (Claude Code, Cursor, VSCode) + # - No session ID + initialize → stateful (so client gets mcp-session-id) + # - No session ID + other → stateless (curl, Inspector, Notion) + session_id = _get_session_id_from_scope(scope) + is_initialize = False + consumed_messages: List[Message] = [] - async with _gateway_initialize_instructions_request_scope( - user_api_key_auth, - mcp_servers, - _client_ip, - ): - await session_manager.handle_request(scope, receive, send) + # Owner-binding: a live stateful session may only be driven by the + # caller that created it. Reject mismatches with 403 so a leaked + # mcp-session-id cannot be hijacked by another authenticated user. + # + # Run before ``_handle_stale_mcp_session`` so a non-owner cannot + # force-clean another caller's residual tracking entries via a + # stale DELETE, and before peeking the request body so the 403 + # response sees a pristine ``receive`` channel. + if session_id: + expected_owner = _stateful_session_owners.get(session_id) + request_owner = _owner_fingerprint_for( + user_api_key_auth, oauth2_headers, _client_ip + ) + if expected_owner is not None and expected_owner != request_owner: + verbose_logger.warning( + "Rejecting MCP request: session '%s' owner mismatch.", + session_id, + ) + forbidden_response = JSONResponse( + status_code=403, + content={ + "error": "Forbidden", + "details": "mcp-session-id is bound to a different caller.", + }, + ) + await forbidden_response(scope, receive, send) + return + + # Handle stale session IDs before choosing a target manager. Stale + # non-DELETE requests have their session header stripped and should + # be routed as no-session requests. + if session_id: + handled = await _handle_stale_mcp_session( + scope, receive, send, session_manager_stateful + ) + if handled: + # Request was fully handled (e.g., DELETE on non-existent session) + return + session_id = _get_session_id_from_scope(scope) + + if scope.get("method") == "POST": + consumed_messages, body = await _read_request_body_for_routing(receive) + is_initialize = _is_initialize_request(body) + + use_stateful = bool(session_id or is_initialize) + target_manager = ( + session_manager_stateful if use_stateful else session_manager_stateless + ) + + verbose_logger.debug( + f"MCP routing to {'stateful' if use_stateful else 'stateless'} manager" + + (f" (session={session_id[:8]}...)" if session_id else "") + + (" (initialize)" if is_initialize else "") + ) + + # A new `initialize` (no session id) is about to create a stateful + # session. Cap how many a single caller can hold so an authenticated + # client cannot spam `initialize` and exhaust memory. + if is_initialize and not session_id: + request_owner = _owner_fingerprint_for( + user_api_key_auth, oauth2_headers, _client_ip + ) + if not await _enforce_stateful_session_cap_for_owner(request_owner): + verbose_logger.warning( + "Rejecting MCP initialize: caller already holds the maximum " + "number of active stateful sessions." + ) + too_many_response = JSONResponse( + status_code=429, + content={ + "error": "Too Many Requests", + "details": "Too many active MCP sessions for this caller.", + }, + ) + await too_many_response(scope, receive, send) + return + + # Replay body messages if we consumed them for peeking + original_receive = receive + if consumed_messages: + + async def wrapped_receive(): + if consumed_messages: + return consumed_messages.pop(0) + return await original_receive() + + receive = wrapped_receive + + # Serialize requests on the same stateful session so concurrent + # callers don't clobber each other's auth context mid-flight. + # + # Skip the lock for streaming GETs (SSE channels held open for the + # life of the session): holding a per-session lock for a long-lived + # stream would block every subsequent POST on the same session. + # POST/DELETE are the methods that actually mutate the shared + # auth context, so serializing those is sufficient for the + # clobbering race between concurrent JSON-RPC calls. + session_lock: Optional[asyncio.Lock] = None + request_method = (scope.get("method") or "").upper() + if use_stateful and session_id and request_method in ("POST", "DELETE"): + session_lock = _stateful_session_locks.setdefault( + session_id, asyncio.Lock() + ) + + active_request_session_ids: List[str] = [] + + def _increment_active_request_session(session_id_to_track: str) -> None: + if session_id_to_track in active_request_session_ids: + return + active_request_session_ids.append(session_id_to_track) + _stateful_session_active_request_counts[session_id_to_track] = ( + _stateful_session_active_request_counts.get(session_id_to_track, 0) + + 1 + ) + + if use_stateful and session_id: + _increment_active_request_session(session_id) + + def _track_initialized_stateful_session( + initialized_session_id: str, + ) -> None: + _increment_active_request_session(initialized_session_id) + + async def _dispatch() -> None: + auth_user = _set_or_update_auth_context( + user_api_key_auth=user_api_key_auth, + mcp_auth_header=mcp_auth_header, + mcp_servers=mcp_servers, + mcp_server_auth_headers=mcp_server_auth_headers, + oauth2_headers=oauth2_headers, + raw_headers=raw_headers, + client_ip=_client_ip, + session_id=session_id if use_stateful else None, + touch_last_seen=(scope.get("method") or "").upper() != "DELETE", + copy_existing_session_auth_context=is_initialize, + ) + local_send = send + if use_stateful and is_initialize: + local_send = _wrap_send_with_stateful_session_auth_context( + local_send, + auth_user, + _owner_fingerprint_for( + user_api_key_auth, oauth2_headers, _client_ip + ), + _track_initialized_stateful_session, + ) + + async with _gateway_initialize_instructions_request_scope( + user_api_key_auth, + mcp_servers, + _client_ip, + ): + await target_manager.handle_request(scope, receive, local_send) + if use_stateful and session_id and scope.get("method") == "DELETE": + _remove_stateful_session_tracking(session_id) + + try: + if session_lock is not None: + async with session_lock: + await _dispatch() + else: + await _dispatch() + finally: + for active_request_session_id in active_request_session_ids: + active_request_count = ( + _stateful_session_active_request_counts.get( + active_request_session_id, 0 + ) + - 1 + ) + if active_request_count > 0: + _stateful_session_active_request_counts[ + active_request_session_id + ] = active_request_count + else: + _stateful_session_active_request_counts.pop( + active_request_session_id, None + ) + + if ( + scope.get("method") != "DELETE" + and active_request_session_id in _stateful_session_auth_contexts + ): + _stateful_session_auth_context_last_seen[ + active_request_session_id + ] = time.monotonic() + + # Periodic cleanup iterates _stateful_session_auth_context_last_seen, + # so locks for untracked sessions must be dropped here. + if ( + active_request_count <= 0 + and active_request_session_id + not in _stateful_session_auth_contexts + ): + _stateful_session_locks.pop(active_request_session_id, None) except HTTPException: # Re-raise HTTP exceptions to preserve status codes and details raise @@ -2879,7 +3590,6 @@ if MCP_AVAILABLE: verbose_logger.exception(f"Error handling MCP request: {e}") # Try to send a graceful error response for non-HTTP exceptions try: - from starlette.responses import JSONResponse from starlette.status import HTTP_500_INTERNAL_SERVER_ERROR error_response = JSONResponse( @@ -2985,8 +3695,9 @@ if MCP_AVAILABLE: ############ Auth Context Functions #################### ######################################################## - def set_auth_context( - user_api_key_auth: UserAPIKeyAuth, + def _update_auth_context( + auth_user: MCPAuthenticatedUser, + user_api_key_auth: Optional[UserAPIKeyAuth], mcp_auth_header: Optional[str] = None, mcp_servers: Optional[List[str]] = None, mcp_server_auth_headers: Optional[Dict[str, Dict[str, str]]] = None, @@ -2994,6 +3705,23 @@ if MCP_AVAILABLE: raw_headers: Optional[Dict[str, str]] = None, client_ip: Optional[str] = None, ) -> None: + auth_user.user_api_key_auth = user_api_key_auth + auth_user.mcp_auth_header = mcp_auth_header + auth_user.mcp_servers = mcp_servers + auth_user.mcp_server_auth_headers = mcp_server_auth_headers or {} + auth_user.oauth2_headers = oauth2_headers + auth_user.raw_headers = raw_headers + auth_user.client_ip = client_ip + + def set_auth_context( + user_api_key_auth: Optional[UserAPIKeyAuth], + mcp_auth_header: Optional[str] = None, + mcp_servers: Optional[List[str]] = None, + mcp_server_auth_headers: Optional[Dict[str, Dict[str, str]]] = None, + oauth2_headers: Optional[Dict[str, str]] = None, + raw_headers: Optional[Dict[str, str]] = None, + client_ip: Optional[str] = None, + ) -> MCPAuthenticatedUser: """ Set the UserAPIKeyAuth in the auth context variable. @@ -3014,6 +3742,84 @@ if MCP_AVAILABLE: client_ip=client_ip, ) auth_context_var.set(auth_user) + return auth_user + + def _set_or_update_auth_context( + user_api_key_auth: Optional[UserAPIKeyAuth], + mcp_auth_header: Optional[str] = None, + mcp_servers: Optional[List[str]] = None, + mcp_server_auth_headers: Optional[Dict[str, Dict[str, str]]] = None, + oauth2_headers: Optional[Dict[str, str]] = None, + raw_headers: Optional[Dict[str, str]] = None, + client_ip: Optional[str] = None, + session_id: Optional[str] = None, + touch_last_seen: bool = True, + copy_existing_session_auth_context: bool = False, + ) -> MCPAuthenticatedUser: + auth_user = ( + _stateful_session_auth_contexts.get(session_id) if session_id else None + ) + if auth_user is not None and session_id is not None: + if touch_last_seen: + _stateful_session_auth_context_last_seen[session_id] = time.monotonic() + if copy_existing_session_auth_context: + return set_auth_context( + user_api_key_auth=user_api_key_auth, + mcp_auth_header=mcp_auth_header, + mcp_servers=mcp_servers, + mcp_server_auth_headers=mcp_server_auth_headers, + oauth2_headers=oauth2_headers, + raw_headers=raw_headers, + client_ip=client_ip, + ) + _update_auth_context( + auth_user=auth_user, + user_api_key_auth=user_api_key_auth, + mcp_auth_header=mcp_auth_header, + mcp_servers=mcp_servers, + mcp_server_auth_headers=mcp_server_auth_headers, + oauth2_headers=oauth2_headers, + raw_headers=raw_headers, + client_ip=client_ip, + ) + auth_context_var.set(auth_user) + return auth_user + return set_auth_context( + user_api_key_auth=user_api_key_auth, + mcp_auth_header=mcp_auth_header, + mcp_servers=mcp_servers, + mcp_server_auth_headers=mcp_server_auth_headers, + oauth2_headers=oauth2_headers, + raw_headers=raw_headers, + client_ip=client_ip, + ) + + def _wrap_send_with_stateful_session_auth_context( + send: Send, + auth_user: MCPAuthenticatedUser, + owner_fingerprint: str, + on_session_registered: Optional[Callable[[str], None]] = None, + ) -> Send: + async def wrapped_send(message: Message) -> None: + if message.get("type") == "http.response.start": + for key, value in message.get("headers", []): + header_name = key if isinstance(key, bytes) else str(key).encode() + if header_name.lower() == b"mcp-session-id": + session_id = ( + value.decode() if isinstance(value, bytes) else str(value) + ) + if on_session_registered is not None: + on_session_registered(session_id) + auth_context_var.set(auth_user) + _stateful_session_auth_contexts[session_id] = auth_user + _stateful_session_auth_context_last_seen[session_id] = ( + time.monotonic() + ) + _stateful_session_owners[session_id] = owner_fingerprint + break + await send(message) + + return wrapped_send def get_auth_context() -> Tuple[ Optional[UserAPIKeyAuth], diff --git a/litellm/proxy/_experimental/mcp_server/tool_registry.py b/litellm/proxy/_experimental/mcp_server/tool_registry.py index 829be5be979..bb30ff55c5c 100644 --- a/litellm/proxy/_experimental/mcp_server/tool_registry.py +++ b/litellm/proxy/_experimental/mcp_server/tool_registry.py @@ -92,13 +92,20 @@ class MCPToolRegistry: ] def load_tools_from_config( - self, mcp_tools_config: Optional[Dict[str, Any]] = None + self, + mcp_tools_config: Optional[Dict[str, Any]] = None, + config_file_path: Optional[str] = None, ) -> None: """ Load and register tools from the proxy config Args: mcp_tools_config: The mcp_tools config from the proxy config + config_file_path: Path to the operator's config.yaml. Threaded + through to ``get_instance_fn`` so an ``s3://``/``gcs://`` + ``handler`` declared in the YAML resolves; callers from a + non-YAML path must leave this ``None`` so the runtime gate + fires. """ if mcp_tools_config is None: raise ValueError( @@ -121,7 +128,7 @@ class MCPToolRegistry: # First check if it's a module path (e.g., "module.submodule.function") if handler_name is None: raise ValueError(f"handler is required for tool {name}") - handler = get_instance_fn(handler_name) + handler = get_instance_fn(handler_name, config_file_path) if handler is None: verbose_logger.warning( diff --git a/litellm/proxy/_experimental/mcp_server/utils.py b/litellm/proxy/_experimental/mcp_server/utils.py index df5705c3425..b8b9207555e 100644 --- a/litellm/proxy/_experimental/mcp_server/utils.py +++ b/litellm/proxy/_experimental/mcp_server/utils.py @@ -2,7 +2,8 @@ MCP Server Utilities """ -from typing import Any, Dict, Iterator, Mapping, Optional, Tuple +import re +from typing import Any, Dict, Iterator, Mapping, Optional, Tuple, Union import hashlib import importlib @@ -117,6 +118,50 @@ def normalize_server_name(server_name: str) -> str: return server_name.replace(" ", "_") +_MCP_ALIAS_HEADER_INVALID_RE = re.compile(r"[^a-z0-9_]") + + +def sanitize_mcp_alias_for_header(alias: str) -> str: + """ + Sanitize an MCP server alias for x-mcp-{alias}-{header} HTTP headers. + + Must stay in sync with ui/litellm-dashboard/src/utils/mcpHeaderUtils.ts. + """ + sanitized = _MCP_ALIAS_HEADER_INVALID_RE.sub("_", alias.lower().strip()) + sanitized = re.sub(r"_+", "_", sanitized) + return sanitized.strip("_") + + +def lookup_mcp_server_auth_in_headers( + mcp_server_auth_headers: Mapping[str, Union[str, Dict[str, str]]], + *, + alias: Optional[str] = None, + server_name: Optional[str] = None, +) -> Optional[Union[str, Dict[str, str]]]: + """ + Resolve server-specific auth headers with case-insensitive matching. + + Tries the raw alias/server_name (lowercased) and the header-safe sanitized + alias so dashboard clients using sanitize_mcp_alias_for_header() still match. + """ + if not mcp_server_auth_headers: + return None + + normalized_headers = {k.lower(): v for k, v in mcp_server_auth_headers.items()} + + for identifier in (alias, server_name): + if not identifier: + continue + keys_to_try = [identifier.lower()] + sanitized = sanitize_mcp_alias_for_header(identifier) + if sanitized and sanitized not in keys_to_try: + keys_to_try.append(sanitized) + for key in keys_to_try: + if key in normalized_headers: + return normalized_headers[key] + return None + + def validate_and_normalize_mcp_server_payload(payload: Any) -> None: """ Validate and normalize MCP server payload fields (server_name and alias). diff --git a/litellm/proxy/_experimental/out/404.html b/litellm/proxy/_experimental/out/404.html index 17fdb3b3582..f27612ff54e 100644 --- a/litellm/proxy/_experimental/out/404.html +++ b/litellm/proxy/_experimental/out/404.html @@ -1 +1 @@ -404: This page could not be found.LiteLLM Dashboard

404

This page could not be found.

\ No newline at end of file +404: This page could not be found.LiteLLM Dashboard

404

This page could not be found.

\ No newline at end of file diff --git a/litellm/proxy/_experimental/out/404/index.html b/litellm/proxy/_experimental/out/404/index.html new file mode 100644 index 00000000000..f27612ff54e --- /dev/null +++ b/litellm/proxy/_experimental/out/404/index.html @@ -0,0 +1 @@ +404: This page could not be found.LiteLLM Dashboard

404

This page could not be found.

\ No newline at end of file diff --git a/litellm/proxy/_experimental/out/__next.__PAGE__.txt b/litellm/proxy/_experimental/out/__next.__PAGE__.txt index cf19a1edd6c..c024136e8dc 100644 --- a/litellm/proxy/_experimental/out/__next.__PAGE__.txt +++ b/litellm/proxy/_experimental/out/__next.__PAGE__.txt @@ -1,30 +1,10 @@ 1:"$Sreact.fragment" 2:I[347257,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"ClientPageRoot"] -3:I[952683,["/litellm-asset-prefix/_next/static/chunks/9e09de50158b3159.js","/litellm-asset-prefix/_next/static/chunks/7e5fe5584502da06.js","/litellm-asset-prefix/_next/static/chunks/0493aafc4891dd29.js","/litellm-asset-prefix/_next/static/chunks/f7e1d08418645368.js","/litellm-asset-prefix/_next/static/chunks/b3d198d6c56a21b8.js","/litellm-asset-prefix/_next/static/chunks/403c4d96324c23a6.js","/litellm-asset-prefix/_next/static/chunks/37e77c06e99eb8ff.js","/litellm-asset-prefix/_next/static/chunks/adb8beb738574863.js","/litellm-asset-prefix/_next/static/chunks/0549bc9afa7d4888.js","/litellm-asset-prefix/_next/static/chunks/0b470ffc60999bf4.js","/litellm-asset-prefix/_next/static/chunks/c847ecdf8c790b0b.js","/litellm-asset-prefix/_next/static/chunks/baadbd26839e7b66.js","/litellm-asset-prefix/_next/static/chunks/ee5f9a39a526e423.js","/litellm-asset-prefix/_next/static/chunks/6eee262391715440.js","/litellm-asset-prefix/_next/static/chunks/4e17b625d75327a7.js","/litellm-asset-prefix/_next/static/chunks/7b788dd93ad868b3.js","/litellm-asset-prefix/_next/static/chunks/a06cc76a774dd182.js","/litellm-asset-prefix/_next/static/chunks/264fd32eefec52b6.js","/litellm-asset-prefix/_next/static/chunks/86828bdbafb8b581.js","/litellm-asset-prefix/_next/static/chunks/10dc4591ef08a91f.js","/litellm-asset-prefix/_next/static/chunks/e099566e8bd4ee4e.js","/litellm-asset-prefix/_next/static/chunks/fbe12a36d22e9554.js","/litellm-asset-prefix/_next/static/chunks/2971c4658f1bcd7d.js","/litellm-asset-prefix/_next/static/chunks/134f728fa7099e3e.js","/litellm-asset-prefix/_next/static/chunks/679dbd657c8b5aef.js","/litellm-asset-prefix/_next/static/chunks/94f7208f5087e27c.js","/litellm-asset-prefix/_next/static/chunks/43f6fc3c2ab9cf23.js","/litellm-asset-prefix/_next/static/chunks/4e06277331e725da.js","/litellm-asset-prefix/_next/static/chunks/3b30ab8eaa03bc21.js","/litellm-asset-prefix/_next/static/chunks/ac3cf77acb5bf234.js","/litellm-asset-prefix/_next/static/chunks/fb125648f2dae104.js","/litellm-asset-prefix/_next/static/chunks/7e417dd24c8becd0.js","/litellm-asset-prefix/_next/static/chunks/a09028cd611c08ef.js","/litellm-asset-prefix/_next/static/chunks/6967a3b4ecbd3785.js","/litellm-asset-prefix/_next/static/chunks/3e917c79aadd945b.js","/litellm-asset-prefix/_next/static/chunks/9bbebdeb3f1cb03f.js","/litellm-asset-prefix/_next/static/chunks/0a65da2cd24e2ab6.js","/litellm-asset-prefix/_next/static/chunks/908828a91f602d8b.js","/litellm-asset-prefix/_next/static/chunks/5f2d62a75803a3f7.js","/litellm-asset-prefix/_next/static/chunks/ca5fbafaf3826374.js","/litellm-asset-prefix/_next/static/chunks/d3ac82723ec9e30d.js","/litellm-asset-prefix/_next/static/chunks/9b0ee76cbdef1a2a.js","/litellm-asset-prefix/_next/static/chunks/1bc2898be56acd1b.js","/litellm-asset-prefix/_next/static/chunks/fcdf7322b0aa3e2e.js","/litellm-asset-prefix/_next/static/chunks/878832edb30e99a4.js","/litellm-asset-prefix/_next/static/chunks/496b84010c33cf69.js","/litellm-asset-prefix/_next/static/chunks/8e3d0ce9505a304f.js","/litellm-asset-prefix/_next/static/chunks/e1f23fd814ac3500.js","/litellm-asset-prefix/_next/static/chunks/88c74f8b4b20d25a.js","/litellm-asset-prefix/_next/static/chunks/99cf9cf99df5ccfc.js","/litellm-asset-prefix/_next/static/chunks/4980372eaa37b78b.js","/litellm-asset-prefix/_next/static/chunks/0cdfadbcf4b8c9e4.js","/litellm-asset-prefix/_next/static/chunks/8f3bf592254c6c3b.js","/litellm-asset-prefix/_next/static/chunks/8c17e934bd227606.js","/litellm-asset-prefix/_next/static/chunks/b98447395b5d37ef.js"],"default"] -1a:I[897367,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"OutletBoundary"] -1b:"$Sreact.suspense" +3:I[952683,["/litellm-asset-prefix/_next/static/chunks/aefca6f40ea185cd.js","/litellm-asset-prefix/_next/static/chunks/632b4c8e836bd956.js","/litellm-asset-prefix/_next/static/chunks/52c4ecc57f72065e.js","/litellm-asset-prefix/_next/static/chunks/4bb663ff806dc32f.js","/litellm-asset-prefix/_next/static/chunks/ee8f89c672745c59.js","/litellm-asset-prefix/_next/static/chunks/2954392b7a60a6a1.js","/litellm-asset-prefix/_next/static/chunks/04711b0f8ffa7bbd.js","/litellm-asset-prefix/_next/static/chunks/9710770c6333a72f.js","/litellm-asset-prefix/_next/static/chunks/88c74f8b4b20d25a.js","/litellm-asset-prefix/_next/static/chunks/d734cb3d5659b0da.js","/litellm-asset-prefix/_next/static/chunks/a1ef280b7ad5ae6a.js","/litellm-asset-prefix/_next/static/chunks/09c1f51da7e82268.js","/litellm-asset-prefix/_next/static/chunks/542a1a209eb732c6.js","/litellm-asset-prefix/_next/static/chunks/d028f8c28935d281.js","/litellm-asset-prefix/_next/static/chunks/61aa637257592262.js","/litellm-asset-prefix/_next/static/chunks/bf962cd5264be987.js","/litellm-asset-prefix/_next/static/chunks/37f229ef9335f8c3.js","/litellm-asset-prefix/_next/static/chunks/36d1c027ba991a4f.js","/litellm-asset-prefix/_next/static/chunks/786e88f4abdd5c58.js","/litellm-asset-prefix/_next/static/chunks/43c3db1352241a8b.js","/litellm-asset-prefix/_next/static/chunks/31275eb5c6f6332f.js","/litellm-asset-prefix/_next/static/chunks/0ac09b227f50edb4.js","/litellm-asset-prefix/_next/static/chunks/4c848b12d4ecda3d.js","/litellm-asset-prefix/_next/static/chunks/91c828abd7c0aff5.js","/litellm-asset-prefix/_next/static/chunks/08c348f8e09a5cb0.js","/litellm-asset-prefix/_next/static/chunks/7e417dd24c8becd0.js","/litellm-asset-prefix/_next/static/chunks/b323e0ef008e6348.js","/litellm-asset-prefix/_next/static/chunks/4ac3235460262f36.js","/litellm-asset-prefix/_next/static/chunks/51494a4a4b6fc437.js","/litellm-asset-prefix/_next/static/chunks/878832edb30e99a4.js","/litellm-asset-prefix/_next/static/chunks/9f1486622270556b.js"],"default"] +6:I[897367,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"OutletBoundary"] +7:"$Sreact.suspense" :HL["/litellm-asset-prefix/_next/static/chunks/3f3fa56b5786d58c.css","style"] -0:{"buildId":"8TZ2JbOi7SZ6BCj9ScTHW","rsc":["$","$1","c",{"children":[["$","$L2",null,{"Component":"$3","serverProvidedParams":{"searchParams":{},"params":{},"promises":["$@4","$@5"]}}],[["$","link","0",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/3f3fa56b5786d58c.css","precedence":"next"}],["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/0493aafc4891dd29.js","async":true}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/f7e1d08418645368.js","async":true}],["$","script","script-2",{"src":"/litellm-asset-prefix/_next/static/chunks/b3d198d6c56a21b8.js","async":true}],["$","script","script-3",{"src":"/litellm-asset-prefix/_next/static/chunks/403c4d96324c23a6.js","async":true}],["$","script","script-4",{"src":"/litellm-asset-prefix/_next/static/chunks/37e77c06e99eb8ff.js","async":true}],["$","script","script-5",{"src":"/litellm-asset-prefix/_next/static/chunks/adb8beb738574863.js","async":true}],["$","script","script-6",{"src":"/litellm-asset-prefix/_next/static/chunks/0549bc9afa7d4888.js","async":true}],["$","script","script-7",{"src":"/litellm-asset-prefix/_next/static/chunks/0b470ffc60999bf4.js","async":true}],["$","script","script-8",{"src":"/litellm-asset-prefix/_next/static/chunks/c847ecdf8c790b0b.js","async":true}],["$","script","script-9",{"src":"/litellm-asset-prefix/_next/static/chunks/baadbd26839e7b66.js","async":true}],["$","script","script-10",{"src":"/litellm-asset-prefix/_next/static/chunks/ee5f9a39a526e423.js","async":true}],["$","script","script-11",{"src":"/litellm-asset-prefix/_next/static/chunks/6eee262391715440.js","async":true}],["$","script","script-12",{"src":"/litellm-asset-prefix/_next/static/chunks/4e17b625d75327a7.js","async":true}],["$","script","script-13",{"src":"/litellm-asset-prefix/_next/static/chunks/7b788dd93ad868b3.js","async":true}],["$","script","script-14",{"src":"/litellm-asset-prefix/_next/static/chunks/a06cc76a774dd182.js","async":true}],["$","script","script-15",{"src":"/litellm-asset-prefix/_next/static/chunks/264fd32eefec52b6.js","async":true}],["$","script","script-16",{"src":"/litellm-asset-prefix/_next/static/chunks/86828bdbafb8b581.js","async":true}],["$","script","script-17",{"src":"/litellm-asset-prefix/_next/static/chunks/10dc4591ef08a91f.js","async":true}],["$","script","script-18",{"src":"/litellm-asset-prefix/_next/static/chunks/e099566e8bd4ee4e.js","async":true}],["$","script","script-19",{"src":"/litellm-asset-prefix/_next/static/chunks/fbe12a36d22e9554.js","async":true}],["$","script","script-20",{"src":"/litellm-asset-prefix/_next/static/chunks/2971c4658f1bcd7d.js","async":true}],["$","script","script-21",{"src":"/litellm-asset-prefix/_next/static/chunks/134f728fa7099e3e.js","async":true}],["$","script","script-22",{"src":"/litellm-asset-prefix/_next/static/chunks/679dbd657c8b5aef.js","async":true}],["$","script","script-23",{"src":"/litellm-asset-prefix/_next/static/chunks/94f7208f5087e27c.js","async":true}],["$","script","script-24",{"src":"/litellm-asset-prefix/_next/static/chunks/43f6fc3c2ab9cf23.js","async":true}],["$","script","script-25",{"src":"/litellm-asset-prefix/_next/static/chunks/4e06277331e725da.js","async":true}],["$","script","script-26",{"src":"/litellm-asset-prefix/_next/static/chunks/3b30ab8eaa03bc21.js","async":true}],["$","script","script-27",{"src":"/litellm-asset-prefix/_next/static/chunks/ac3cf77acb5bf234.js","async":true}],["$","script","script-28",{"src":"/litellm-asset-prefix/_next/static/chunks/fb125648f2dae104.js","async":true}],["$","script","script-29",{"src":"/litellm-asset-prefix/_next/static/chunks/7e417dd24c8becd0.js","async":true}],["$","script","script-30",{"src":"/litellm-asset-prefix/_next/static/chunks/a09028cd611c08ef.js","async":true}],["$","script","script-31",{"src":"/litellm-asset-prefix/_next/static/chunks/6967a3b4ecbd3785.js","async":true}],["$","script","script-32",{"src":"/litellm-asset-prefix/_next/static/chunks/3e917c79aadd945b.js","async":true}],["$","script","script-33",{"src":"/litellm-asset-prefix/_next/static/chunks/9bbebdeb3f1cb03f.js","async":true}],"$L6","$L7","$L8","$L9","$La","$Lb","$Lc","$Ld","$Le","$Lf","$L10","$L11","$L12","$L13","$L14","$L15","$L16","$L17","$L18"],"$L19"]}],"loading":null,"isPartial":false} +0:{"buildId":"TrcGiQpTupSbDYFFfkFHY","rsc":["$","$1","c",{"children":[["$","$L2",null,{"Component":"$3","serverProvidedParams":{"searchParams":{},"params":{},"promises":["$@4","$@5"]}}],[["$","link","0",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/3f3fa56b5786d58c.css","precedence":"next"}],["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/52c4ecc57f72065e.js","async":true}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/4bb663ff806dc32f.js","async":true}],["$","script","script-2",{"src":"/litellm-asset-prefix/_next/static/chunks/ee8f89c672745c59.js","async":true}],["$","script","script-3",{"src":"/litellm-asset-prefix/_next/static/chunks/2954392b7a60a6a1.js","async":true}],["$","script","script-4",{"src":"/litellm-asset-prefix/_next/static/chunks/04711b0f8ffa7bbd.js","async":true}],["$","script","script-5",{"src":"/litellm-asset-prefix/_next/static/chunks/9710770c6333a72f.js","async":true}],["$","script","script-6",{"src":"/litellm-asset-prefix/_next/static/chunks/88c74f8b4b20d25a.js","async":true}],["$","script","script-7",{"src":"/litellm-asset-prefix/_next/static/chunks/d734cb3d5659b0da.js","async":true}],["$","script","script-8",{"src":"/litellm-asset-prefix/_next/static/chunks/a1ef280b7ad5ae6a.js","async":true}],["$","script","script-9",{"src":"/litellm-asset-prefix/_next/static/chunks/09c1f51da7e82268.js","async":true}],["$","script","script-10",{"src":"/litellm-asset-prefix/_next/static/chunks/542a1a209eb732c6.js","async":true}],["$","script","script-11",{"src":"/litellm-asset-prefix/_next/static/chunks/d028f8c28935d281.js","async":true}],["$","script","script-12",{"src":"/litellm-asset-prefix/_next/static/chunks/61aa637257592262.js","async":true}],["$","script","script-13",{"src":"/litellm-asset-prefix/_next/static/chunks/bf962cd5264be987.js","async":true}],["$","script","script-14",{"src":"/litellm-asset-prefix/_next/static/chunks/37f229ef9335f8c3.js","async":true}],["$","script","script-15",{"src":"/litellm-asset-prefix/_next/static/chunks/36d1c027ba991a4f.js","async":true}],["$","script","script-16",{"src":"/litellm-asset-prefix/_next/static/chunks/786e88f4abdd5c58.js","async":true}],["$","script","script-17",{"src":"/litellm-asset-prefix/_next/static/chunks/43c3db1352241a8b.js","async":true}],["$","script","script-18",{"src":"/litellm-asset-prefix/_next/static/chunks/31275eb5c6f6332f.js","async":true}],["$","script","script-19",{"src":"/litellm-asset-prefix/_next/static/chunks/0ac09b227f50edb4.js","async":true}],["$","script","script-20",{"src":"/litellm-asset-prefix/_next/static/chunks/4c848b12d4ecda3d.js","async":true}],["$","script","script-21",{"src":"/litellm-asset-prefix/_next/static/chunks/91c828abd7c0aff5.js","async":true}],["$","script","script-22",{"src":"/litellm-asset-prefix/_next/static/chunks/08c348f8e09a5cb0.js","async":true}],["$","script","script-23",{"src":"/litellm-asset-prefix/_next/static/chunks/7e417dd24c8becd0.js","async":true}],["$","script","script-24",{"src":"/litellm-asset-prefix/_next/static/chunks/b323e0ef008e6348.js","async":true}],["$","script","script-25",{"src":"/litellm-asset-prefix/_next/static/chunks/4ac3235460262f36.js","async":true}],["$","script","script-26",{"src":"/litellm-asset-prefix/_next/static/chunks/51494a4a4b6fc437.js","async":true}],["$","script","script-27",{"src":"/litellm-asset-prefix/_next/static/chunks/878832edb30e99a4.js","async":true}],["$","script","script-28",{"src":"/litellm-asset-prefix/_next/static/chunks/9f1486622270556b.js","async":true}]],["$","$L6",null,{"children":["$","$7",null,{"name":"Next.MetadataOutlet","children":"$@8"}]}]]}],"loading":null,"isPartial":false} 4:{} 5:"$0:rsc:props:children:0:props:serverProvidedParams:params" -6:["$","script","script-34",{"src":"/litellm-asset-prefix/_next/static/chunks/0a65da2cd24e2ab6.js","async":true}] -7:["$","script","script-35",{"src":"/litellm-asset-prefix/_next/static/chunks/908828a91f602d8b.js","async":true}] -8:["$","script","script-36",{"src":"/litellm-asset-prefix/_next/static/chunks/5f2d62a75803a3f7.js","async":true}] -9:["$","script","script-37",{"src":"/litellm-asset-prefix/_next/static/chunks/ca5fbafaf3826374.js","async":true}] -a:["$","script","script-38",{"src":"/litellm-asset-prefix/_next/static/chunks/d3ac82723ec9e30d.js","async":true}] -b:["$","script","script-39",{"src":"/litellm-asset-prefix/_next/static/chunks/9b0ee76cbdef1a2a.js","async":true}] -c:["$","script","script-40",{"src":"/litellm-asset-prefix/_next/static/chunks/1bc2898be56acd1b.js","async":true}] -d:["$","script","script-41",{"src":"/litellm-asset-prefix/_next/static/chunks/fcdf7322b0aa3e2e.js","async":true}] -e:["$","script","script-42",{"src":"/litellm-asset-prefix/_next/static/chunks/878832edb30e99a4.js","async":true}] -f:["$","script","script-43",{"src":"/litellm-asset-prefix/_next/static/chunks/496b84010c33cf69.js","async":true}] -10:["$","script","script-44",{"src":"/litellm-asset-prefix/_next/static/chunks/8e3d0ce9505a304f.js","async":true}] -11:["$","script","script-45",{"src":"/litellm-asset-prefix/_next/static/chunks/e1f23fd814ac3500.js","async":true}] -12:["$","script","script-46",{"src":"/litellm-asset-prefix/_next/static/chunks/88c74f8b4b20d25a.js","async":true}] -13:["$","script","script-47",{"src":"/litellm-asset-prefix/_next/static/chunks/99cf9cf99df5ccfc.js","async":true}] -14:["$","script","script-48",{"src":"/litellm-asset-prefix/_next/static/chunks/4980372eaa37b78b.js","async":true}] -15:["$","script","script-49",{"src":"/litellm-asset-prefix/_next/static/chunks/0cdfadbcf4b8c9e4.js","async":true}] -16:["$","script","script-50",{"src":"/litellm-asset-prefix/_next/static/chunks/8f3bf592254c6c3b.js","async":true}] -17:["$","script","script-51",{"src":"/litellm-asset-prefix/_next/static/chunks/8c17e934bd227606.js","async":true}] -18:["$","script","script-52",{"src":"/litellm-asset-prefix/_next/static/chunks/b98447395b5d37ef.js","async":true}] -19:["$","$L1a",null,{"children":["$","$1b",null,{"name":"Next.MetadataOutlet","children":"$@1c"}]}] -1c:null +8:null diff --git a/litellm/proxy/_experimental/out/__next._full.txt b/litellm/proxy/_experimental/out/__next._full.txt index 6b75bee8839..0c119086b9e 100644 --- a/litellm/proxy/_experimental/out/__next._full.txt +++ b/litellm/proxy/_experimental/out/__next._full.txt @@ -1,62 +1,39 @@ 1:"$Sreact.fragment" -2:I[867271,["/litellm-asset-prefix/_next/static/chunks/9e09de50158b3159.js","/litellm-asset-prefix/_next/static/chunks/7e5fe5584502da06.js"],"default"] -3:I[71195,["/litellm-asset-prefix/_next/static/chunks/9e09de50158b3159.js","/litellm-asset-prefix/_next/static/chunks/7e5fe5584502da06.js"],"default"] -4:I[339756,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"default"] -5:I[837457,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"default"] -6:I[347257,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"ClientPageRoot"] -7:I[952683,["/litellm-asset-prefix/_next/static/chunks/9e09de50158b3159.js","/litellm-asset-prefix/_next/static/chunks/7e5fe5584502da06.js","/litellm-asset-prefix/_next/static/chunks/0493aafc4891dd29.js","/litellm-asset-prefix/_next/static/chunks/f7e1d08418645368.js","/litellm-asset-prefix/_next/static/chunks/b3d198d6c56a21b8.js","/litellm-asset-prefix/_next/static/chunks/403c4d96324c23a6.js","/litellm-asset-prefix/_next/static/chunks/37e77c06e99eb8ff.js","/litellm-asset-prefix/_next/static/chunks/adb8beb738574863.js","/litellm-asset-prefix/_next/static/chunks/0549bc9afa7d4888.js","/litellm-asset-prefix/_next/static/chunks/0b470ffc60999bf4.js","/litellm-asset-prefix/_next/static/chunks/c847ecdf8c790b0b.js","/litellm-asset-prefix/_next/static/chunks/baadbd26839e7b66.js","/litellm-asset-prefix/_next/static/chunks/ee5f9a39a526e423.js","/litellm-asset-prefix/_next/static/chunks/6eee262391715440.js","/litellm-asset-prefix/_next/static/chunks/4e17b625d75327a7.js","/litellm-asset-prefix/_next/static/chunks/7b788dd93ad868b3.js","/litellm-asset-prefix/_next/static/chunks/a06cc76a774dd182.js","/litellm-asset-prefix/_next/static/chunks/264fd32eefec52b6.js","/litellm-asset-prefix/_next/static/chunks/86828bdbafb8b581.js","/litellm-asset-prefix/_next/static/chunks/10dc4591ef08a91f.js","/litellm-asset-prefix/_next/static/chunks/e099566e8bd4ee4e.js","/litellm-asset-prefix/_next/static/chunks/fbe12a36d22e9554.js","/litellm-asset-prefix/_next/static/chunks/2971c4658f1bcd7d.js","/litellm-asset-prefix/_next/static/chunks/134f728fa7099e3e.js","/litellm-asset-prefix/_next/static/chunks/679dbd657c8b5aef.js","/litellm-asset-prefix/_next/static/chunks/94f7208f5087e27c.js","/litellm-asset-prefix/_next/static/chunks/43f6fc3c2ab9cf23.js","/litellm-asset-prefix/_next/static/chunks/4e06277331e725da.js","/litellm-asset-prefix/_next/static/chunks/3b30ab8eaa03bc21.js","/litellm-asset-prefix/_next/static/chunks/ac3cf77acb5bf234.js","/litellm-asset-prefix/_next/static/chunks/fb125648f2dae104.js","/litellm-asset-prefix/_next/static/chunks/7e417dd24c8becd0.js","/litellm-asset-prefix/_next/static/chunks/a09028cd611c08ef.js","/litellm-asset-prefix/_next/static/chunks/6967a3b4ecbd3785.js","/litellm-asset-prefix/_next/static/chunks/3e917c79aadd945b.js","/litellm-asset-prefix/_next/static/chunks/9bbebdeb3f1cb03f.js","/litellm-asset-prefix/_next/static/chunks/0a65da2cd24e2ab6.js","/litellm-asset-prefix/_next/static/chunks/908828a91f602d8b.js","/litellm-asset-prefix/_next/static/chunks/5f2d62a75803a3f7.js","/litellm-asset-prefix/_next/static/chunks/ca5fbafaf3826374.js","/litellm-asset-prefix/_next/static/chunks/d3ac82723ec9e30d.js","/litellm-asset-prefix/_next/static/chunks/9b0ee76cbdef1a2a.js","/litellm-asset-prefix/_next/static/chunks/1bc2898be56acd1b.js","/litellm-asset-prefix/_next/static/chunks/fcdf7322b0aa3e2e.js","/litellm-asset-prefix/_next/static/chunks/878832edb30e99a4.js","/litellm-asset-prefix/_next/static/chunks/496b84010c33cf69.js","/litellm-asset-prefix/_next/static/chunks/8e3d0ce9505a304f.js","/litellm-asset-prefix/_next/static/chunks/e1f23fd814ac3500.js","/litellm-asset-prefix/_next/static/chunks/88c74f8b4b20d25a.js","/litellm-asset-prefix/_next/static/chunks/99cf9cf99df5ccfc.js","/litellm-asset-prefix/_next/static/chunks/4980372eaa37b78b.js","/litellm-asset-prefix/_next/static/chunks/0cdfadbcf4b8c9e4.js","/litellm-asset-prefix/_next/static/chunks/8f3bf592254c6c3b.js","/litellm-asset-prefix/_next/static/chunks/8c17e934bd227606.js","/litellm-asset-prefix/_next/static/chunks/b98447395b5d37ef.js"],"default"] -31:I[168027,[],"default"] +2:I[867271,["/litellm-asset-prefix/_next/static/chunks/aefca6f40ea185cd.js","/litellm-asset-prefix/_next/static/chunks/632b4c8e836bd956.js"],"default"] +3:I[71195,["/litellm-asset-prefix/_next/static/chunks/aefca6f40ea185cd.js","/litellm-asset-prefix/_next/static/chunks/632b4c8e836bd956.js"],"default"] +4:I[557951,["/litellm-asset-prefix/_next/static/chunks/aefca6f40ea185cd.js","/litellm-asset-prefix/_next/static/chunks/632b4c8e836bd956.js"],"AuthProvider"] +5:I[339756,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"default"] +6:I[837457,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"default"] +7:I[347257,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"ClientPageRoot"] +8:I[952683,["/litellm-asset-prefix/_next/static/chunks/aefca6f40ea185cd.js","/litellm-asset-prefix/_next/static/chunks/632b4c8e836bd956.js","/litellm-asset-prefix/_next/static/chunks/52c4ecc57f72065e.js","/litellm-asset-prefix/_next/static/chunks/4bb663ff806dc32f.js","/litellm-asset-prefix/_next/static/chunks/ee8f89c672745c59.js","/litellm-asset-prefix/_next/static/chunks/2954392b7a60a6a1.js","/litellm-asset-prefix/_next/static/chunks/04711b0f8ffa7bbd.js","/litellm-asset-prefix/_next/static/chunks/9710770c6333a72f.js","/litellm-asset-prefix/_next/static/chunks/88c74f8b4b20d25a.js","/litellm-asset-prefix/_next/static/chunks/d734cb3d5659b0da.js","/litellm-asset-prefix/_next/static/chunks/a1ef280b7ad5ae6a.js","/litellm-asset-prefix/_next/static/chunks/09c1f51da7e82268.js","/litellm-asset-prefix/_next/static/chunks/542a1a209eb732c6.js","/litellm-asset-prefix/_next/static/chunks/d028f8c28935d281.js","/litellm-asset-prefix/_next/static/chunks/61aa637257592262.js","/litellm-asset-prefix/_next/static/chunks/bf962cd5264be987.js","/litellm-asset-prefix/_next/static/chunks/37f229ef9335f8c3.js","/litellm-asset-prefix/_next/static/chunks/36d1c027ba991a4f.js","/litellm-asset-prefix/_next/static/chunks/786e88f4abdd5c58.js","/litellm-asset-prefix/_next/static/chunks/43c3db1352241a8b.js","/litellm-asset-prefix/_next/static/chunks/31275eb5c6f6332f.js","/litellm-asset-prefix/_next/static/chunks/0ac09b227f50edb4.js","/litellm-asset-prefix/_next/static/chunks/4c848b12d4ecda3d.js","/litellm-asset-prefix/_next/static/chunks/91c828abd7c0aff5.js","/litellm-asset-prefix/_next/static/chunks/08c348f8e09a5cb0.js","/litellm-asset-prefix/_next/static/chunks/7e417dd24c8becd0.js","/litellm-asset-prefix/_next/static/chunks/b323e0ef008e6348.js","/litellm-asset-prefix/_next/static/chunks/4ac3235460262f36.js","/litellm-asset-prefix/_next/static/chunks/51494a4a4b6fc437.js","/litellm-asset-prefix/_next/static/chunks/878832edb30e99a4.js","/litellm-asset-prefix/_next/static/chunks/9f1486622270556b.js"],"default"] +1a:I[168027,[],"default"] :HL["/litellm-asset-prefix/_next/static/chunks/4e20891f2fd03463.css","style"] -:HL["/litellm-asset-prefix/_next/static/chunks/91037395c95e366d.css","style"] +:HL["/litellm-asset-prefix/_next/static/chunks/1bcca3c38c9deb02.css","style"] :HL["/litellm-asset-prefix/_next/static/media/83afe278b6a6bb3c-s.p.3a6ba036.woff2","font",{"crossOrigin":"","type":"font/woff2"}] :HL["/litellm-asset-prefix/_next/static/chunks/3f3fa56b5786d58c.css","style"] -0:{"P":null,"b":"8TZ2JbOi7SZ6BCj9ScTHW","c":["",""],"q":"","i":false,"f":[[["",{"children":["__PAGE__",{}]},"$undefined","$undefined",true],[["$","$1","c",{"children":[[["$","link","0",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/4e20891f2fd03463.css","precedence":"next","crossOrigin":"$undefined","nonce":"$undefined"}],["$","link","1",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/91037395c95e366d.css","precedence":"next","crossOrigin":"$undefined","nonce":"$undefined"}],["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/9e09de50158b3159.js","async":true,"nonce":"$undefined"}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/7e5fe5584502da06.js","async":true,"nonce":"$undefined"}]],["$","html",null,{"lang":"en","children":["$","body",null,{"className":"inter_5972bc34-module__OU16Qa__className","children":["$","$L2",null,{"children":["$","$L3",null,{"children":["$","$L4",null,{"parallelRouterKey":"children","error":"$undefined","errorStyles":"$undefined","errorScripts":"$undefined","template":["$","$L5",null,{}],"templateStyles":"$undefined","templateScripts":"$undefined","notFound":[[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":404}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],[]],"forbidden":"$undefined","unauthorized":"$undefined"}]}]}]}]}]]}],{"children":[["$","$1","c",{"children":[["$","$L6",null,{"Component":"$7","serverProvidedParams":{"searchParams":{},"params":{},"promises":["$@8","$@9"]}}],[["$","link","0",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/3f3fa56b5786d58c.css","precedence":"next","crossOrigin":"$undefined","nonce":"$undefined"}],["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/0493aafc4891dd29.js","async":true,"nonce":"$undefined"}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/f7e1d08418645368.js","async":true,"nonce":"$undefined"}],["$","script","script-2",{"src":"/litellm-asset-prefix/_next/static/chunks/b3d198d6c56a21b8.js","async":true,"nonce":"$undefined"}],["$","script","script-3",{"src":"/litellm-asset-prefix/_next/static/chunks/403c4d96324c23a6.js","async":true,"nonce":"$undefined"}],["$","script","script-4",{"src":"/litellm-asset-prefix/_next/static/chunks/37e77c06e99eb8ff.js","async":true,"nonce":"$undefined"}],["$","script","script-5",{"src":"/litellm-asset-prefix/_next/static/chunks/adb8beb738574863.js","async":true,"nonce":"$undefined"}],["$","script","script-6",{"src":"/litellm-asset-prefix/_next/static/chunks/0549bc9afa7d4888.js","async":true,"nonce":"$undefined"}],["$","script","script-7",{"src":"/litellm-asset-prefix/_next/static/chunks/0b470ffc60999bf4.js","async":true,"nonce":"$undefined"}],["$","script","script-8",{"src":"/litellm-asset-prefix/_next/static/chunks/c847ecdf8c790b0b.js","async":true,"nonce":"$undefined"}],["$","script","script-9",{"src":"/litellm-asset-prefix/_next/static/chunks/baadbd26839e7b66.js","async":true,"nonce":"$undefined"}],["$","script","script-10",{"src":"/litellm-asset-prefix/_next/static/chunks/ee5f9a39a526e423.js","async":true,"nonce":"$undefined"}],["$","script","script-11",{"src":"/litellm-asset-prefix/_next/static/chunks/6eee262391715440.js","async":true,"nonce":"$undefined"}],["$","script","script-12",{"src":"/litellm-asset-prefix/_next/static/chunks/4e17b625d75327a7.js","async":true,"nonce":"$undefined"}],["$","script","script-13",{"src":"/litellm-asset-prefix/_next/static/chunks/7b788dd93ad868b3.js","async":true,"nonce":"$undefined"}],["$","script","script-14",{"src":"/litellm-asset-prefix/_next/static/chunks/a06cc76a774dd182.js","async":true,"nonce":"$undefined"}],["$","script","script-15",{"src":"/litellm-asset-prefix/_next/static/chunks/264fd32eefec52b6.js","async":true,"nonce":"$undefined"}],"$La","$Lb","$Lc","$Ld","$Le","$Lf","$L10","$L11","$L12","$L13","$L14","$L15","$L16","$L17","$L18","$L19","$L1a","$L1b","$L1c","$L1d","$L1e","$L1f","$L20","$L21","$L22","$L23","$L24","$L25","$L26","$L27","$L28","$L29","$L2a","$L2b","$L2c","$L2d","$L2e"],"$L2f"]}],{},null,false,false]},null,false,false],"$L30",false]],"m":"$undefined","G":["$31",[]],"S":true} -32:I[897367,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"OutletBoundary"] -33:"$Sreact.suspense" -35:I[897367,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"ViewportBoundary"] -37:I[897367,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"MetadataBoundary"] -a:["$","script","script-16",{"src":"/litellm-asset-prefix/_next/static/chunks/86828bdbafb8b581.js","async":true,"nonce":"$undefined"}] -b:["$","script","script-17",{"src":"/litellm-asset-prefix/_next/static/chunks/10dc4591ef08a91f.js","async":true,"nonce":"$undefined"}] -c:["$","script","script-18",{"src":"/litellm-asset-prefix/_next/static/chunks/e099566e8bd4ee4e.js","async":true,"nonce":"$undefined"}] -d:["$","script","script-19",{"src":"/litellm-asset-prefix/_next/static/chunks/fbe12a36d22e9554.js","async":true,"nonce":"$undefined"}] -e:["$","script","script-20",{"src":"/litellm-asset-prefix/_next/static/chunks/2971c4658f1bcd7d.js","async":true,"nonce":"$undefined"}] -f:["$","script","script-21",{"src":"/litellm-asset-prefix/_next/static/chunks/134f728fa7099e3e.js","async":true,"nonce":"$undefined"}] -10:["$","script","script-22",{"src":"/litellm-asset-prefix/_next/static/chunks/679dbd657c8b5aef.js","async":true,"nonce":"$undefined"}] -11:["$","script","script-23",{"src":"/litellm-asset-prefix/_next/static/chunks/94f7208f5087e27c.js","async":true,"nonce":"$undefined"}] -12:["$","script","script-24",{"src":"/litellm-asset-prefix/_next/static/chunks/43f6fc3c2ab9cf23.js","async":true,"nonce":"$undefined"}] -13:["$","script","script-25",{"src":"/litellm-asset-prefix/_next/static/chunks/4e06277331e725da.js","async":true,"nonce":"$undefined"}] -14:["$","script","script-26",{"src":"/litellm-asset-prefix/_next/static/chunks/3b30ab8eaa03bc21.js","async":true,"nonce":"$undefined"}] -15:["$","script","script-27",{"src":"/litellm-asset-prefix/_next/static/chunks/ac3cf77acb5bf234.js","async":true,"nonce":"$undefined"}] -16:["$","script","script-28",{"src":"/litellm-asset-prefix/_next/static/chunks/fb125648f2dae104.js","async":true,"nonce":"$undefined"}] -17:["$","script","script-29",{"src":"/litellm-asset-prefix/_next/static/chunks/7e417dd24c8becd0.js","async":true,"nonce":"$undefined"}] -18:["$","script","script-30",{"src":"/litellm-asset-prefix/_next/static/chunks/a09028cd611c08ef.js","async":true,"nonce":"$undefined"}] -19:["$","script","script-31",{"src":"/litellm-asset-prefix/_next/static/chunks/6967a3b4ecbd3785.js","async":true,"nonce":"$undefined"}] -1a:["$","script","script-32",{"src":"/litellm-asset-prefix/_next/static/chunks/3e917c79aadd945b.js","async":true,"nonce":"$undefined"}] -1b:["$","script","script-33",{"src":"/litellm-asset-prefix/_next/static/chunks/9bbebdeb3f1cb03f.js","async":true,"nonce":"$undefined"}] -1c:["$","script","script-34",{"src":"/litellm-asset-prefix/_next/static/chunks/0a65da2cd24e2ab6.js","async":true,"nonce":"$undefined"}] -1d:["$","script","script-35",{"src":"/litellm-asset-prefix/_next/static/chunks/908828a91f602d8b.js","async":true,"nonce":"$undefined"}] -1e:["$","script","script-36",{"src":"/litellm-asset-prefix/_next/static/chunks/5f2d62a75803a3f7.js","async":true,"nonce":"$undefined"}] -1f:["$","script","script-37",{"src":"/litellm-asset-prefix/_next/static/chunks/ca5fbafaf3826374.js","async":true,"nonce":"$undefined"}] -20:["$","script","script-38",{"src":"/litellm-asset-prefix/_next/static/chunks/d3ac82723ec9e30d.js","async":true,"nonce":"$undefined"}] -21:["$","script","script-39",{"src":"/litellm-asset-prefix/_next/static/chunks/9b0ee76cbdef1a2a.js","async":true,"nonce":"$undefined"}] -22:["$","script","script-40",{"src":"/litellm-asset-prefix/_next/static/chunks/1bc2898be56acd1b.js","async":true,"nonce":"$undefined"}] -23:["$","script","script-41",{"src":"/litellm-asset-prefix/_next/static/chunks/fcdf7322b0aa3e2e.js","async":true,"nonce":"$undefined"}] -24:["$","script","script-42",{"src":"/litellm-asset-prefix/_next/static/chunks/878832edb30e99a4.js","async":true,"nonce":"$undefined"}] -25:["$","script","script-43",{"src":"/litellm-asset-prefix/_next/static/chunks/496b84010c33cf69.js","async":true,"nonce":"$undefined"}] -26:["$","script","script-44",{"src":"/litellm-asset-prefix/_next/static/chunks/8e3d0ce9505a304f.js","async":true,"nonce":"$undefined"}] -27:["$","script","script-45",{"src":"/litellm-asset-prefix/_next/static/chunks/e1f23fd814ac3500.js","async":true,"nonce":"$undefined"}] -28:["$","script","script-46",{"src":"/litellm-asset-prefix/_next/static/chunks/88c74f8b4b20d25a.js","async":true,"nonce":"$undefined"}] -29:["$","script","script-47",{"src":"/litellm-asset-prefix/_next/static/chunks/99cf9cf99df5ccfc.js","async":true,"nonce":"$undefined"}] -2a:["$","script","script-48",{"src":"/litellm-asset-prefix/_next/static/chunks/4980372eaa37b78b.js","async":true,"nonce":"$undefined"}] -2b:["$","script","script-49",{"src":"/litellm-asset-prefix/_next/static/chunks/0cdfadbcf4b8c9e4.js","async":true,"nonce":"$undefined"}] -2c:["$","script","script-50",{"src":"/litellm-asset-prefix/_next/static/chunks/8f3bf592254c6c3b.js","async":true,"nonce":"$undefined"}] -2d:["$","script","script-51",{"src":"/litellm-asset-prefix/_next/static/chunks/8c17e934bd227606.js","async":true,"nonce":"$undefined"}] -2e:["$","script","script-52",{"src":"/litellm-asset-prefix/_next/static/chunks/b98447395b5d37ef.js","async":true,"nonce":"$undefined"}] -2f:["$","$L32",null,{"children":["$","$33",null,{"name":"Next.MetadataOutlet","children":"$@34"}]}] -30:["$","$1","h",{"children":[null,["$","$L35",null,{"children":"$L36"}],["$","div",null,{"hidden":true,"children":["$","$L37",null,{"children":["$","$33",null,{"name":"Next.Metadata","children":"$L38"}]}]}],["$","meta",null,{"name":"next-size-adjust","content":""}]]}] -8:{} -9:"$0:f:0:1:1:children:0:props:children:0:props:serverProvidedParams:params" -36:[["$","meta","0",{"charSet":"utf-8"}],["$","meta","1",{"name":"viewport","content":"width=device-width, initial-scale=1"}]] -39:I[27201,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"IconMark"] -34:null -38:[["$","title","0",{"children":"LiteLLM Dashboard"}],["$","meta","1",{"name":"description","content":"LiteLLM Proxy Admin UI"}],["$","link","2",{"rel":"icon","href":"/favicon.ico?favicon.1d32c690.ico","sizes":"48x48","type":"image/x-icon"}],["$","link","3",{"rel":"icon","href":"./favicon.ico"}],["$","$L39","4",{}]] +0:{"P":null,"b":"TrcGiQpTupSbDYFFfkFHY","c":["",""],"q":"","i":false,"f":[[["",{"children":["__PAGE__",{}]},"$undefined","$undefined",true],[["$","$1","c",{"children":[[["$","link","0",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/4e20891f2fd03463.css","precedence":"next","crossOrigin":"$undefined","nonce":"$undefined"}],["$","link","1",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/1bcca3c38c9deb02.css","precedence":"next","crossOrigin":"$undefined","nonce":"$undefined"}],["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/aefca6f40ea185cd.js","async":true,"nonce":"$undefined"}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/632b4c8e836bd956.js","async":true,"nonce":"$undefined"}]],["$","html",null,{"lang":"en","children":["$","body",null,{"className":"inter_5972bc34-module__OU16Qa__className","children":["$","$L2",null,{"children":["$","$L3",null,{"children":["$","$L4",null,{"children":["$","$L5",null,{"parallelRouterKey":"children","error":"$undefined","errorStyles":"$undefined","errorScripts":"$undefined","template":["$","$L6",null,{}],"templateStyles":"$undefined","templateScripts":"$undefined","notFound":[[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":404}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],[]],"forbidden":"$undefined","unauthorized":"$undefined"}]}]}]}]}]}]]}],{"children":[["$","$1","c",{"children":[["$","$L7",null,{"Component":"$8","serverProvidedParams":{"searchParams":{},"params":{},"promises":["$@9","$@a"]}}],[["$","link","0",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/3f3fa56b5786d58c.css","precedence":"next","crossOrigin":"$undefined","nonce":"$undefined"}],["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/52c4ecc57f72065e.js","async":true,"nonce":"$undefined"}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/4bb663ff806dc32f.js","async":true,"nonce":"$undefined"}],["$","script","script-2",{"src":"/litellm-asset-prefix/_next/static/chunks/ee8f89c672745c59.js","async":true,"nonce":"$undefined"}],["$","script","script-3",{"src":"/litellm-asset-prefix/_next/static/chunks/2954392b7a60a6a1.js","async":true,"nonce":"$undefined"}],["$","script","script-4",{"src":"/litellm-asset-prefix/_next/static/chunks/04711b0f8ffa7bbd.js","async":true,"nonce":"$undefined"}],["$","script","script-5",{"src":"/litellm-asset-prefix/_next/static/chunks/9710770c6333a72f.js","async":true,"nonce":"$undefined"}],["$","script","script-6",{"src":"/litellm-asset-prefix/_next/static/chunks/88c74f8b4b20d25a.js","async":true,"nonce":"$undefined"}],["$","script","script-7",{"src":"/litellm-asset-prefix/_next/static/chunks/d734cb3d5659b0da.js","async":true,"nonce":"$undefined"}],["$","script","script-8",{"src":"/litellm-asset-prefix/_next/static/chunks/a1ef280b7ad5ae6a.js","async":true,"nonce":"$undefined"}],["$","script","script-9",{"src":"/litellm-asset-prefix/_next/static/chunks/09c1f51da7e82268.js","async":true,"nonce":"$undefined"}],["$","script","script-10",{"src":"/litellm-asset-prefix/_next/static/chunks/542a1a209eb732c6.js","async":true,"nonce":"$undefined"}],["$","script","script-11",{"src":"/litellm-asset-prefix/_next/static/chunks/d028f8c28935d281.js","async":true,"nonce":"$undefined"}],["$","script","script-12",{"src":"/litellm-asset-prefix/_next/static/chunks/61aa637257592262.js","async":true,"nonce":"$undefined"}],["$","script","script-13",{"src":"/litellm-asset-prefix/_next/static/chunks/bf962cd5264be987.js","async":true,"nonce":"$undefined"}],["$","script","script-14",{"src":"/litellm-asset-prefix/_next/static/chunks/37f229ef9335f8c3.js","async":true,"nonce":"$undefined"}],["$","script","script-15",{"src":"/litellm-asset-prefix/_next/static/chunks/36d1c027ba991a4f.js","async":true,"nonce":"$undefined"}],"$Lb","$Lc","$Ld","$Le","$Lf","$L10","$L11","$L12","$L13","$L14","$L15","$L16","$L17"],"$L18"]}],{},null,false,false]},null,false,false],"$L19",false]],"m":"$undefined","G":["$1a",[]],"S":true} +1b:I[897367,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"OutletBoundary"] +1c:"$Sreact.suspense" +1e:I[897367,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"ViewportBoundary"] +20:I[897367,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"MetadataBoundary"] +b:["$","script","script-16",{"src":"/litellm-asset-prefix/_next/static/chunks/786e88f4abdd5c58.js","async":true,"nonce":"$undefined"}] +c:["$","script","script-17",{"src":"/litellm-asset-prefix/_next/static/chunks/43c3db1352241a8b.js","async":true,"nonce":"$undefined"}] +d:["$","script","script-18",{"src":"/litellm-asset-prefix/_next/static/chunks/31275eb5c6f6332f.js","async":true,"nonce":"$undefined"}] +e:["$","script","script-19",{"src":"/litellm-asset-prefix/_next/static/chunks/0ac09b227f50edb4.js","async":true,"nonce":"$undefined"}] +f:["$","script","script-20",{"src":"/litellm-asset-prefix/_next/static/chunks/4c848b12d4ecda3d.js","async":true,"nonce":"$undefined"}] +10:["$","script","script-21",{"src":"/litellm-asset-prefix/_next/static/chunks/91c828abd7c0aff5.js","async":true,"nonce":"$undefined"}] +11:["$","script","script-22",{"src":"/litellm-asset-prefix/_next/static/chunks/08c348f8e09a5cb0.js","async":true,"nonce":"$undefined"}] +12:["$","script","script-23",{"src":"/litellm-asset-prefix/_next/static/chunks/7e417dd24c8becd0.js","async":true,"nonce":"$undefined"}] +13:["$","script","script-24",{"src":"/litellm-asset-prefix/_next/static/chunks/b323e0ef008e6348.js","async":true,"nonce":"$undefined"}] +14:["$","script","script-25",{"src":"/litellm-asset-prefix/_next/static/chunks/4ac3235460262f36.js","async":true,"nonce":"$undefined"}] +15:["$","script","script-26",{"src":"/litellm-asset-prefix/_next/static/chunks/51494a4a4b6fc437.js","async":true,"nonce":"$undefined"}] +16:["$","script","script-27",{"src":"/litellm-asset-prefix/_next/static/chunks/878832edb30e99a4.js","async":true,"nonce":"$undefined"}] +17:["$","script","script-28",{"src":"/litellm-asset-prefix/_next/static/chunks/9f1486622270556b.js","async":true,"nonce":"$undefined"}] +18:["$","$L1b",null,{"children":["$","$1c",null,{"name":"Next.MetadataOutlet","children":"$@1d"}]}] +19:["$","$1","h",{"children":[null,["$","$L1e",null,{"children":"$L1f"}],["$","div",null,{"hidden":true,"children":["$","$L20",null,{"children":["$","$1c",null,{"name":"Next.Metadata","children":"$L21"}]}]}],["$","meta",null,{"name":"next-size-adjust","content":""}]]}] +9:{} +a:"$0:f:0:1:1:children:0:props:children:0:props:serverProvidedParams:params" +1f:[["$","meta","0",{"charSet":"utf-8"}],["$","meta","1",{"name":"viewport","content":"width=device-width, initial-scale=1"}]] +22:I[27201,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"IconMark"] +1d:null +21:[["$","title","0",{"children":"LiteLLM Dashboard"}],["$","meta","1",{"name":"description","content":"LiteLLM Proxy Admin UI"}],["$","link","2",{"rel":"icon","href":"/favicon.ico?favicon.1d32c690.ico","sizes":"48x48","type":"image/x-icon"}],["$","link","3",{"rel":"icon","href":"./favicon.ico"}],["$","$L22","4",{}]] diff --git a/litellm/proxy/_experimental/out/__next._head.txt b/litellm/proxy/_experimental/out/__next._head.txt index 680119eb918..ea6e5095458 100644 --- a/litellm/proxy/_experimental/out/__next._head.txt +++ b/litellm/proxy/_experimental/out/__next._head.txt @@ -3,4 +3,4 @@ 3:I[897367,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"MetadataBoundary"] 4:"$Sreact.suspense" 5:I[27201,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"IconMark"] -0:{"buildId":"8TZ2JbOi7SZ6BCj9ScTHW","rsc":["$","$1","h",{"children":[null,["$","$L2",null,{"children":[["$","meta","0",{"charSet":"utf-8"}],["$","meta","1",{"name":"viewport","content":"width=device-width, initial-scale=1"}]]}],["$","div",null,{"hidden":true,"children":["$","$L3",null,{"children":["$","$4",null,{"name":"Next.Metadata","children":[["$","title","0",{"children":"LiteLLM Dashboard"}],["$","meta","1",{"name":"description","content":"LiteLLM Proxy Admin UI"}],["$","link","2",{"rel":"icon","href":"/favicon.ico?favicon.1d32c690.ico","sizes":"48x48","type":"image/x-icon"}],["$","link","3",{"rel":"icon","href":"./favicon.ico"}],["$","$L5","4",{}]]}]}]}],["$","meta",null,{"name":"next-size-adjust","content":""}]]}],"loading":null,"isPartial":false} +0:{"buildId":"TrcGiQpTupSbDYFFfkFHY","rsc":["$","$1","h",{"children":[null,["$","$L2",null,{"children":[["$","meta","0",{"charSet":"utf-8"}],["$","meta","1",{"name":"viewport","content":"width=device-width, initial-scale=1"}]]}],["$","div",null,{"hidden":true,"children":["$","$L3",null,{"children":["$","$4",null,{"name":"Next.Metadata","children":[["$","title","0",{"children":"LiteLLM Dashboard"}],["$","meta","1",{"name":"description","content":"LiteLLM Proxy Admin UI"}],["$","link","2",{"rel":"icon","href":"/favicon.ico?favicon.1d32c690.ico","sizes":"48x48","type":"image/x-icon"}],["$","link","3",{"rel":"icon","href":"./favicon.ico"}],["$","$L5","4",{}]]}]}]}],["$","meta",null,{"name":"next-size-adjust","content":""}]]}],"loading":null,"isPartial":false} diff --git a/litellm/proxy/_experimental/out/__next._index.txt b/litellm/proxy/_experimental/out/__next._index.txt index f23ca44427c..ebf6d8fec08 100644 --- a/litellm/proxy/_experimental/out/__next._index.txt +++ b/litellm/proxy/_experimental/out/__next._index.txt @@ -1,8 +1,9 @@ 1:"$Sreact.fragment" -2:I[867271,["/litellm-asset-prefix/_next/static/chunks/9e09de50158b3159.js","/litellm-asset-prefix/_next/static/chunks/7e5fe5584502da06.js"],"default"] -3:I[71195,["/litellm-asset-prefix/_next/static/chunks/9e09de50158b3159.js","/litellm-asset-prefix/_next/static/chunks/7e5fe5584502da06.js"],"default"] -4:I[339756,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"default"] -5:I[837457,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"default"] +2:I[867271,["/litellm-asset-prefix/_next/static/chunks/aefca6f40ea185cd.js","/litellm-asset-prefix/_next/static/chunks/632b4c8e836bd956.js"],"default"] +3:I[71195,["/litellm-asset-prefix/_next/static/chunks/aefca6f40ea185cd.js","/litellm-asset-prefix/_next/static/chunks/632b4c8e836bd956.js"],"default"] +4:I[557951,["/litellm-asset-prefix/_next/static/chunks/aefca6f40ea185cd.js","/litellm-asset-prefix/_next/static/chunks/632b4c8e836bd956.js"],"AuthProvider"] +5:I[339756,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"default"] +6:I[837457,["/litellm-asset-prefix/_next/static/chunks/d96012bcfc98706a.js","/litellm-asset-prefix/_next/static/chunks/dbca964212122d58.js"],"default"] :HL["/litellm-asset-prefix/_next/static/chunks/4e20891f2fd03463.css","style"] -:HL["/litellm-asset-prefix/_next/static/chunks/91037395c95e366d.css","style"] -0:{"buildId":"8TZ2JbOi7SZ6BCj9ScTHW","rsc":["$","$1","c",{"children":[[["$","link","0",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/4e20891f2fd03463.css","precedence":"next"}],["$","link","1",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/91037395c95e366d.css","precedence":"next"}],["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/9e09de50158b3159.js","async":true}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/7e5fe5584502da06.js","async":true}]],["$","html",null,{"lang":"en","children":["$","body",null,{"className":"inter_5972bc34-module__OU16Qa__className","children":["$","$L2",null,{"children":["$","$L3",null,{"children":["$","$L4",null,{"parallelRouterKey":"children","template":["$","$L5",null,{}],"notFound":[[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":404}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],[]]}]}]}]}]}]]}],"loading":null,"isPartial":false} +:HL["/litellm-asset-prefix/_next/static/chunks/1bcca3c38c9deb02.css","style"] +0:{"buildId":"TrcGiQpTupSbDYFFfkFHY","rsc":["$","$1","c",{"children":[[["$","link","0",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/4e20891f2fd03463.css","precedence":"next"}],["$","link","1",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/1bcca3c38c9deb02.css","precedence":"next"}],["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/aefca6f40ea185cd.js","async":true}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/632b4c8e836bd956.js","async":true}]],["$","html",null,{"lang":"en","children":["$","body",null,{"className":"inter_5972bc34-module__OU16Qa__className","children":["$","$L2",null,{"children":["$","$L3",null,{"children":["$","$L4",null,{"children":["$","$L5",null,{"parallelRouterKey":"children","template":["$","$L6",null,{}],"notFound":[[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":404}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],[]]}]}]}]}]}]}]]}],"loading":null,"isPartial":false} diff --git a/litellm/proxy/_experimental/out/__next._tree.txt b/litellm/proxy/_experimental/out/__next._tree.txt index 45dbee66029..4a08f4f9e11 100644 --- a/litellm/proxy/_experimental/out/__next._tree.txt +++ b/litellm/proxy/_experimental/out/__next._tree.txt @@ -1,5 +1,5 @@ :HL["/litellm-asset-prefix/_next/static/chunks/4e20891f2fd03463.css","style"] -:HL["/litellm-asset-prefix/_next/static/chunks/91037395c95e366d.css","style"] +:HL["/litellm-asset-prefix/_next/static/chunks/1bcca3c38c9deb02.css","style"] :HL["/litellm-asset-prefix/_next/static/media/83afe278b6a6bb3c-s.p.3a6ba036.woff2","font",{"crossOrigin":"","type":"font/woff2"}] :HL["/litellm-asset-prefix/_next/static/chunks/3f3fa56b5786d58c.css","style"] -0:{"buildId":"8TZ2JbOi7SZ6BCj9ScTHW","tree":{"name":"","paramType":null,"paramKey":"","hasRuntimePrefetch":false,"slots":{"children":{"name":"__PAGE__","paramType":null,"paramKey":"__PAGE__","hasRuntimePrefetch":false,"slots":null,"isRootLayout":false}},"isRootLayout":true},"staleTime":300} +0:{"buildId":"TrcGiQpTupSbDYFFfkFHY","tree":{"name":"","paramType":null,"paramKey":"","hasRuntimePrefetch":false,"slots":{"children":{"name":"__PAGE__","paramType":null,"paramKey":"__PAGE__","hasRuntimePrefetch":false,"slots":null,"isRootLayout":false}},"isRootLayout":true},"staleTime":300} diff --git a/litellm/proxy/_experimental/out/_next/static/8TZ2JbOi7SZ6BCj9ScTHW/_buildManifest.js b/litellm/proxy/_experimental/out/_next/static/TrcGiQpTupSbDYFFfkFHY/_buildManifest.js similarity index 100% rename from litellm/proxy/_experimental/out/_next/static/8TZ2JbOi7SZ6BCj9ScTHW/_buildManifest.js rename to litellm/proxy/_experimental/out/_next/static/TrcGiQpTupSbDYFFfkFHY/_buildManifest.js diff --git a/litellm/proxy/_experimental/out/_next/static/8TZ2JbOi7SZ6BCj9ScTHW/_clientMiddlewareManifest.json b/litellm/proxy/_experimental/out/_next/static/TrcGiQpTupSbDYFFfkFHY/_clientMiddlewareManifest.json similarity index 100% rename from litellm/proxy/_experimental/out/_next/static/8TZ2JbOi7SZ6BCj9ScTHW/_clientMiddlewareManifest.json rename to litellm/proxy/_experimental/out/_next/static/TrcGiQpTupSbDYFFfkFHY/_clientMiddlewareManifest.json diff --git a/litellm/proxy/_experimental/out/_next/static/8TZ2JbOi7SZ6BCj9ScTHW/_ssgManifest.js b/litellm/proxy/_experimental/out/_next/static/TrcGiQpTupSbDYFFfkFHY/_ssgManifest.js similarity index 100% rename from litellm/proxy/_experimental/out/_next/static/8TZ2JbOi7SZ6BCj9ScTHW/_ssgManifest.js rename to litellm/proxy/_experimental/out/_next/static/TrcGiQpTupSbDYFFfkFHY/_ssgManifest.js diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/00ff280cdb7d7ee5.js b/litellm/proxy/_experimental/out/_next/static/chunks/00ff280cdb7d7ee5.js deleted file mode 100644 index ef84e7aadbe..00000000000 --- a/litellm/proxy/_experimental/out/_next/static/chunks/00ff280cdb7d7ee5.js +++ /dev/null @@ -1 +0,0 @@ -(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,829087,397126,229315,343084,953760,e=>{"use strict";e.i(247167);var t=e.i(271645);new WeakMap,new WeakMap;var n='input:not([inert]):not([inert] *),select:not([inert]):not([inert] *),textarea:not([inert]):not([inert] *),a[href]:not([inert]):not([inert] *),button:not([inert]):not([inert] *),[tabindex]:not(slot):not([inert]):not([inert] *),audio[controls]:not([inert]):not([inert] *),video[controls]:not([inert]):not([inert] *),[contenteditable]:not([contenteditable="false"]):not([inert]):not([inert] *),details>summary:first-of-type:not([inert]):not([inert] *),details:not([inert]):not([inert] *)',r="u"typeof window&&void 0!==window.CSS&&"function"==typeof window.CSS.escape)t=r(window.CSS.escape(e.name));else try{t=r(e.name)}catch(e){return console.error("Looks like you have a radio button with a name attribute containing invalid CSS selector characters and need the CSS.escape polyfill: %s",e.message),!1}var o=h(t,e.form);return!o||o===e},v=function(e){return m(e)&&"radio"===e.type&&!g(e)},y=function(e){var t,n,r,o,l,u,a,c=e&&i(e),s=null==(t=c)?void 0:t.host,f=!1;if(c&&c!==e)for(f=!!(null!=(n=s)&&null!=(r=n.ownerDocument)&&r.contains(s)||null!=e&&null!=(o=e.ownerDocument)&&o.contains(e));!f&&s;)f=!!(null!=(u=s=null==(l=c=i(s))?void 0:l.host)&&null!=(a=u.ownerDocument)&&a.contains(s));return f},w=function(e){var t=e.getBoundingClientRect(),n=t.width,r=t.height;return 0===n&&0===r},b=function(e,t){var n=t.displayCheck,r=t.getShadowRoot;if("full-native"===n&&"checkVisibility"in e)return!e.checkVisibility({checkOpacity:!1,opacityProperty:!1,contentVisibilityAuto:!0,visibilityProperty:!0,checkVisibilityCSS:!0});if("hidden"===getComputedStyle(e).visibility)return!0;var l=o.call(e,"details>summary:first-of-type")?e.parentElement:e;if(o.call(l,"details:not([open]) *"))return!0;if(n&&"full"!==n&&"full-native"!==n&&"legacy-full"!==n){if("non-zero-area"===n)return w(e)}else{if("function"==typeof r){for(var u=e;e;){var a=e.parentElement,c=i(e);if(a&&!a.shadowRoot&&!0===r(a))return w(e);e=e.assignedSlot?e.assignedSlot:a||c===e.ownerDocument?a:c.host}e=u}if(y(e))return!e.getClientRects().length;if("legacy-full"!==n)return!0}return!1},x=function(e){if(/^(INPUT|BUTTON|SELECT|TEXTAREA)$/.test(e.tagName))for(var t=e.parentElement;t;){if("FIELDSET"===t.tagName&&t.disabled){for(var n=0;nf(t))&&!!E(e,t)},S=function(e){var t=parseInt(e.getAttribute("tabindex"),10);return!!isNaN(t)||!!(t>=0)},T=function(e){var t=[],n=[];return e.forEach(function(e,r){var o=!!e.scopeParent,i=o?e.scopeParent:e,l=d(i,o),u=o?T(e.candidates):i;0===l?o?t.push.apply(t,u):t.push(i):n.push({documentOrder:r,tabIndex:l,item:e,isScope:o,content:u})}),n.sort(p).reduce(function(e,t){return t.isScope?e.push.apply(e,t.content):e.push(t.content),e},[]).concat(t)},L=function(e,t){return T((t=t||{}).getShadowRoot?c([e],t.includeContainer,{filter:R.bind(null,t),flatten:!1,getShadowRoot:t.getShadowRoot,shadowRootFilter:S}):a(e,t.includeContainer,R.bind(null,t)))},A=function(e,t){if(t=t||{},!e)throw Error("No node provided");return!1!==o.call(e,n)&&R(t,e)};e.s(["isTabbable",()=>A,"tabbable",()=>L],397126);var C=e.i(174080);function P(){return"u">typeof window}function O(e){return M(e)?(e.nodeName||"").toLowerCase():"#document"}function k(e){var t;return(null==e||null==(t=e.ownerDocument)?void 0:t.defaultView)||window}function D(e){var t;return null==(t=(M(e)?e.ownerDocument:e.document)||window.document)?void 0:t.documentElement}function M(e){return!!P()&&(e instanceof Node||e instanceof k(e).Node)}function N(e){return!!P()&&(e instanceof Element||e instanceof k(e).Element)}function F(e){return!!P()&&(e instanceof HTMLElement||e instanceof k(e).HTMLElement)}function I(e){return!(!P()||"u"{try{return e.matches(t)}catch(e){return!1}})}let z=["transform","translate","scale","rotate","perspective"],K=["transform","translate","scale","rotate","perspective","filter"],U=["paint","layout","strict","content"];function X(e){let t=$(),n=N(e)?J(e):e;return z.some(e=>!!n[e]&&"none"!==n[e])||!!n.containerType&&"normal"!==n.containerType||!t&&!!n.backdropFilter&&"none"!==n.backdropFilter||!t&&!!n.filter&&"none"!==n.filter||K.some(e=>(n.willChange||"").includes(e))||U.some(e=>(n.contain||"").includes(e))}function Y(e){let t=Z(e);for(;F(t)&&!G(t);){if(X(t))return t;if(j(t))break;t=Z(t)}return null}function $(){return!("u"J,"getContainingBlock",()=>Y,"getDocumentElement",()=>D,"getFrameElement",()=>et,"getNodeName",()=>O,"getNodeScroll",()=>Q,"getOverflowAncestors",()=>ee,"getParentNode",()=>Z,"getWindow",()=>k,"isContainingBlock",()=>X,"isElement",()=>N,"isHTMLElement",()=>F,"isLastTraversableNode",()=>G,"isOverflowElement",()=>W,"isShadowRoot",()=>I,"isTableElement",()=>V,"isTopLayer",()=>j,"isWebKit",()=>$],229315);let en=["top","right","bottom","left"],er=en.reduce((e,t)=>e.concat(t,t+"-start",t+"-end"),[]),eo=Math.min,ei=Math.max,el=Math.round,eu=Math.floor,ea=e=>({x:e,y:e}),ec={left:"right",right:"left",bottom:"top",top:"bottom"},es={start:"end",end:"start"};function ef(e,t,n){return ei(e,eo(t,n))}function ed(e,t){return"function"==typeof e?e(t):e}function ep(e){return e.split("-")[0]}function em(e){return e.split("-")[1]}function eh(e){return"x"===e?"y":"x"}function eg(e){return"y"===e?"height":"width"}let ev=new Set(["top","bottom"]);function ey(e){return ev.has(ep(e))?"y":"x"}function ew(e){return eh(ey(e))}function eb(e,t,n){void 0===n&&(n=!1);let r=em(e),o=ew(e),i=eg(o),l="x"===o?r===(n?"end":"start")?"right":"left":"start"===r?"bottom":"top";return t.reference[i]>t.floating[i]&&(l=eC(l)),[l,eC(l)]}function ex(e){let t=eC(e);return[eE(e),t,eE(t)]}function eE(e){return e.replace(/start|end/g,e=>es[e])}let eR=["left","right"],eS=["right","left"],eT=["top","bottom"],eL=["bottom","top"];function eA(e,t,n,r){let o=em(e),i=function(e,t,n){switch(e){case"top":case"bottom":if(n)return t?eS:eR;return t?eR:eS;case"left":case"right":return t?eT:eL;default:return[]}}(ep(e),"start"===n,r);return o&&(i=i.map(e=>e+"-"+o),t&&(i=i.concat(i.map(eE)))),i}function eC(e){return e.replace(/left|right|bottom|top/g,e=>ec[e])}function eP(e){return"number"!=typeof e?{top:0,right:0,bottom:0,left:0,...e}:{top:e,right:e,bottom:e,left:e}}function eO(e){let{x:t,y:n,width:r,height:o}=e;return{width:r,height:o,top:n,left:t,right:t+r,bottom:n+o,x:t,y:n}}function ek(e,t,n){let r,{reference:o,floating:i}=e,l=ey(t),u=ew(t),a=eg(u),c=ep(t),s="y"===l,f=o.x+o.width/2-i.width/2,d=o.y+o.height/2-i.height/2,p=o[a]/2-i[a]/2;switch(c){case"top":r={x:f,y:o.y-i.height};break;case"bottom":r={x:f,y:o.y+o.height};break;case"right":r={x:o.x+o.width,y:d};break;case"left":r={x:o.x-i.width,y:d};break;default:r={x:o.x,y:o.y}}switch(em(t)){case"start":r[u]-=p*(n&&s?-1:1);break;case"end":r[u]+=p*(n&&s?-1:1)}return r}async function eD(e,t){var n;void 0===t&&(t={});let{x:r,y:o,platform:i,rects:l,elements:u,strategy:a}=e,{boundary:c="clippingAncestors",rootBoundary:s="viewport",elementContext:f="floating",altBoundary:d=!1,padding:p=0}=ed(t,e),m=eP(p),h=u[d?"floating"===f?"reference":"floating":f],g=eO(await i.getClippingRect({element:null==(n=await (null==i.isElement?void 0:i.isElement(h)))||n?h:h.contextElement||await (null==i.getDocumentElement?void 0:i.getDocumentElement(u.floating)),boundary:c,rootBoundary:s,strategy:a})),v="floating"===f?{x:r,y:o,width:l.floating.width,height:l.floating.height}:l.reference,y=await (null==i.getOffsetParent?void 0:i.getOffsetParent(u.floating)),w=await (null==i.isElement?void 0:i.isElement(y))&&await (null==i.getScale?void 0:i.getScale(y))||{x:1,y:1},b=eO(i.convertOffsetParentRelativeRectToViewportRelativeRect?await i.convertOffsetParentRelativeRectToViewportRelativeRect({elements:u,rect:v,offsetParent:y,strategy:a}):v);return{top:(g.top-b.top+m.top)/w.y,bottom:(b.bottom-g.bottom+m.bottom)/w.y,left:(g.left-b.left+m.left)/w.x,right:(b.right-g.right+m.right)/w.x}}e.s(["clamp",()=>ef,"createCoords",()=>ea,"evaluate",()=>ed,"floor",()=>eu,"getAlignment",()=>em,"getAlignmentAxis",()=>ew,"getAlignmentSides",()=>eb,"getAxisLength",()=>eg,"getExpandedPlacements",()=>ex,"getOppositeAlignmentPlacement",()=>eE,"getOppositeAxis",()=>eh,"getOppositeAxisPlacements",()=>eA,"getOppositePlacement",()=>eC,"getPaddingObject",()=>eP,"getSide",()=>ep,"getSideAxis",()=>ey,"max",()=>ei,"min",()=>eo,"placements",()=>er,"rectToClientRect",()=>eO,"round",()=>el,"sides",()=>en],343084);let eM=async(e,t,n)=>{let{placement:r="bottom",strategy:o="absolute",middleware:i=[],platform:l}=n,u=i.filter(Boolean),a=await (null==l.isRTL?void 0:l.isRTL(t)),c=await l.getElementRects({reference:e,floating:t,strategy:o}),{x:s,y:f}=ek(c,r,a),d=r,p={},m=0;for(let n=0;ne[t]>=0)}function eI(e){let t=eo(...e.map(e=>e.left)),n=eo(...e.map(e=>e.top));return{x:t,y:n,width:ei(...e.map(e=>e.right))-t,height:ei(...e.map(e=>e.bottom))-n}}let eB=new Set(["left","top"]);async function eW(e,t){let{placement:n,platform:r,elements:o}=e,i=await (null==r.isRTL?void 0:r.isRTL(o.floating)),l=ep(n),u=em(n),a="y"===ey(n),c=eB.has(l)?-1:1,s=i&&a?-1:1,f=ed(t,e),{mainAxis:d,crossAxis:p,alignmentAxis:m}="number"==typeof f?{mainAxis:f,crossAxis:0,alignmentAxis:null}:{mainAxis:f.mainAxis||0,crossAxis:f.crossAxis||0,alignmentAxis:f.alignmentAxis};return u&&"number"==typeof m&&(p="end"===u?-1*m:m),a?{x:p*s,y:d*c}:{x:d*c,y:p*s}}function eH(e){let t=J(e),n=parseFloat(t.width)||0,r=parseFloat(t.height)||0,o=F(e),i=o?e.offsetWidth:n,l=o?e.offsetHeight:r,u=el(n)!==i||el(r)!==l;return u&&(n=i,r=l),{width:n,height:r,$:u}}function eV(e){return N(e)?e:e.contextElement}function e_(e){let t=eV(e);if(!F(t))return ea(1);let n=t.getBoundingClientRect(),{width:r,height:o,$:i}=eH(t),l=(i?el(n.width):n.width)/r,u=(i?el(n.height):n.height)/o;return l&&Number.isFinite(l)||(l=1),u&&Number.isFinite(u)||(u=1),{x:l,y:u}}let ej=ea(0);function ez(e){let t=k(e);return $()&&t.visualViewport?{x:t.visualViewport.offsetLeft,y:t.visualViewport.offsetTop}:ej}function eK(e,t,n,r){var o;void 0===t&&(t=!1),void 0===n&&(n=!1);let i=e.getBoundingClientRect(),l=eV(e),u=ea(1);t&&(r?N(r)&&(u=e_(r)):u=e_(e));let a=(void 0===(o=n)&&(o=!1),r&&(!o||r===k(l))&&o)?ez(l):ea(0),c=(i.left+a.x)/u.x,s=(i.top+a.y)/u.y,f=i.width/u.x,d=i.height/u.y;if(l){let e=k(l),t=r&&N(r)?k(r):r,n=e,o=et(n);for(;o&&r&&t!==n;){let e=e_(o),t=o.getBoundingClientRect(),r=J(o),i=t.left+(o.clientLeft+parseFloat(r.paddingLeft))*e.x,l=t.top+(o.clientTop+parseFloat(r.paddingTop))*e.y;c*=e.x,s*=e.y,f*=e.x,d*=e.y,c+=i,s+=l,o=et(n=k(o))}}return eO({width:f,height:d,x:c,y:s})}function eU(e,t){let n=Q(e).scrollLeft;return t?t.left+n:eK(D(e)).left+n}function eX(e,t){let n=e.getBoundingClientRect();return{x:n.left+t.scrollLeft-eU(e,n),y:n.top+t.scrollTop}}let eY=new Set(["absolute","fixed"]);function e$(e,t,n){var r;let o;if("viewport"===t)o=function(e,t){let n=k(e),r=D(e),o=n.visualViewport,i=r.clientWidth,l=r.clientHeight,u=0,a=0;if(o){i=o.width,l=o.height;let e=$();(!e||e&&"fixed"===t)&&(u=o.offsetLeft,a=o.offsetTop)}let c=eU(r);if(c<=0){let e=r.ownerDocument,t=e.body,n=getComputedStyle(t),o="CSS1Compat"===e.compatMode&&parseFloat(n.marginLeft)+parseFloat(n.marginRight)||0,l=Math.abs(r.clientWidth-t.clientWidth-o);l<=25&&(i-=l)}else c<=25&&(i+=c);return{width:i,height:l,x:u,y:a}}(e,n);else if("document"===t){let t,n,i,l,u,a,c;r=D(e),t=D(r),n=Q(r),i=r.ownerDocument.body,l=ei(t.scrollWidth,t.clientWidth,i.scrollWidth,i.clientWidth),u=ei(t.scrollHeight,t.clientHeight,i.scrollHeight,i.clientHeight),a=-n.scrollLeft+eU(r),c=-n.scrollTop,"rtl"===J(i).direction&&(a+=ei(t.clientWidth,i.clientWidth)-l),o={width:l,height:u,x:a,y:c}}else if(N(t)){let e,r,i,l,u,a;r=(e=eK(t,!0,"fixed"===n)).top+t.clientTop,i=e.left+t.clientLeft,l=F(t)?e_(t):ea(1),u=t.clientWidth*l.x,a=t.clientHeight*l.y,o={width:u,height:a,x:i*l.x,y:r*l.y}}else{let n=ez(e);o={x:t.x-n.x,y:t.y-n.y,width:t.width,height:t.height}}return eO(o)}function eq(e){return"static"===J(e).position}function eG(e,t){if(!F(e)||"fixed"===J(e).position)return null;if(t)return t(e);let n=e.offsetParent;return D(e)===n&&(n=n.ownerDocument.body),n}function eJ(e,t){let n=k(e);if(j(e))return n;if(!F(e)){let t=Z(e);for(;t&&!G(t);){if(N(t)&&!eq(t))return t;t=Z(t)}return n}let r=eG(e,t);for(;r&&V(r)&&eq(r);)r=eG(r,t);return r&&G(r)&&eq(r)&&!X(r)?n:r||Y(e)||n}let eQ=async function(e){let t=this.getOffsetParent||eJ,n=this.getDimensions,r=await n(e.floating);return{reference:function(e,t,n){let r=F(t),o=D(t),i="fixed"===n,l=eK(e,!0,i,t),u={scrollLeft:0,scrollTop:0},a=ea(0);if(r||!r&&!i)if(("body"!==O(t)||W(o))&&(u=Q(t)),r){let e=eK(t,!0,i,t);a.x=e.x+t.clientLeft,a.y=e.y+t.clientTop}else o&&(a.x=eU(o));i&&!r&&o&&(a.x=eU(o));let c=!o||r||i?ea(0):eX(o,u);return{x:l.left+u.scrollLeft-a.x-c.x,y:l.top+u.scrollTop-a.y-c.y,width:l.width,height:l.height}}(e.reference,await t(e.floating),e.strategy),floating:{x:0,y:0,width:r.width,height:r.height}}},eZ={convertOffsetParentRelativeRectToViewportRelativeRect:function(e){let{elements:t,rect:n,offsetParent:r,strategy:o}=e,i="fixed"===o,l=D(r),u=!!t&&j(t.floating);if(r===l||u&&i)return n;let a={scrollLeft:0,scrollTop:0},c=ea(1),s=ea(0),f=F(r);if((f||!f&&!i)&&(("body"!==O(r)||W(l))&&(a=Q(r)),F(r))){let e=eK(r);c=e_(r),s.x=e.x+r.clientLeft,s.y=e.y+r.clientTop}let d=!l||f||i?ea(0):eX(l,a);return{width:n.width*c.x,height:n.height*c.y,x:n.x*c.x-a.scrollLeft*c.x+s.x+d.x,y:n.y*c.y-a.scrollTop*c.y+s.y+d.y}},getDocumentElement:D,getClippingRect:function(e){let{element:t,boundary:n,rootBoundary:r,strategy:o}=e,i=[..."clippingAncestors"===n?j(t)?[]:function(e,t){let n=t.get(e);if(n)return n;let r=ee(e,[],!1).filter(e=>N(e)&&"body"!==O(e)),o=null,i="fixed"===J(e).position,l=i?Z(e):e;for(;N(l)&&!G(l);){let t=J(l),n=X(l);n||"fixed"!==t.position||(o=null),(i?!n&&!o:!n&&"static"===t.position&&!!o&&eY.has(o.position)||W(l)&&!n&&function e(t,n){let r=Z(t);return!(r===n||!N(r)||G(r))&&("fixed"===J(r).position||e(r,n))}(e,l))?r=r.filter(e=>e!==l):o=t,l=Z(l)}return t.set(e,r),r}(t,this._c):[].concat(n),r],l=i[0],u=i.reduce((e,n)=>{let r=e$(t,n,o);return e.top=ei(r.top,e.top),e.right=eo(r.right,e.right),e.bottom=eo(r.bottom,e.bottom),e.left=ei(r.left,e.left),e},e$(t,l,o));return{width:u.right-u.left,height:u.bottom-u.top,x:u.left,y:u.top}},getOffsetParent:eJ,getElementRects:eQ,getClientRects:function(e){return Array.from(e.getClientRects())},getDimensions:function(e){let{width:t,height:n}=eH(e);return{width:t,height:n}},getScale:e_,isElement:N,isRTL:function(e){return"rtl"===J(e).direction}};function e0(e,t){return e.x===t.x&&e.y===t.y&&e.width===t.width&&e.height===t.height}function e1(e,t,n,r){let o;void 0===r&&(r={});let{ancestorScroll:i=!0,ancestorResize:l=!0,elementResize:u="function"==typeof ResizeObserver,layoutShift:a="function"==typeof IntersectionObserver,animationFrame:c=!1}=r,s=eV(e),f=i||l?[...s?ee(s):[],...ee(t)]:[];f.forEach(e=>{i&&e.addEventListener("scroll",n,{passive:!0}),l&&e.addEventListener("resize",n)});let d=s&&a?function(e,t){let n,r=null,o=D(e);function i(){var e;clearTimeout(n),null==(e=r)||e.disconnect(),r=null}return!function l(u,a){void 0===u&&(u=!1),void 0===a&&(a=1),i();let c=e.getBoundingClientRect(),{left:s,top:f,width:d,height:p}=c;if(u||t(),!d||!p)return;let m={rootMargin:-eu(f)+"px "+-eu(o.clientWidth-(s+d))+"px "+-eu(o.clientHeight-(f+p))+"px "+-eu(s)+"px",threshold:ei(0,eo(1,a))||1},h=!0;function g(t){let r=t[0].intersectionRatio;if(r!==a){if(!h)return l();r?l(!1,r):n=setTimeout(()=>{l(!1,1e-7)},1e3)}1!==r||e0(c,e.getBoundingClientRect())||l(),h=!1}try{r=new IntersectionObserver(g,{...m,root:o.ownerDocument})}catch(e){r=new IntersectionObserver(g,m)}r.observe(e)}(!0),i}(s,n):null,p=-1,m=null;u&&(m=new ResizeObserver(e=>{let[r]=e;r&&r.target===s&&m&&(m.unobserve(t),cancelAnimationFrame(p),p=requestAnimationFrame(()=>{var e;null==(e=m)||e.observe(t)})),n()}),s&&!c&&m.observe(s),m.observe(t));let h=c?eK(e):null;return c&&function t(){let r=eK(e);h&&!e0(h,r)&&n(),h=r,o=requestAnimationFrame(t)}(),n(),()=>{var e;f.forEach(e=>{i&&e.removeEventListener("scroll",n),l&&e.removeEventListener("resize",n)}),null==d||d(),null==(e=m)||e.disconnect(),m=null,c&&cancelAnimationFrame(o)}}let e2=function(e){return void 0===e&&(e=0),{name:"offset",options:e,async fn(t){var n,r;let{x:o,y:i,placement:l,middlewareData:u}=t,a=await eW(t,e);return l===(null==(n=u.offset)?void 0:n.placement)&&null!=(r=u.arrow)&&r.alignmentOffset?{}:{x:o+a.x,y:i+a.y,data:{...a,placement:l}}}}},e3=function(e){return void 0===e&&(e={}),{name:"autoPlacement",options:e,async fn(t){var n,r,o,i;let{rects:l,middlewareData:u,placement:a,platform:c,elements:s}=t,{crossAxis:f=!1,alignment:d,allowedPlacements:p=er,autoAlignment:m=!0,...h}=ed(e,t),g=void 0!==d||p===er?((i=d||null)?[...p.filter(e=>em(e)===i),...p.filter(e=>em(e)!==i)]:p.filter(e=>ep(e)===e)).filter(e=>!i||em(e)===i||!!m&&eE(e)!==e):p,v=await c.detectOverflow(t,h),y=(null==(n=u.autoPlacement)?void 0:n.index)||0,w=g[y];if(null==w)return{};let b=eb(w,l,await (null==c.isRTL?void 0:c.isRTL(s.floating)));if(a!==w)return{reset:{placement:g[0]}};let x=[v[ep(w)],v[b[0]],v[b[1]]],E=[...(null==(r=u.autoPlacement)?void 0:r.overflows)||[],{placement:w,overflows:x}],R=g[y+1];if(R)return{data:{index:y+1,overflows:E},reset:{placement:R}};let S=E.map(e=>{let t=em(e.placement);return[e.placement,t&&f?e.overflows.slice(0,2).reduce((e,t)=>e+t,0):e.overflows[0],e.overflows]}).sort((e,t)=>e[1]-t[1]),T=(null==(o=S.filter(e=>e[2].slice(0,em(e[0])?2:3).every(e=>e<=0))[0])?void 0:o[0])||S[0][0];return T!==a?{data:{index:y+1,overflows:E},reset:{placement:T}}:{}}}},e5=function(e){return void 0===e&&(e={}),{name:"shift",options:e,async fn(t){let{x:n,y:r,placement:o,platform:i}=t,{mainAxis:l=!0,crossAxis:u=!1,limiter:a={fn:e=>{let{x:t,y:n}=e;return{x:t,y:n}}},...c}=ed(e,t),s={x:n,y:r},f=await i.detectOverflow(t,c),d=ey(ep(o)),p=eh(d),m=s[p],h=s[d];if(l){let e="y"===p?"top":"left",t="y"===p?"bottom":"right",n=m+f[e],r=m-f[t];m=ef(n,m,r)}if(u){let e="y"===d?"top":"left",t="y"===d?"bottom":"right",n=h+f[e],r=h-f[t];h=ef(n,h,r)}let g=a.fn({...t,[p]:m,[d]:h});return{...g,data:{x:g.x-n,y:g.y-r,enabled:{[p]:l,[d]:u}}}}}},e7=function(e){return void 0===e&&(e={}),{name:"flip",options:e,async fn(t){var n,r,o,i,l;let{placement:u,middlewareData:a,rects:c,initialPlacement:s,platform:f,elements:d}=t,{mainAxis:p=!0,crossAxis:m=!0,fallbackPlacements:h,fallbackStrategy:g="bestFit",fallbackAxisSideDirection:v="none",flipAlignment:y=!0,...w}=ed(e,t);if(null!=(n=a.arrow)&&n.alignmentOffset)return{};let b=ep(u),x=ey(s),E=ep(s)===s,R=await (null==f.isRTL?void 0:f.isRTL(d.floating)),S=h||(E||!y?[eC(s)]:ex(s)),T="none"!==v;!h&&T&&S.push(...eA(s,y,v,R));let L=[s,...S],A=await f.detectOverflow(t,w),C=[],P=(null==(r=a.flip)?void 0:r.overflows)||[];if(p&&C.push(A[b]),m){let e=eb(u,c,R);C.push(A[e[0]],A[e[1]])}if(P=[...P,{placement:u,overflows:C}],!C.every(e=>e<=0)){let e=((null==(o=a.flip)?void 0:o.index)||0)+1,t=L[e];if(t&&("alignment"!==m||x===ey(t)||P.every(e=>ey(e.placement)!==x||e.overflows[0]>0)))return{data:{index:e,overflows:P},reset:{placement:t}};let n=null==(i=P.filter(e=>e.overflows[0]<=0).sort((e,t)=>e.overflows[1]-t.overflows[1])[0])?void 0:i.placement;if(!n)switch(g){case"bestFit":{let e=null==(l=P.filter(e=>{if(T){let t=ey(e.placement);return t===x||"y"===t}return!0}).map(e=>[e.placement,e.overflows.filter(e=>e>0).reduce((e,t)=>e+t,0)]).sort((e,t)=>e[1]-t[1])[0])?void 0:l[0];e&&(n=e);break}case"initialPlacement":n=s}if(u!==n)return{reset:{placement:n}}}return{}}}},e4=function(e){return void 0===e&&(e={}),{name:"size",options:e,async fn(t){var n,r;let o,i,{placement:l,rects:u,platform:a,elements:c}=t,{apply:s=()=>{},...f}=ed(e,t),d=await a.detectOverflow(t,f),p=ep(l),m=em(l),h="y"===ey(l),{width:g,height:v}=u.floating;"top"===p||"bottom"===p?(o=p,i=m===(await (null==a.isRTL?void 0:a.isRTL(c.floating))?"start":"end")?"left":"right"):(i=p,o="end"===m?"top":"bottom");let y=v-d.top-d.bottom,w=g-d.left-d.right,b=eo(v-d[o],y),x=eo(g-d[i],w),E=!t.middlewareData.shift,R=b,S=x;if(null!=(n=t.middlewareData.shift)&&n.enabled.x&&(S=w),null!=(r=t.middlewareData.shift)&&r.enabled.y&&(R=y),E&&!m){let e=ei(d.left,0),t=ei(d.right,0),n=ei(d.top,0),r=ei(d.bottom,0);h?S=g-2*(0!==e||0!==t?e+t:ei(d.left,d.right)):R=v-2*(0!==n||0!==r?n+r:ei(d.top,d.bottom))}await s({...t,availableWidth:S,availableHeight:R});let T=await a.getDimensions(c.floating);return g!==T.width||v!==T.height?{reset:{rects:!0}}:{}}}},e9=function(e){return void 0===e&&(e={}),{name:"hide",options:e,async fn(t){let{rects:n,platform:r}=t,{strategy:o="referenceHidden",...i}=ed(e,t);switch(o){case"referenceHidden":{let e=eN(await r.detectOverflow(t,{...i,elementContext:"reference"}),n.reference);return{data:{referenceHiddenOffsets:e,referenceHidden:eF(e)}}}case"escaped":{let e=eN(await r.detectOverflow(t,{...i,altBoundary:!0}),n.floating);return{data:{escapedOffsets:e,escaped:eF(e)}}}default:return{}}}}},e8=e=>({name:"arrow",options:e,async fn(t){let{x:n,y:r,placement:o,rects:i,platform:l,elements:u,middlewareData:a}=t,{element:c,padding:s=0}=ed(e,t)||{};if(null==c)return{};let f=eP(s),d={x:n,y:r},p=ew(o),m=eg(p),h=await l.getDimensions(c),g="y"===p,v=g?"clientHeight":"clientWidth",y=i.reference[m]+i.reference[p]-d[p]-i.floating[m],w=d[p]-i.reference[p],b=await (null==l.getOffsetParent?void 0:l.getOffsetParent(c)),x=b?b[v]:0;x&&await (null==l.isElement?void 0:l.isElement(b))||(x=u.floating[v]||i.floating[m]);let E=x/2-h[m]/2-1,R=eo(f[g?"top":"left"],E),S=eo(f[g?"bottom":"right"],E),T=x-h[m]-S,L=x/2-h[m]/2+(y/2-w/2),A=ef(R,L,T),C=!a.arrow&&null!=em(o)&&L!==A&&i.reference[m]/2-(Le.y-t.y),n=[],r=null;for(let e=0;er.height/2?n.push([o]):n[n.length-1].push(o),r=o}return n.map(e=>eO(eI(e)))}(s),d=eO(eI(s)),p=eP(u),m=await i.getElementRects({reference:{getBoundingClientRect:function(){if(2===f.length&&f[0].left>f[1].right&&null!=a&&null!=c)return f.find(e=>a>e.left-p.left&&ae.top-p.top&&c=2){if("y"===ey(n)){let e=f[0],t=f[f.length-1],r="top"===ep(n),o=e.top,i=t.bottom,l=r?e.left:t.left,u=r?e.right:t.right;return{top:o,bottom:i,left:l,right:u,width:u-l,height:i-o,x:l,y:o}}let e="left"===ep(n),t=ei(...f.map(e=>e.right)),r=eo(...f.map(e=>e.left)),o=f.filter(n=>e?n.left===r:n.right===t),i=o[0].top,l=o[o.length-1].bottom;return{top:i,bottom:l,left:r,right:t,width:t-r,height:l-i,x:r,y:i}}return d}},floating:r.floating,strategy:l});return o.reference.x!==m.reference.x||o.reference.y!==m.reference.y||o.reference.width!==m.reference.width||o.reference.height!==m.reference.height?{reset:{rects:m}}:{}}}},te=function(e){return void 0===e&&(e={}),{options:e,fn(t){let{x:n,y:r,placement:o,rects:i,middlewareData:l}=t,{offset:u=0,mainAxis:a=!0,crossAxis:c=!0}=ed(e,t),s={x:n,y:r},f=ey(o),d=eh(f),p=s[d],m=s[f],h=ed(u,t),g="number"==typeof h?{mainAxis:h,crossAxis:0}:{mainAxis:0,crossAxis:0,...h};if(a){let e="y"===d?"height":"width",t=i.reference[d]-i.floating[e]+g.mainAxis,n=i.reference[d]+i.reference[e]-g.mainAxis;pn&&(p=n)}if(c){var v,y;let e="y"===d?"width":"height",t=eB.has(ep(o)),n=i.reference[f]-i.floating[e]+(t&&(null==(v=l.offset)?void 0:v[f])||0)+(t?0:g.crossAxis),r=i.reference[f]+i.reference[e]+(t?0:(null==(y=l.offset)?void 0:y[f])||0)-(t?g.crossAxis:0);mr&&(m=r)}return{[d]:p,[f]:m}}}},tt=(e,t,n)=>{let r=new Map,o={platform:eZ,...n},i={...o.platform,_c:r};return eM(e,t,{...o,platform:i})};e.s(["arrow",()=>e8,"autoPlacement",()=>e3,"autoUpdate",()=>e1,"computePosition",()=>tt,"detectOverflow",()=>eD,"flip",()=>e7,"hide",()=>e9,"inline",()=>e6,"limitShift",()=>te,"offset",()=>e2,"shift",()=>e5,"size",()=>e4],953760);var tn="u">typeof document?t.useLayoutEffect:t.useEffect;function tr(e,t){let n,r,o;if(e===t)return!0;if(typeof e!=typeof t)return!1;if("function"==typeof e&&e.toString()===t.toString())return!0;if(e&&t&&"object"==typeof e){if(Array.isArray(e)){if((n=e.length)!=t.length)return!1;for(r=n;0!=r--;)if(!tr(e[r],t[r]))return!1;return!0}if((n=(o=Object.keys(e)).length)!==Object.keys(t).length)return!1;for(r=n;0!=r--;)if(!Object.prototype.hasOwnProperty.call(t,o[r]))return!1;for(r=n;0!=r--;){let n=o[r];if(("_owner"!==n||!e.$$typeof)&&!tr(e[n],t[n]))return!1}return!0}return e!=e&&t!=t}function to(e){let n=t.useRef(e);return tn(()=>{n.current=e}),n}var ti="u">typeof document?t.useLayoutEffect:t.useEffect;let tl=!1,tu=0,ta=()=>"floating-ui-"+tu++,tc=t["useId".toString()]||function(){let[e,n]=t.useState(()=>tl?ta():void 0);return ti(()=>{null==e&&n(ta())},[]),t.useEffect(()=>{tl||(tl=!0)},[]),e},ts=t.createContext(null),tf=t.createContext(null),td=()=>{var e;return(null==(e=t.useContext(ts))?void 0:e.id)||null};function tp(e){return(null==e?void 0:e.ownerDocument)||document}function tm(e){return tp(e).defaultView||window}function th(e){return!!e&&e instanceof tm(e).Element}function tg(e){return!!e&&e instanceof tm(e).HTMLElement}function tv(e,t){let n=["mouse","pen"];return t||n.push("",void 0),n.includes(e)}function ty(e){let n=(0,t.useRef)(e);return ti(()=>{n.current=e}),n}let tw="data-floating-ui-safe-polygon";function tb(e,t,n){return n&&!tv(n)?0:"number"==typeof e?e:null==e?void 0:e[t]}let tx=function(e,n){let{enabled:r=!0,delay:o=0,handleClose:i=null,mouseOnly:l=!1,restMs:u=0,move:a=!0}=void 0===n?{}:n,{open:c,onOpenChange:s,dataRef:f,events:d,elements:{domReference:p,floating:m},refs:h}=e,g=t.useContext(tf),v=td(),y=ty(i),w=ty(o),b=t.useRef(),x=t.useRef(),E=t.useRef(),R=t.useRef(),S=t.useRef(!0),T=t.useRef(!1),L=t.useRef(()=>{}),A=t.useCallback(()=>{var e;let t=null==(e=f.current.openEvent)?void 0:e.type;return(null==t?void 0:t.includes("mouse"))&&"mousedown"!==t},[f]);t.useEffect(()=>{if(r)return d.on("dismiss",e),()=>{d.off("dismiss",e)};function e(){clearTimeout(x.current),clearTimeout(R.current),S.current=!0}},[r,d]),t.useEffect(()=>{if(!r||!y.current||!c)return;function e(){A()&&s(!1)}let t=tp(m).documentElement;return t.addEventListener("mouseleave",e),()=>{t.removeEventListener("mouseleave",e)}},[m,c,s,r,y,f,A]);let C=t.useCallback(function(e){void 0===e&&(e=!0);let t=tb(w.current,"close",b.current);t&&!E.current?(clearTimeout(x.current),x.current=setTimeout(()=>s(!1),t)):e&&(clearTimeout(x.current),s(!1))},[w,s]),P=t.useCallback(()=>{L.current(),E.current=void 0},[]),O=t.useCallback(()=>{if(T.current){let e=tp(h.floating.current).body;e.style.pointerEvents="",e.removeAttribute(tw),T.current=!1}},[h]);return t.useEffect(()=>{if(r&&th(p))return c&&p.addEventListener("mouseleave",i),null==m||m.addEventListener("mouseleave",i),a&&p.addEventListener("mousemove",n,{once:!0}),p.addEventListener("mouseenter",n),p.addEventListener("mouseleave",o),()=>{c&&p.removeEventListener("mouseleave",i),null==m||m.removeEventListener("mouseleave",i),a&&p.removeEventListener("mousemove",n),p.removeEventListener("mouseenter",n),p.removeEventListener("mouseleave",o)};function t(){return!!f.current.openEvent&&["click","mousedown"].includes(f.current.openEvent.type)}function n(e){if(clearTimeout(x.current),S.current=!1,l&&!tv(b.current)||u>0&&0===tb(w.current,"open"))return;f.current.openEvent=e;let t=tb(w.current,"open",b.current);t?x.current=setTimeout(()=>{s(!0)},t):s(!0)}function o(n){if(t())return;L.current();let r=tp(m);if(clearTimeout(R.current),y.current){c||clearTimeout(x.current),E.current=y.current({...e,tree:g,x:n.clientX,y:n.clientY,onClose(){O(),P(),C()}});let t=E.current;r.addEventListener("mousemove",t),L.current=()=>{r.removeEventListener("mousemove",t)};return}C()}function i(n){t()||null==y.current||y.current({...e,tree:g,x:n.clientX,y:n.clientY,onClose(){O(),P(),C()}})(n)}},[p,m,r,e,l,u,a,C,P,O,s,c,g,w,y,f]),ti(()=>{var e,t,n;if(r&&c&&null!=(e=y.current)&&e.__options.blockPointerEvents&&A()){let e=tp(m).body;if(e.setAttribute(tw,""),e.style.pointerEvents="none",T.current=!0,th(p)&&m){let e=null==g||null==(t=g.nodesRef.current.find(e=>e.id===v))||null==(n=t.context)?void 0:n.elements.floating;return e&&(e.style.pointerEvents=""),p.style.pointerEvents="auto",m.style.pointerEvents="auto",()=>{p.style.pointerEvents="",m.style.pointerEvents=""}}}},[r,c,v,m,p,g,y,f,A]),ti(()=>{c||(b.current=void 0,P(),O())},[c,P,O]),t.useEffect(()=>()=>{P(),clearTimeout(x.current),clearTimeout(R.current),O()},[r,P,O]),t.useMemo(()=>{if(!r)return{};function e(e){b.current=e.pointerType}return{reference:{onPointerDown:e,onPointerEnter:e,onMouseMove(){c||0===u||(clearTimeout(R.current),R.current=setTimeout(()=>{S.current||s(!0)},u))}},floating:{onMouseEnter(){clearTimeout(x.current)},onMouseLeave(){d.emit("dismiss",{type:"mouseLeave",data:{returnFocus:!1}}),C(!1)}}}},[d,r,u,c,s,C])};function tE(e,t){if(!e||!t)return!1;let n=t.getRootNode&&t.getRootNode();if(e.contains(t))return!0;if(n&&function(e){if("u"{var n;return e.parentId===t&&(null==(n=e.context)?void 0:n.open)})||[],r=n;for(;r.length;)r=e.filter(e=>{var t;return null==(t=r)?void 0:t.some(t=>{var n;return e.parentId===t.id&&(null==(n=e.context)?void 0:n.open)})})||[],n=n.concat(r);return n}let tS=t["useInsertionEffect".toString()]||(e=>e());function tT(e){let n=t.useRef(()=>{});return tS(()=>{n.current=e}),t.useCallback(function(){for(var e=arguments.length,t=Array(e),r=0;r!1),E="function"==typeof p?x:p,R=t.useRef(!1),{escapeKeyBubbles:S,outsidePressBubbles:T}=tP(y);return t.useEffect(()=>{if(!r||!f)return;function e(e){if("Escape"===e.key){let e=w?tR(w.nodesRef.current,l):[];if(e.length>0){let t=!0;if(e.forEach(e=>{var n;if(null!=(n=e.context)&&n.open&&!e.context.dataRef.current.__escapeKeyBubbles){t=!1;return}}),!t)return}i.emit("dismiss",{type:"escapeKey",data:{returnFocus:{preventScroll:!1}}}),o(!1)}}function t(e){var t;let n=R.current;if(R.current=!1,n||"function"==typeof E&&!E(e))return;let r="composedPath"in e?e.composedPath()[0]:e.target;if(tg(r)&&c){let t=c.ownerDocument.defaultView||window,n=r.scrollWidth>r.clientWidth,o=r.scrollHeight>r.clientHeight,i=o&&e.offsetX>r.clientWidth;if(o&&"rtl"===t.getComputedStyle(r).direction&&(i=e.offsetX<=r.offsetWidth-r.clientWidth),i||n&&e.offsetY>r.clientHeight)return}let u=w&&tR(w.nodesRef.current,l).some(t=>{var n;return tL(e,null==(n=t.context)?void 0:n.elements.floating)});if(tL(e,c)||tL(e,a)||u)return;let s=w?tR(w.nodesRef.current,l):[];if(s.length>0){let e=!0;if(s.forEach(t=>{var n;if(null!=(n=t.context)&&n.open&&!t.context.dataRef.current.__outsidePressBubbles){e=!1;return}}),!e)return}i.emit("dismiss",{type:"outsidePress",data:{returnFocus:b?{preventScroll:!0}:function(e){let t,n;if(0===e.mozInputSource&&e.isTrusted)return!0;let r=/Android/i;return(r.test(null!=(n=navigator.userAgentData)&&n.platform?n.platform:navigator.platform)||r.test((t=navigator.userAgentData)&&Array.isArray(t.brands)?t.brands.map(e=>{let{brand:t,version:n}=e;return t+"/"+n}).join(" "):navigator.userAgent))&&e.pointerType?"click"===e.type&&1===e.buttons:0===e.detail&&!e.pointerType}(e)||0===(t=e).width&&0===t.height||1===t.width&&1===t.height&&0===t.pressure&&0===t.detail&&"mouse"!==t.pointerType||t.width<1&&t.height<1&&0===t.pressure&&0===t.detail}}),o(!1)}function n(){o(!1)}s.current.__escapeKeyBubbles=S,s.current.__outsidePressBubbles=T;let p=tp(c);d&&p.addEventListener("keydown",e),E&&p.addEventListener(m,t);let h=[];return v&&(th(a)&&(h=ee(a)),th(c)&&(h=h.concat(ee(c))),!th(u)&&u&&u.contextElement&&(h=h.concat(ee(u.contextElement)))),(h=h.filter(e=>{var t;return e!==(null==(t=p.defaultView)?void 0:t.visualViewport)})).forEach(e=>{e.addEventListener("scroll",n,{passive:!0})}),()=>{d&&p.removeEventListener("keydown",e),E&&p.removeEventListener(m,t),h.forEach(e=>{e.removeEventListener("scroll",n)})}},[s,c,a,u,d,E,m,i,w,l,r,o,v,f,S,T,b]),t.useEffect(()=>{R.current=!1},[E,m]),t.useMemo(()=>f?{reference:{[tA[g]]:()=>{h&&(i.emit("dismiss",{type:"referencePress",data:{returnFocus:!1}}),o(!1))}},floating:{[tC[m]]:()=>{R.current=!0}}}:{},[f,i,h,m,g,o])},tk=function(e,n){let{open:r,onOpenChange:o,dataRef:i,events:l,refs:u,elements:{floating:a,domReference:c}}=e,{enabled:s=!0,keyboardOnly:f=!0}=void 0===n?{}:n,d=t.useRef(""),p=t.useRef(!1),m=t.useRef();return t.useEffect(()=>{if(!s)return;let e=tp(a).defaultView||window;function t(){!r&&tg(c)&&c===function(e){let t=e.activeElement;for(;(null==(n=t)||null==(r=n.shadowRoot)?void 0:r.activeElement)!=null;){var n,r;t=t.shadowRoot.activeElement}return t}(tp(c))&&(p.current=!0)}return e.addEventListener("blur",t),()=>{e.removeEventListener("blur",t)}},[a,c,r,s]),t.useEffect(()=>{if(s)return l.on("dismiss",e),()=>{l.off("dismiss",e)};function e(e){("referencePress"===e.type||"escapeKey"===e.type)&&(p.current=!0)}},[l,s]),t.useEffect(()=>()=>{clearTimeout(m.current)},[]),t.useMemo(()=>s?{reference:{onPointerDown(e){let{pointerType:t}=e;d.current=t,p.current=!!(t&&f)},onMouseLeave(){p.current=!1},onFocus(e){var t;p.current||"focus"===e.type&&(null==(t=i.current.openEvent)?void 0:t.type)==="mousedown"&&i.current.openEvent&&tL(i.current.openEvent,c)||(i.current.openEvent=e.nativeEvent,o(!0))},onBlur(e){p.current=!1;let t=e.relatedTarget,n=th(t)&&t.hasAttribute("data-floating-ui-focus-guard")&&"outside"===t.getAttribute("data-type");m.current=setTimeout(()=>{tE(u.floating.current,t)||tE(c,t)||n||o(!1)})}}}:{},[s,f,c,u,i,o])},tD=function(e,n){let{open:r}=e,{enabled:o=!0,role:i="dialog"}=void 0===n?{}:n,l=tc(),u=tc();return t.useMemo(()=>{let e={id:l,role:i};return o?"tooltip"===i?{reference:{"aria-describedby":r?l:void 0},floating:e}:{reference:{"aria-expanded":r?"true":"false","aria-haspopup":"alertdialog"===i?"dialog":i,"aria-controls":r?l:void 0,..."listbox"===i&&{role:"combobox"},..."menu"===i&&{id:u}},floating:{...e,..."menu"===i&&{"aria-labelledby":u}}}:{}},[o,i,r,l,u])};function tM(e,t,n){let r=new Map;return{..."floating"===n&&{tabIndex:-1},...e,...t.map(e=>e?e[n]:null).concat(e).reduce((e,t)=>(t&&Object.entries(t).forEach(t=>{let[n,o]=t;if(0===n.indexOf("on")){if(r.has(n)||r.set(n,[]),"function"==typeof o){var i;null==(i=r.get(n))||i.push(o),e[n]=function(){for(var e,t=arguments.length,o=Array(t),i=0;ie(...o))}}}else e[n]=o}),e),{})}}let tN=function(e){void 0===e&&(e=[]);let n=e,r=t.useCallback(t=>tM(t,e,"reference"),n),o=t.useCallback(t=>tM(t,e,"floating"),n),i=t.useCallback(t=>tM(t,e,"item"),e.map(e=>null==e?void 0:e.item));return t.useMemo(()=>({getReferenceProps:r,getFloatingProps:o,getItemProps:i}),[r,o,i])};var tF=e.i(444755);let tI=e=>{let[n,r]=(0,t.useState)(!1),[o,i]=(0,t.useState)(),{x:l,y:u,refs:a,strategy:c,context:s}=function(e){void 0===e&&(e={});let{open:n=!1,onOpenChange:r,nodeId:o}=e,i=function(e){void 0===e&&(e={});let{placement:n="bottom",strategy:r="absolute",middleware:o=[],platform:i,whileElementsMounted:l,open:u}=e,[a,c]=t.useState({x:null,y:null,strategy:r,placement:n,middlewareData:{},isPositioned:!1}),[s,f]=t.useState(o);tr(s,o)||f(o);let d=t.useRef(null),p=t.useRef(null),m=t.useRef(a),h=to(l),g=to(i),[v,y]=t.useState(null),[w,b]=t.useState(null),x=t.useCallback(e=>{d.current!==e&&(d.current=e,y(e))},[]),E=t.useCallback(e=>{p.current!==e&&(p.current=e,b(e))},[]),R=t.useCallback(()=>{if(!d.current||!p.current)return;let e={placement:n,strategy:r,middleware:s};g.current&&(e.platform=g.current),tt(d.current,p.current,e).then(e=>{let t={...e,isPositioned:!0};S.current&&!tr(m.current,t)&&(m.current=t,C.flushSync(()=>{c(t)}))})},[s,n,r,g]);tn(()=>{!1===u&&m.current.isPositioned&&(m.current.isPositioned=!1,c(e=>({...e,isPositioned:!1})))},[u]);let S=t.useRef(!1);tn(()=>(S.current=!0,()=>{S.current=!1}),[]),tn(()=>{if(v&&w)if(h.current)return h.current(v,w,R);else R()},[v,w,R,h]);let T=t.useMemo(()=>({reference:d,floating:p,setReference:x,setFloating:E}),[x,E]),L=t.useMemo(()=>({reference:v,floating:w}),[v,w]);return t.useMemo(()=>({...a,update:R,refs:T,elements:L,reference:x,floating:E}),[a,R,T,L,x,E])}(e),l=t.useContext(tf),u=t.useRef(null),a=t.useRef({}),c=t.useState(()=>{let e;return e=new Map,{emit(t,n){var r;null==(r=e.get(t))||r.forEach(e=>e(n))},on(t,n){e.set(t,[...e.get(t)||[],n])},off(t,n){e.set(t,(e.get(t)||[]).filter(e=>e!==n))}}})[0],[s,f]=t.useState(null),d=t.useCallback(e=>{let t=th(e)?{getBoundingClientRect:()=>e.getBoundingClientRect(),contextElement:e}:e;i.refs.setReference(t)},[i.refs]),p=t.useCallback(e=>{(th(e)||null===e)&&(u.current=e,f(e)),(th(i.refs.reference.current)||null===i.refs.reference.current||null!==e&&!th(e))&&i.refs.setReference(e)},[i.refs]),m=t.useMemo(()=>({...i.refs,setReference:p,setPositionReference:d,domReference:u}),[i.refs,p,d]),h=t.useMemo(()=>({...i.elements,domReference:s}),[i.elements,s]),g=tT(r),v=t.useMemo(()=>({...i,refs:m,elements:h,dataRef:a,nodeId:o,events:c,open:n,onOpenChange:g}),[i,o,c,n,g,m,h]);return ti(()=>{let e=null==l?void 0:l.nodesRef.current.find(e=>e.id===o);e&&(e.context=v)}),t.useMemo(()=>({...i,context:v,refs:m,reference:p,positionReference:d}),[i,m,v,p,d])}({open:n,onOpenChange:t=>{t&&e?i(setTimeout(()=>{r(t)},e)):(clearTimeout(o),r(t))},placement:"top",whileElementsMounted:e1,middleware:[e2(5),e7({fallbackAxisSideDirection:"start"}),e5()]}),{getReferenceProps:f,getFloatingProps:d}=tN([tx(s,{move:!1}),tk(s),tO(s),tD(s,{role:"tooltip"})]);return{tooltipProps:{open:n,x:l,y:u,refs:a,strategy:c,getFloatingProps:d},getReferenceProps:f}},tB=({text:e,open:n,x:r,y:o,refs:i,strategy:l,getFloatingProps:u})=>n&&e?t.default.createElement("div",Object.assign({className:(0,tF.tremorTwMerge)("max-w-xs text-sm z-20 rounded-tremor-default opacity-100 px-2.5 py-1","text-white bg-tremor-background-emphasis","dark:text-tremor-content-emphasis dark:bg-white"),ref:i.setFloating,style:{position:l,top:null!=o?o:0,left:null!=r?r:0}},u()),e):null;tB.displayName="Tooltip",e.s(["default",()=>tB,"useTooltip",()=>tI],829087)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/018293fccad2eeda.js b/litellm/proxy/_experimental/out/_next/static/chunks/018293fccad2eeda.js new file mode 100644 index 00000000000..6ef0d01f2bd --- /dev/null +++ b/litellm/proxy/_experimental/out/_next/static/chunks/018293fccad2eeda.js @@ -0,0 +1 @@ +(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,482725,244451,e=>{"use strict";let t;e.i(247167);var n=e.i(271645),r=e.i(343794),i=e.i(242064),o=e.i(763731),s=e.i(174428);let a=80*Math.PI,l=e=>{let{dotClassName:t,style:i,hasCircleCls:o}=e;return n.createElement("circle",{className:(0,r.default)(`${t}-circle`,{[`${t}-circle-bg`]:o}),r:40,cx:50,cy:50,strokeWidth:20,style:i})},c=({percent:e,prefixCls:t})=>{let i=`${t}-dot`,o=`${i}-holder`,c=`${o}-hidden`,[u,d]=n.useState(!1);(0,s.default)(()=>{0!==e&&d(!0)},[0!==e]);let f=Math.max(Math.min(e,100),0);if(!u)return null;let h={strokeDashoffset:`${a/4}`,strokeDasharray:`${a*f/100} ${a*(100-f)/100}`};return n.createElement("span",{className:(0,r.default)(o,`${i}-progress`,f<=0&&c)},n.createElement("svg",{viewBox:"0 0 100 100",role:"progressbar","aria-valuemin":0,"aria-valuemax":100,"aria-valuenow":f},n.createElement(l,{dotClassName:i,hasCircleCls:!0}),n.createElement(l,{dotClassName:i,style:h})))};function u(e){let{prefixCls:t,percent:i=0}=e,o=`${t}-dot`,s=`${o}-holder`,a=`${s}-hidden`;return n.createElement(n.Fragment,null,n.createElement("span",{className:(0,r.default)(s,i>0&&a)},n.createElement("span",{className:(0,r.default)(o,`${t}-dot-spin`)},[1,2,3,4].map(e=>n.createElement("i",{className:`${t}-dot-item`,key:e})))),n.createElement(c,{prefixCls:t,percent:i}))}function d(e){var t;let{prefixCls:i,indicator:s,percent:a}=e,l=`${i}-dot`;return s&&n.isValidElement(s)?(0,o.cloneElement)(s,{className:(0,r.default)(null==(t=s.props)?void 0:t.className,l),percent:a}):n.createElement(u,{prefixCls:i,percent:a})}e.i(296059);var f=e.i(694758),h=e.i(183293),m=e.i(246422),p=e.i(838378);let v=new f.Keyframes("antSpinMove",{to:{opacity:1}}),g=new f.Keyframes("antRotate",{to:{transform:"rotate(405deg)"}}),y=(0,m.genStyleHooks)("Spin",e=>(e=>{let{componentCls:t,calc:n}=e;return{[t]:Object.assign(Object.assign({},(0,h.resetComponent)(e)),{position:"absolute",display:"none",color:e.colorPrimary,fontSize:0,textAlign:"center",verticalAlign:"middle",opacity:0,transition:`transform ${e.motionDurationSlow} ${e.motionEaseInOutCirc}`,"&-spinning":{position:"relative",display:"inline-block",opacity:1},[`${t}-text`]:{fontSize:e.fontSize,paddingTop:n(n(e.dotSize).sub(e.fontSize)).div(2).add(2).equal()},"&-fullscreen":{position:"fixed",width:"100vw",height:"100vh",backgroundColor:e.colorBgMask,zIndex:e.zIndexPopupBase,inset:0,display:"flex",alignItems:"center",flexDirection:"column",justifyContent:"center",opacity:0,visibility:"hidden",transition:`all ${e.motionDurationMid}`,"&-show":{opacity:1,visibility:"visible"},[t]:{[`${t}-dot-holder`]:{color:e.colorWhite},[`${t}-text`]:{color:e.colorTextLightSolid}}},"&-nested-loading":{position:"relative",[`> div > ${t}`]:{position:"absolute",top:0,insetInlineStart:0,zIndex:4,display:"block",width:"100%",height:"100%",maxHeight:e.contentHeight,[`${t}-dot`]:{position:"absolute",top:"50%",insetInlineStart:"50%",margin:n(e.dotSize).mul(-1).div(2).equal()},[`${t}-text`]:{position:"absolute",top:"50%",width:"100%",textShadow:`0 1px 2px ${e.colorBgContainer}`},[`&${t}-show-text ${t}-dot`]:{marginTop:n(e.dotSize).div(2).mul(-1).sub(10).equal()},"&-sm":{[`${t}-dot`]:{margin:n(e.dotSizeSM).mul(-1).div(2).equal()},[`${t}-text`]:{paddingTop:n(n(e.dotSizeSM).sub(e.fontSize)).div(2).add(2).equal()},[`&${t}-show-text ${t}-dot`]:{marginTop:n(e.dotSizeSM).div(2).mul(-1).sub(10).equal()}},"&-lg":{[`${t}-dot`]:{margin:n(e.dotSizeLG).mul(-1).div(2).equal()},[`${t}-text`]:{paddingTop:n(n(e.dotSizeLG).sub(e.fontSize)).div(2).add(2).equal()},[`&${t}-show-text ${t}-dot`]:{marginTop:n(e.dotSizeLG).div(2).mul(-1).sub(10).equal()}}},[`${t}-container`]:{position:"relative",transition:`opacity ${e.motionDurationSlow}`,"&::after":{position:"absolute",top:0,insetInlineEnd:0,bottom:0,insetInlineStart:0,zIndex:10,width:"100%",height:"100%",background:e.colorBgContainer,opacity:0,transition:`all ${e.motionDurationSlow}`,content:'""',pointerEvents:"none"}},[`${t}-blur`]:{clear:"both",opacity:.5,userSelect:"none",pointerEvents:"none","&::after":{opacity:.4,pointerEvents:"auto"}}},"&-tip":{color:e.spinDotDefault},[`${t}-dot-holder`]:{width:"1em",height:"1em",fontSize:e.dotSize,display:"inline-block",transition:`transform ${e.motionDurationSlow} ease, opacity ${e.motionDurationSlow} ease`,transformOrigin:"50% 50%",lineHeight:1,color:e.colorPrimary,"&-hidden":{transform:"scale(0.3)",opacity:0}},[`${t}-dot-progress`]:{position:"absolute",inset:0},[`${t}-dot`]:{position:"relative",display:"inline-block",fontSize:e.dotSize,width:"1em",height:"1em","&-item":{position:"absolute",display:"block",width:n(e.dotSize).sub(n(e.marginXXS).div(2)).div(2).equal(),height:n(e.dotSize).sub(n(e.marginXXS).div(2)).div(2).equal(),background:"currentColor",borderRadius:"100%",transform:"scale(0.75)",transformOrigin:"50% 50%",opacity:.3,animationName:v,animationDuration:"1s",animationIterationCount:"infinite",animationTimingFunction:"linear",animationDirection:"alternate","&:nth-child(1)":{top:0,insetInlineStart:0,animationDelay:"0s"},"&:nth-child(2)":{top:0,insetInlineEnd:0,animationDelay:"0.4s"},"&:nth-child(3)":{insetInlineEnd:0,bottom:0,animationDelay:"0.8s"},"&:nth-child(4)":{bottom:0,insetInlineStart:0,animationDelay:"1.2s"}},"&-spin":{transform:"rotate(45deg)",animationName:g,animationDuration:"1.2s",animationIterationCount:"infinite",animationTimingFunction:"linear"},"&-circle":{strokeLinecap:"round",transition:["stroke-dashoffset","stroke-dasharray","stroke","stroke-width","opacity"].map(t=>`${t} ${e.motionDurationSlow} ease`).join(","),fillOpacity:0,stroke:"currentcolor"},"&-circle-bg":{stroke:e.colorFillSecondary}},[`&-sm ${t}-dot`]:{"&, &-holder":{fontSize:e.dotSizeSM}},[`&-sm ${t}-dot-holder`]:{i:{width:n(n(e.dotSizeSM).sub(n(e.marginXXS).div(2))).div(2).equal(),height:n(n(e.dotSizeSM).sub(n(e.marginXXS).div(2))).div(2).equal()}},[`&-lg ${t}-dot`]:{"&, &-holder":{fontSize:e.dotSizeLG}},[`&-lg ${t}-dot-holder`]:{i:{width:n(n(e.dotSizeLG).sub(e.marginXXS)).div(2).equal(),height:n(n(e.dotSizeLG).sub(e.marginXXS)).div(2).equal()}},[`&${t}-show-text ${t}-text`]:{display:"block"}})}})((0,p.mergeToken)(e,{spinDotDefault:e.colorTextDescription})),e=>{let{controlHeightLG:t,controlHeight:n}=e;return{contentHeight:400,dotSize:t/2,dotSizeSM:.35*t,dotSizeLG:n}}),S=[[30,.05],[70,.03],[96,.01]];var b=function(e,t){var n={};for(var r in e)Object.prototype.hasOwnProperty.call(e,r)&&0>t.indexOf(r)&&(n[r]=e[r]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var i=0,r=Object.getOwnPropertySymbols(e);it.indexOf(r[i])&&Object.prototype.propertyIsEnumerable.call(e,r[i])&&(n[r[i]]=e[r[i]]);return n};let $=e=>{var o;let{prefixCls:s,spinning:a=!0,delay:l=0,className:c,rootClassName:u,size:f="default",tip:h,wrapperClassName:m,style:p,children:v,fullscreen:g=!1,indicator:$,percent:_}=e,w=b(e,["prefixCls","spinning","delay","className","rootClassName","size","tip","wrapperClassName","style","children","fullscreen","indicator","percent"]),{getPrefixCls:O,direction:C,className:x,style:z,indicator:E}=(0,i.useComponentConfig)("spin"),M=O("spin",s),[j,D,T]=y(M),[k,N]=n.useState(()=>a&&(!a||!l||!!Number.isNaN(Number(l)))),R=function(e,t){let[r,i]=n.useState(0),o=n.useRef(null),s="auto"===t;return n.useEffect(()=>(s&&e&&(i(0),o.current=setInterval(()=>{i(e=>{let t=100-e;for(let n=0;n{o.current&&(clearInterval(o.current),o.current=null)}),[s,e]),s?r:t}(k,_);n.useEffect(()=>{if(a){let e=function(e,t,n){var r,i=n||{},o=i.noTrailing,s=void 0!==o&&o,a=i.noLeading,l=void 0!==a&&a,c=i.debounceMode,u=void 0===c?void 0:c,d=!1,f=0;function h(){r&&clearTimeout(r)}function m(){for(var n=arguments.length,i=Array(n),o=0;oe?l?(f=Date.now(),s||(r=setTimeout(u?p:m,e))):m():!0!==s&&(r=setTimeout(u?p:m,void 0===u?e-c:e)))}return m.cancel=function(e){var t=(e||{}).upcomingOnly;h(),d=!(void 0!==t&&t)},m}(l,()=>{N(!0)},{debounceMode:false});return e(),()=>{var t;null==(t=null==e?void 0:e.cancel)||t.call(e)}}N(!1)},[l,a]);let I=n.useMemo(()=>void 0!==v&&!g,[v,g]),A=(0,r.default)(M,x,{[`${M}-sm`]:"small"===f,[`${M}-lg`]:"large"===f,[`${M}-spinning`]:k,[`${M}-show-text`]:!!h,[`${M}-rtl`]:"rtl"===C},c,!g&&u,D,T),F=(0,r.default)(`${M}-container`,{[`${M}-blur`]:k}),P=null!=(o=null!=$?$:E)?o:t,H=Object.assign(Object.assign({},z),p),L=n.createElement("div",Object.assign({},w,{style:H,className:A,"aria-live":"polite","aria-busy":k}),n.createElement(d,{prefixCls:M,indicator:P,percent:R}),h&&(I||g)?n.createElement("div",{className:`${M}-text`},h):null);return j(I?n.createElement("div",Object.assign({},w,{className:(0,r.default)(`${M}-nested-loading`,m,D,T)}),k&&n.createElement("div",{key:"loading"},L),n.createElement("div",{className:F,key:"container"},v)):g?n.createElement("div",{className:(0,r.default)(`${M}-fullscreen`,{[`${M}-fullscreen-show`]:k},u,D,T)},L):L)};$.setDefaultIndicator=e=>{t=e},e.s(["default",0,$],244451),e.s(["Spin",0,$],482725)},597440,e=>{"use strict";e.i(247167);var t=e.i(931067),n=e.i(271645);let r={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M360 184h-8c4.4 0 8-3.6 8-8v8h304v-8c0 4.4 3.6 8 8 8h-8v72h72v-80c0-35.3-28.7-64-64-64H352c-35.3 0-64 28.7-64 64v80h72v-72zm504 72H160c-17.7 0-32 14.3-32 32v32c0 4.4 3.6 8 8 8h60.4l24.7 523c1.6 34.1 29.8 61 63.9 61h454c34.2 0 62.3-26.8 63.9-61l24.7-523H888c4.4 0 8-3.6 8-8v-32c0-17.7-14.3-32-32-32zM731.3 840H292.7l-24.2-512h487l-24.2 512z"}}]},name:"delete",theme:"outlined"};var i=e.i(9583),o=n.forwardRef(function(e,o){return n.createElement(i.default,(0,t.default)({},e,{ref:o,icon:r}))});e.s(["default",0,o],597440)},955135,e=>{"use strict";var t=e.i(597440);e.s(["DeleteOutlined",()=>t.default])},689020,e=>{"use strict";var t=e.i(764205);let n=async e=>{try{let n=await (0,t.modelHubCall)(e);if(console.log("model_info:",n),n?.data.length>0){let e=n.data.map(e=>({model_group:e.model_group,mode:e?.mode}));return e.sort((e,t)=>e.model_group.localeCompare(t.model_group)),e}return[]}catch(e){throw console.error("Error fetching model info:",e),e}};e.s(["fetchAvailableModels",0,n])},981339,e=>{"use strict";var t=e.i(185793);e.s(["Skeleton",()=>t.default])},751904,883552,e=>{"use strict";var t=e.i(401361);e.s(["EditOutlined",()=>t.default],751904),e.i(247167);var n=e.i(271645),r=e.i(562901),i=e.i(343794),o=e.i(914949),s=e.i(529681),a=e.i(242064),l=e.i(829672),c=e.i(285781),u=e.i(836938),d=e.i(920228),f=e.i(62405),h=e.i(408850),m=e.i(87414),p=e.i(310730);let v=(0,e.i(246422).genStyleHooks)("Popconfirm",e=>(e=>{let{componentCls:t,iconCls:n,antCls:r,zIndexPopup:i,colorText:o,colorWarning:s,marginXXS:a,marginXS:l,fontSize:c,fontWeightStrong:u,colorTextHeading:d}=e;return{[t]:{zIndex:i,[`&${r}-popover`]:{fontSize:c},[`${t}-message`]:{marginBottom:l,display:"flex",flexWrap:"nowrap",alignItems:"start",[`> ${t}-message-icon ${n}`]:{color:s,fontSize:c,lineHeight:1,marginInlineEnd:l},[`${t}-title`]:{fontWeight:u,color:d,"&:only-child":{fontWeight:"normal"}},[`${t}-description`]:{marginTop:a,color:o}},[`${t}-buttons`]:{textAlign:"end",whiteSpace:"nowrap",button:{marginInlineStart:l}}}}})(e),e=>{let{zIndexPopupBase:t}=e;return{zIndexPopup:t+60}},{resetStyle:!1});var g=function(e,t){var n={};for(var r in e)Object.prototype.hasOwnProperty.call(e,r)&&0>t.indexOf(r)&&(n[r]=e[r]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var i=0,r=Object.getOwnPropertySymbols(e);it.indexOf(r[i])&&Object.prototype.propertyIsEnumerable.call(e,r[i])&&(n[r[i]]=e[r[i]]);return n};let y=e=>{let{prefixCls:t,okButtonProps:i,cancelButtonProps:o,title:s,description:l,cancelText:p,okText:v,okType:g="primary",icon:y=n.createElement(r.default,null),showCancel:S=!0,close:b,onConfirm:$,onCancel:_,onPopupClick:w}=e,{getPrefixCls:O}=n.useContext(a.ConfigContext),[C]=(0,h.useLocale)("Popconfirm",m.default.Popconfirm),x=(0,u.getRenderPropValue)(s),z=(0,u.getRenderPropValue)(l);return n.createElement("div",{className:`${t}-inner-content`,onClick:w},n.createElement("div",{className:`${t}-message`},y&&n.createElement("span",{className:`${t}-message-icon`},y),n.createElement("div",{className:`${t}-message-text`},x&&n.createElement("div",{className:`${t}-title`},x),z&&n.createElement("div",{className:`${t}-description`},z))),n.createElement("div",{className:`${t}-buttons`},S&&n.createElement(d.default,Object.assign({onClick:_,size:"small"},o),p||(null==C?void 0:C.cancelText)),n.createElement(c.default,{buttonProps:Object.assign(Object.assign({size:"small"},(0,f.convertLegacyProps)(g)),i),actionFn:$,close:b,prefixCls:O("btn"),quitOnNullishReturnValue:!0,emitEvent:!0},v||(null==C?void 0:C.okText))))};var S=function(e,t){var n={};for(var r in e)Object.prototype.hasOwnProperty.call(e,r)&&0>t.indexOf(r)&&(n[r]=e[r]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var i=0,r=Object.getOwnPropertySymbols(e);it.indexOf(r[i])&&Object.prototype.propertyIsEnumerable.call(e,r[i])&&(n[r[i]]=e[r[i]]);return n};let b=n.forwardRef((e,t)=>{var c,u;let{prefixCls:d,placement:f="top",trigger:h="click",okType:m="primary",icon:p=n.createElement(r.default,null),children:g,overlayClassName:b,onOpenChange:$,onVisibleChange:_,overlayStyle:w,styles:O,classNames:C}=e,x=S(e,["prefixCls","placement","trigger","okType","icon","children","overlayClassName","onOpenChange","onVisibleChange","overlayStyle","styles","classNames"]),{getPrefixCls:z,className:E,style:M,classNames:j,styles:D}=(0,a.useComponentConfig)("popconfirm"),[T,k]=(0,o.default)(!1,{value:null!=(c=e.open)?c:e.visible,defaultValue:null!=(u=e.defaultOpen)?u:e.defaultVisible}),N=(e,t)=>{k(e,!0),null==_||_(e),null==$||$(e,t)},R=z("popconfirm",d),I=(0,i.default)(R,E,b,j.root,null==C?void 0:C.root),A=(0,i.default)(j.body,null==C?void 0:C.body),[F]=v(R);return F(n.createElement(l.default,Object.assign({},(0,s.default)(x,["title"]),{trigger:h,placement:f,onOpenChange:(t,n)=>{let{disabled:r=!1}=e;r||N(t,n)},open:T,ref:t,classNames:{root:I,body:A},styles:{root:Object.assign(Object.assign(Object.assign(Object.assign({},D.root),M),w),null==O?void 0:O.root),body:Object.assign(Object.assign({},D.body),null==O?void 0:O.body)},content:n.createElement(y,Object.assign({okType:m,icon:p},e,{prefixCls:R,close:e=>{N(!1,e)},onConfirm:t=>{var n;return null==(n=e.onConfirm)?void 0:n.call(void 0,t)},onCancel:t=>{var n;N(!1,t),null==(n=e.onCancel)||n.call(void 0,t)}})),"data-popover-inject":!0}),g))});b._InternalPanelDoNotUseOrYouWillBeFired=e=>{let{prefixCls:t,placement:r,className:o,style:s}=e,l=g(e,["prefixCls","placement","className","style"]),{getPrefixCls:c}=n.useContext(a.ConfigContext),u=c("popconfirm",t),[d]=v(u);return d(n.createElement(p.default,{placement:r,className:(0,i.default)(u,o),style:s,content:n.createElement(y,Object.assign({prefixCls:u},l))}))},e.s(["Popconfirm",0,b],883552)},822315,(e,t,n)=>{e.e,t.exports=function(){"use strict";var e="millisecond",t="second",n="minute",r="hour",i="week",o="month",s="quarter",a="year",l="date",c="Invalid Date",u=/^(\d{4})[-/]?(\d{1,2})?[-/]?(\d{0,2})[Tt\s]*(\d{1,2})?:?(\d{1,2})?:?(\d{1,2})?[.:]?(\d+)?$/,d=/\[([^\]]+)]|Y{1,4}|M{1,4}|D{1,2}|d{1,4}|H{1,2}|h{1,2}|a|A|m{1,2}|s{1,2}|Z{1,2}|SSS/g,f=function(e,t,n){var r=String(e);return!r||r.length>=t?e:""+Array(t+1-r.length).join(n)+e},h="en",m={};m[h]={name:"en",weekdays:"Sunday_Monday_Tuesday_Wednesday_Thursday_Friday_Saturday".split("_"),months:"January_February_March_April_May_June_July_August_September_October_November_December".split("_"),ordinal:function(e){var t=["th","st","nd","rd"],n=e%100;return"["+e+(t[(n-20)%10]||t[n]||t[0])+"]"}};var p="$isDayjsObject",v=function(e){return e instanceof b||!(!e||!e[p])},g=function e(t,n,r){var i;if(!t)return h;if("string"==typeof t){var o=t.toLowerCase();m[o]&&(i=o),n&&(m[o]=n,i=o);var s=t.split("-");if(!i&&s.length>1)return e(s[0])}else{var a=t.name;m[a]=t,i=a}return!r&&i&&(h=i),i||!r&&h},y=function(e,t){if(v(e))return e.clone();var n="object"==typeof t?t:{};return n.date=e,n.args=arguments,new b(n)},S={s:f,z:function(e){var t=-e.utcOffset(),n=Math.abs(t);return(t<=0?"+":"-")+f(Math.floor(n/60),2,"0")+":"+f(n%60,2,"0")},m:function e(t,n){if(t.date(){"use strict";var t=e.i(959013);e.s(["PlusOutlined",()=>t.default])},313603,e=>{"use strict";e.i(247167);var t=e.i(931067),n=e.i(271645);let r={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M924.8 625.7l-65.5-56c3.1-19 4.7-38.4 4.7-57.8s-1.6-38.8-4.7-57.8l65.5-56a32.03 32.03 0 009.3-35.2l-.9-2.6a443.74 443.74 0 00-79.7-137.9l-1.8-2.1a32.12 32.12 0 00-35.1-9.5l-81.3 28.9c-30-24.6-63.5-44-99.7-57.6l-15.7-85a32.05 32.05 0 00-25.8-25.7l-2.7-.5c-52.1-9.4-106.9-9.4-159 0l-2.7.5a32.05 32.05 0 00-25.8 25.7l-15.8 85.4a351.86 351.86 0 00-99 57.4l-81.9-29.1a32 32 0 00-35.1 9.5l-1.8 2.1a446.02 446.02 0 00-79.7 137.9l-.9 2.6c-4.5 12.5-.8 26.5 9.3 35.2l66.3 56.6c-3.1 18.8-4.6 38-4.6 57.1 0 19.2 1.5 38.4 4.6 57.1L99 625.5a32.03 32.03 0 00-9.3 35.2l.9 2.6c18.1 50.4 44.9 96.9 79.7 137.9l1.8 2.1a32.12 32.12 0 0035.1 9.5l81.9-29.1c29.8 24.5 63.1 43.9 99 57.4l15.8 85.4a32.05 32.05 0 0025.8 25.7l2.7.5a449.4 449.4 0 00159 0l2.7-.5a32.05 32.05 0 0025.8-25.7l15.7-85a350 350 0 0099.7-57.6l81.3 28.9a32 32 0 0035.1-9.5l1.8-2.1c34.8-41.1 61.6-87.5 79.7-137.9l.9-2.6c4.5-12.3.8-26.3-9.3-35zM788.3 465.9c2.5 15.1 3.8 30.6 3.8 46.1s-1.3 31-3.8 46.1l-6.6 40.1 74.7 63.9a370.03 370.03 0 01-42.6 73.6L721 702.8l-31.4 25.8c-23.9 19.6-50.5 35-79.3 45.8l-38.1 14.3-17.9 97a377.5 377.5 0 01-85 0l-17.9-97.2-37.8-14.5c-28.5-10.8-55-26.2-78.7-45.7l-31.4-25.9-93.4 33.2c-17-22.9-31.2-47.6-42.6-73.6l75.5-64.5-6.5-40c-2.4-14.9-3.7-30.3-3.7-45.5 0-15.3 1.2-30.6 3.7-45.5l6.5-40-75.5-64.5c11.3-26.1 25.6-50.7 42.6-73.6l93.4 33.2 31.4-25.9c23.7-19.5 50.2-34.9 78.7-45.7l37.9-14.3 17.9-97.2c28.1-3.2 56.8-3.2 85 0l17.9 97 38.1 14.3c28.7 10.8 55.4 26.2 79.3 45.8l31.4 25.8 92.8-32.9c17 22.9 31.2 47.6 42.6 73.6L781.8 426l6.5 39.9zM512 326c-97.2 0-176 78.8-176 176s78.8 176 176 176 176-78.8 176-176-78.8-176-176-176zm79.2 255.2A111.6 111.6 0 01512 614c-29.9 0-58-11.7-79.2-32.8A111.6 111.6 0 01400 502c0-29.9 11.7-58 32.8-79.2C454 401.6 482.1 390 512 390c29.9 0 58 11.6 79.2 32.8A111.6 111.6 0 01624 502c0 29.9-11.7 58-32.8 79.2z"}}]},name:"setting",theme:"outlined"};var i=e.i(9583),o=n.forwardRef(function(e,o){return n.createElement(i.default,(0,t.default)({},e,{ref:o,icon:r}))});e.s(["SettingOutlined",0,o],313603)},362024,e=>{"use strict";var t=e.i(988122);e.s(["Collapse",()=>t.default])},240647,e=>{"use strict";var t=e.i(286612);e.s(["RightOutlined",()=>t.default])},516015,(e,t,n)=>{},898547,(e,t,n)=>{var r=e.i(247167);e.r(516015);var i=e.r(271645),o=i&&"object"==typeof i&&"default"in i?i:{default:i},s=void 0!==r.default&&r.default.env&&!0,a=function(e){return"[object String]"===Object.prototype.toString.call(e)},l=function(){function e(e){var t=void 0===e?{}:e,n=t.name,r=void 0===n?"stylesheet":n,i=t.optimizeForSpeed,o=void 0===i?s:i;c(a(r),"`name` must be a string"),this._name=r,this._deletedRulePlaceholder="#"+r+"-deleted-rule____{}",c("boolean"==typeof o,"`optimizeForSpeed` must be a boolean"),this._optimizeForSpeed=o,this._serverSheet=void 0,this._tags=[],this._injected=!1,this._rulesCount=0;var l="u">typeof window&&document.querySelector('meta[property="csp-nonce"]');this._nonce=l?l.getAttribute("content"):null}var t,n=e.prototype;return n.setOptimizeForSpeed=function(e){c("boolean"==typeof e,"`setOptimizeForSpeed` accepts a boolean"),c(0===this._rulesCount,"optimizeForSpeed cannot be when rules have already been inserted"),this.flush(),this._optimizeForSpeed=e,this.inject()},n.isOptimizeForSpeed=function(){return this._optimizeForSpeed},n.inject=function(){var e=this;if(c(!this._injected,"sheet already injected"),this._injected=!0,"u">typeof window&&this._optimizeForSpeed){this._tags[0]=this.makeStyleTag(this._name),this._optimizeForSpeed="insertRule"in this.getSheet(),this._optimizeForSpeed||(s||console.warn("StyleSheet: optimizeForSpeed mode not supported falling back to standard mode."),this.flush(),this._injected=!0);return}this._serverSheet={cssRules:[],insertRule:function(t,n){return"number"==typeof n?e._serverSheet.cssRules[n]={cssText:t}:e._serverSheet.cssRules.push({cssText:t}),n},deleteRule:function(t){e._serverSheet.cssRules[t]=null}}},n.getSheetForTag=function(e){if(e.sheet)return e.sheet;for(var t=0;ttypeof window?this.getSheet():this._serverSheet;if(t.trim()||(t=this._deletedRulePlaceholder),!n.cssRules[e])return e;n.deleteRule(e);try{n.insertRule(t,e)}catch(r){s||console.warn("StyleSheet: illegal rule: \n\n"+t+"\n\nSee https://stackoverflow.com/q/20007992 for more info"),n.insertRule(this._deletedRulePlaceholder,e)}}else{var r=this._tags[e];c(r,"old rule at index `"+e+"` not found"),r.textContent=t}return e},n.deleteRule=function(e){if("u"typeof window?(this._tags.forEach(function(e){return e&&e.parentNode.removeChild(e)}),this._tags=[]):this._serverSheet.cssRules=[]},n.cssRules=function(){var e=this;return"u">>0},d={};function f(e,t){if(!t)return"jsx-"+e;var n=String(t),r=e+n;return d[r]||(d[r]="jsx-"+u(e+"-"+n)),d[r]}function h(e,t){"u"typeof window&&!this._fromServer&&(this._fromServer=this.selectFromServer(),this._instancesCounts=Object.keys(this._fromServer).reduce(function(e,t){return e[t]=0,e},{}));var n=this.getIdAndRules(e),r=n.styleId,i=n.rules;if(r in this._instancesCounts){this._instancesCounts[r]+=1;return}var o=i.map(function(e){return t._sheet.insertRule(e)}).filter(function(e){return -1!==e});this._indices[r]=o,this._instancesCounts[r]=1},t.remove=function(e){var t=this,n=this.getIdAndRules(e).styleId;if(function(e,t){if(!e)throw Error("StyleSheetRegistry: "+t+".")}(n in this._instancesCounts,"styleId: `"+n+"` not found"),this._instancesCounts[n]-=1,this._instancesCounts[n]<1){var r=this._fromServer&&this._fromServer[n];r?(r.parentNode.removeChild(r),delete this._fromServer[n]):(this._indices[n].forEach(function(e){return t._sheet.deleteRule(e)}),delete this._indices[n]),delete this._instancesCounts[n]}},t.update=function(e,t){this.add(t),this.remove(e)},t.flush=function(){this._sheet.flush(),this._sheet.inject(),this._fromServer=void 0,this._indices={},this._instancesCounts={}},t.cssRules=function(){var e=this,t=this._fromServer?Object.keys(this._fromServer).map(function(t){return[t,e._fromServer[t]]}):[],n=this._sheet.cssRules();return t.concat(Object.keys(this._indices).map(function(t){return[t,e._indices[t].map(function(e){return n[e].cssText}).join(e._optimizeForSpeed?"":"\n")]}).filter(function(e){return!!e[1]}))},t.styles=function(e){var t,n;return t=this.cssRules(),void 0===(n=e)&&(n={}),t.map(function(e){var t=e[0],r=e[1];return o.default.createElement("style",{id:"__"+t,key:"__"+t,nonce:n.nonce?n.nonce:void 0,dangerouslySetInnerHTML:{__html:r}})})},t.getIdAndRules=function(e){var t=e.children,n=e.dynamic,r=e.id;if(n){var i=f(r,n);return{styleId:i,rules:Array.isArray(t)?t.map(function(e){return h(i,e)}):[h(i,t)]}}return{styleId:f(r),rules:Array.isArray(t)?t:[t]}},t.selectFromServer=function(){return Array.prototype.slice.call(document.querySelectorAll('[id^="__jsx-"]')).reduce(function(e,t){return e[t.id.slice(2)]=t,e},{})},e}(),p=i.createContext(null);function v(){return new m}function g(){return i.useContext(p)}p.displayName="StyleSheetContext";var y=o.default.useInsertionEffect||o.default.useLayoutEffect,S="u">typeof window?v():void 0;function b(e){var t=S||g();return t&&("u"{t.exports=e.r(898547).style},434166,e=>{"use strict";function t(e,t){window.sessionStorage.setItem(e,btoa(encodeURIComponent(t).replace(/%([0-9A-F]{2})/g,(e,t)=>String.fromCharCode(parseInt(t,16)))))}function n(e){try{let t=window.sessionStorage.getItem(e);if(null===t)return null;return decodeURIComponent(atob(t).split("").map(e=>"%"+e.charCodeAt(0).toString(16).padStart(2,"0")).join(""))}catch{return null}}e.s(["getSecureItem",()=>n,"setSecureItem",()=>t])},438957,366308,e=>{"use strict";e.i(247167);var t=e.i(931067),n=e.i(271645);let r={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M608 112c-167.9 0-304 136.1-304 304 0 70.3 23.9 135 63.9 186.5l-41.1 41.1-62.3-62.3a8.15 8.15 0 00-11.4 0l-39.8 39.8a8.15 8.15 0 000 11.4l62.3 62.3-44.9 44.9-62.3-62.3a8.15 8.15 0 00-11.4 0l-39.8 39.8a8.15 8.15 0 000 11.4l62.3 62.3-65.3 65.3a8.03 8.03 0 000 11.3l42.3 42.3c3.1 3.1 8.2 3.1 11.3 0l253.6-253.6A304.06 304.06 0 00608 720c167.9 0 304-136.1 304-304S775.9 112 608 112zm161.2 465.2C726.2 620.3 668.9 644 608 644c-60.9 0-118.2-23.7-161.2-66.8-43.1-43-66.8-100.3-66.8-161.2 0-60.9 23.7-118.2 66.8-161.2 43-43.1 100.3-66.8 161.2-66.8 60.9 0 118.2 23.7 161.2 66.8 43.1 43 66.8 100.3 66.8 161.2 0 60.9-23.7 118.2-66.8 161.2z"}}]},name:"key",theme:"outlined"};var i=e.i(9583),o=n.forwardRef(function(e,o){return n.createElement(i.default,(0,t.default)({},e,{ref:o,icon:r}))});e.s(["KeyOutlined",0,o],438957);let s={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M876.6 239.5c-.5-.9-1.2-1.8-2-2.5-5-5-13.1-5-18.1 0L684.2 409.3l-67.9-67.9L788.7 169c.8-.8 1.4-1.6 2-2.5 3.6-6.1 1.6-13.9-4.5-17.5-98.2-58-226.8-44.7-311.3 39.7-67 67-89.2 162-66.5 247.4l-293 293c-3 3-2.8 7.9.3 11l169.7 169.7c3.1 3.1 8.1 3.3 11 .3l292.9-292.9c85.5 22.8 180.5.7 247.6-66.4 84.4-84.5 97.7-213.1 39.7-311.3zM786 499.8c-58.1 58.1-145.3 69.3-214.6 33.6l-8.8 8.8-.1-.1-274 274.1-79.2-79.2 230.1-230.1s0 .1.1.1l52.8-52.8c-35.7-69.3-24.5-156.5 33.6-214.6a184.2 184.2 0 01144-53.5L537 318.9a32.05 32.05 0 000 45.3l124.5 124.5a32.05 32.05 0 0045.3 0l132.8-132.8c3.7 51.8-14.4 104.8-53.6 143.9z"}}]},name:"tool",theme:"outlined"};var a=n.forwardRef(function(e,r){return n.createElement(i.default,(0,t.default)({},e,{ref:r,icon:s}))});e.s(["ToolOutlined",0,a],366308)},477189,e=>{"use strict";e.i(247167);var t=e.i(931067),n=e.i(271645);let r={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M464 144H160c-8.8 0-16 7.2-16 16v304c0 8.8 7.2 16 16 16h304c8.8 0 16-7.2 16-16V160c0-8.8-7.2-16-16-16zm-52 268H212V212h200v200zm452-268H560c-8.8 0-16 7.2-16 16v304c0 8.8 7.2 16 16 16h304c8.8 0 16-7.2 16-16V160c0-8.8-7.2-16-16-16zm-52 268H612V212h200v200zM464 544H160c-8.8 0-16 7.2-16 16v304c0 8.8 7.2 16 16 16h304c8.8 0 16-7.2 16-16V560c0-8.8-7.2-16-16-16zm-52 268H212V612h200v200zm452-268H560c-8.8 0-16 7.2-16 16v304c0 8.8 7.2 16 16 16h304c8.8 0 16-7.2 16-16V560c0-8.8-7.2-16-16-16zm-52 268H612V612h200v200z"}}]},name:"appstore",theme:"outlined"};var i=e.i(9583),o=n.forwardRef(function(e,o){return n.createElement(i.default,(0,t.default)({},e,{ref:o,icon:r}))});e.s(["AppstoreOutlined",0,o],477189)},264843,292335,122520,165615,779129,e=>{"use strict";e.i(247167);var t=e.i(931067),n=e.i(271645);let r={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M464 512a48 48 0 1096 0 48 48 0 10-96 0zm200 0a48 48 0 1096 0 48 48 0 10-96 0zm-400 0a48 48 0 1096 0 48 48 0 10-96 0zm661.2-173.6c-22.6-53.7-55-101.9-96.3-143.3a444.35 444.35 0 00-143.3-96.3C630.6 75.7 572.2 64 512 64h-2c-60.6.3-119.3 12.3-174.5 35.9a445.35 445.35 0 00-142 96.5c-40.9 41.3-73 89.3-95.2 142.8-23 55.4-34.6 114.3-34.3 174.9A449.4 449.4 0 00112 714v152a46 46 0 0046 46h152.1A449.4 449.4 0 00510 960h2.1c59.9 0 118-11.6 172.7-34.3a444.48 444.48 0 00142.8-95.2c41.3-40.9 73.8-88.7 96.5-142 23.6-55.2 35.6-113.9 35.9-174.5.3-60.9-11.5-120-34.8-175.6zm-151.1 438C704 845.8 611 884 512 884h-1.7c-60.3-.3-120.2-15.3-173.1-43.5l-8.4-4.5H188V695.2l-4.5-8.4C155.3 633.9 140.3 574 140 513.7c-.4-99.7 37.7-193.3 107.6-263.8 69.8-70.5 163.1-109.5 262.8-109.9h1.7c50 0 98.5 9.7 144.2 28.9 44.6 18.7 84.6 45.6 119 80 34.3 34.3 61.3 74.4 80 119 19.4 46.2 29.1 95.2 28.9 145.8-.6 99.6-39.7 192.9-110.1 262.7z"}}]},name:"message",theme:"outlined"};var i=e.i(9583),o=n.forwardRef(function(e,o){return n.createElement(i.default,(0,t.default)({},e,{ref:o,icon:r}))});e.s(["MessageOutlined",0,o],264843);let s={NONE:"none",API_KEY:"api_key",BEARER_TOKEN:"bearer_token",TOKEN:"token",BASIC:"basic",OAUTH2:"oauth2",AWS_SIGV4:"aws_sigv4"},a={SSE:"sse",HTTP:"http",STDIO:"stdio",OPENAPI:"openapi"};function l(e){if(e instanceof Error)return e.message;if(e&&"object"==typeof e){let t=e.detail;return"string"==typeof t?t:Array.isArray(t)?t.map(e=>e&&"object"==typeof e?"string"==typeof e.msg?e.msg:JSON.stringify(e):String(e)).join("; "):t&&"object"==typeof t&&"string"==typeof t.error?t.error:"string"==typeof e.message?e.message:JSON.stringify(e)}return String(e)}e.s(["AUTH_TYPE",0,s,"OAUTH_FLOW",0,{INTERACTIVE:"interactive",M2M:"m2m"},"TRANSPORT",0,a,"handleAuth",0,e=>null==e?s.NONE:e,"handleTransport",0,(e,t)=>null==e?a.SSE:t&&e!==a.STDIO?a.OPENAPI:e],292335),e.s(["extractErrorMessage",()=>l],122520);let c=e=>{let t=new Uint8Array(e),n="";return t.forEach(e=>n+=String.fromCharCode(e)),btoa(n).replace(/\+/g,"-").replace(/\//g,"_").replace(/=+$/,"")},u=async e=>{let t=new TextEncoder().encode(e);return c(await window.crypto.subtle.digest("SHA-256",t))};e.s(["generateCodeChallenge",0,u,"generateCodeVerifier",0,()=>{let e=new Uint8Array(32);return window.crypto.getRandomValues(e),c(e.buffer)}],165615),e.i(764205),e.s(["buildCallbackUrl",0,()=>{{let e=window.location.pathname||"",t=e.indexOf("/ui"),n=t>=0?e.slice(0,t+3).replace(/\/+$/,""):"";return`${window.location.origin}${n}/mcp/oauth/callback`}},"clearStorage",0,(...e)=>{e.forEach(e=>{try{window.sessionStorage.removeItem(e)}catch(e){}})}],779129)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/01a2d4575f32b1b4.js b/litellm/proxy/_experimental/out/_next/static/chunks/01a2d4575f32b1b4.js new file mode 100644 index 00000000000..1a3f4b3b3c9 --- /dev/null +++ b/litellm/proxy/_experimental/out/_next/static/chunks/01a2d4575f32b1b4.js @@ -0,0 +1,8 @@ +(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,282786,836938,310730,829672,e=>{"use strict";e.i(247167);var t=e.i(271645),r=e.i(343794),i=e.i(914949),n=e.i(404948);let s=e=>e?"function"==typeof e?e():e:null;e.s(["getRenderPropValue",0,s],836938);var a=e.i(613541),l=e.i(763731),o=e.i(242064),u=e.i(491816);e.i(793154);var c=e.i(880476),d=e.i(183293),h=e.i(717356),p=e.i(320560),f=e.i(307358),g=e.i(246422),m=e.i(838378),b=e.i(617933);let y=(0,g.genStyleHooks)("Popover",e=>{let{colorBgElevated:t,colorText:r}=e,i=(0,m.mergeToken)(e,{popoverBg:t,popoverColor:r});return[(e=>{let{componentCls:t,popoverColor:r,titleMinWidth:i,fontWeightStrong:n,innerPadding:s,boxShadowSecondary:a,colorTextHeading:l,borderRadiusLG:o,zIndexPopup:u,titleMarginBottom:c,colorBgElevated:h,popoverBg:f,titleBorderBottom:g,innerContentPadding:m,titlePadding:b}=e;return[{[t]:Object.assign(Object.assign({},(0,d.resetComponent)(e)),{position:"absolute",top:0,left:{_skip_check_:!0,value:0},zIndex:u,fontWeight:"normal",whiteSpace:"normal",textAlign:"start",cursor:"auto",userSelect:"text","--valid-offset-x":"var(--arrow-offset-horizontal, var(--arrow-x))",transformOrigin:"var(--valid-offset-x, 50%) var(--arrow-y, 50%)","--antd-arrow-background-color":h,width:"max-content",maxWidth:"100vw","&-rtl":{direction:"rtl"},"&-hidden":{display:"none"},[`${t}-content`]:{position:"relative"},[`${t}-inner`]:{backgroundColor:f,backgroundClip:"padding-box",borderRadius:o,boxShadow:a,padding:s},[`${t}-title`]:{minWidth:i,marginBottom:c,color:l,fontWeight:n,borderBottom:g,padding:b},[`${t}-inner-content`]:{color:r,padding:m}})},(0,p.default)(e,"var(--antd-arrow-background-color)"),{[`${t}-pure`]:{position:"relative",maxWidth:"none",margin:e.sizePopupArrow,display:"inline-block",[`${t}-content`]:{display:"inline-block"}}}]})(i),(e=>{let{componentCls:t}=e;return{[t]:b.PresetColors.map(r=>{let i=e[`${r}6`];return{[`&${t}-${r}`]:{"--antd-arrow-background-color":i,[`${t}-inner`]:{backgroundColor:i},[`${t}-arrow`]:{background:"transparent"}}}})}})(i),(0,h.initZoomMotion)(i,"zoom-big")]},e=>{let{lineWidth:t,controlHeight:r,fontHeight:i,padding:n,wireframe:s,zIndexPopupBase:a,borderRadiusLG:l,marginXS:o,lineType:u,colorSplit:c,paddingSM:d}=e,h=r-i;return Object.assign(Object.assign(Object.assign({titleMinWidth:177,zIndexPopup:a+30},(0,f.getArrowToken)(e)),(0,p.getArrowOffsetToken)({contentRadius:l,limitVerticalRadius:!0})),{innerPadding:12*!s,titleMarginBottom:s?0:o,titlePadding:s?`${h/2}px ${n}px ${h/2-t}px`:0,titleBorderBottom:s?`${t}px ${u} ${c}`:"none",innerContentPadding:s?`${d}px ${n}px`:0})},{resetStyle:!1,deprecatedTokens:[["width","titleMinWidth"],["minWidth","titleMinWidth"]]});var v=function(e,t){var r={};for(var i in e)Object.prototype.hasOwnProperty.call(e,i)&&0>t.indexOf(i)&&(r[i]=e[i]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var n=0,i=Object.getOwnPropertySymbols(e);nt.indexOf(i[n])&&Object.prototype.propertyIsEnumerable.call(e,i[n])&&(r[i[n]]=e[i[n]]);return r};let O=({title:e,content:r,prefixCls:i})=>e||r?t.createElement(t.Fragment,null,e&&t.createElement("div",{className:`${i}-title`},e),r&&t.createElement("div",{className:`${i}-inner-content`},r)):null,$=e=>{let{hashId:i,prefixCls:n,className:a,style:l,placement:o="top",title:u,content:d,children:h}=e,p=s(u),f=s(d),g=(0,r.default)(i,n,`${n}-pure`,`${n}-placement-${o}`,a);return t.createElement("div",{className:g,style:l},t.createElement("div",{className:`${n}-arrow`}),t.createElement(c.Popup,Object.assign({},e,{className:i,prefixCls:n}),h||t.createElement(O,{prefixCls:n,title:p,content:f})))},R=e=>{let{prefixCls:i,className:n}=e,s=v(e,["prefixCls","className"]),{getPrefixCls:a}=t.useContext(o.ConfigContext),l=a("popover",i),[u,c,d]=y(l);return u(t.createElement($,Object.assign({},s,{prefixCls:l,hashId:c,className:(0,r.default)(n,d)})))};e.s(["Overlay",0,O,"default",0,R],310730);var C=function(e,t){var r={};for(var i in e)Object.prototype.hasOwnProperty.call(e,i)&&0>t.indexOf(i)&&(r[i]=e[i]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var n=0,i=Object.getOwnPropertySymbols(e);nt.indexOf(i[n])&&Object.prototype.propertyIsEnumerable.call(e,i[n])&&(r[i[n]]=e[i[n]]);return r};let w=t.forwardRef((e,c)=>{var d,h;let{prefixCls:p,title:f,content:g,overlayClassName:m,placement:b="top",trigger:v="hover",children:$,mouseEnterDelay:R=.1,mouseLeaveDelay:w=.1,onOpenChange:x,overlayStyle:E={},styles:k,classNames:j}=e,S=C(e,["prefixCls","title","content","overlayClassName","placement","trigger","children","mouseEnterDelay","mouseLeaveDelay","onOpenChange","overlayStyle","styles","classNames"]),{getPrefixCls:I,className:T,style:Q,classNames:q,styles:U}=(0,o.useComponentConfig)("popover"),P=I("popover",p),[N,M,D]=y(P),F=I(),W=(0,r.default)(m,M,D,T,q.root,null==j?void 0:j.root),L=(0,r.default)(q.body,null==j?void 0:j.body),[A,B]=(0,i.default)(!1,{value:null!=(d=e.open)?d:e.visible,defaultValue:null!=(h=e.defaultOpen)?h:e.defaultVisible}),z=(e,t)=>{B(e,!0),null==x||x(e,t)},_=s(f),H=s(g);return N(t.createElement(u.default,Object.assign({placement:b,trigger:v,mouseEnterDelay:R,mouseLeaveDelay:w},S,{prefixCls:P,classNames:{root:W,body:L},styles:{root:Object.assign(Object.assign(Object.assign(Object.assign({},U.root),Q),E),null==k?void 0:k.root),body:Object.assign(Object.assign({},U.body),null==k?void 0:k.body)},ref:c,open:A,onOpenChange:e=>{z(e)},overlay:_||H?t.createElement(O,{prefixCls:P,title:_,content:H}):null,transitionName:(0,a.getTransitionName)(F,"zoom-big",S.transitionName),"data-popover-inject":!0}),(0,l.cloneElement)($,{onKeyDown:e=>{var r,i;(0,t.isValidElement)($)&&(null==(i=null==$?void 0:(r=$.props).onKeyDown)||i.call(r,e)),e.keyCode===n.default.ESC&&z(!1,e)}})))});w._InternalPanelDoNotUseOrYouWillBeFired=R,e.s(["default",0,w],829672),e.s(["Popover",0,w],282786)},618566,(e,t,r)=>{t.exports=e.r(976562)},612256,869230,469637,266027,243652,e=>{"use strict";let t;var r=e.i(764205),i=e.i(175555),n=e.i(540143),s=e.i(286491),a=e.i(915823),l=e.i(793803),o=e.i(619273),u=e.i(180166),c=class extends a.Subscribable{constructor(e,t){super(),this.options=t,this.#e=e,this.#t=null,this.#r=(0,l.pendingThenable)(),this.bindMethods(),this.setOptions(t)}#e;#i=void 0;#n=void 0;#s=void 0;#a;#l;#r;#t;#o;#u;#c;#d;#h;#p;#f=new Set;bindMethods(){this.refetch=this.refetch.bind(this)}onSubscribe(){1===this.listeners.size&&(this.#i.addObserver(this),d(this.#i,this.options)?this.#g():this.updateResult(),this.#m())}onUnsubscribe(){this.hasListeners()||this.destroy()}shouldFetchOnReconnect(){return h(this.#i,this.options,this.options.refetchOnReconnect)}shouldFetchOnWindowFocus(){return h(this.#i,this.options,this.options.refetchOnWindowFocus)}destroy(){this.listeners=new Set,this.#b(),this.#y(),this.#i.removeObserver(this)}setOptions(e){let t=this.options,r=this.#i;if(this.options=this.#e.defaultQueryOptions(e),void 0!==this.options.enabled&&"boolean"!=typeof this.options.enabled&&"function"!=typeof this.options.enabled&&"boolean"!=typeof(0,o.resolveEnabled)(this.options.enabled,this.#i))throw Error("Expected enabled to be a boolean or a callback that returns a boolean");this.#v(),this.#i.setOptions(this.options),t._defaulted&&!(0,o.shallowEqualObjects)(this.options,t)&&this.#e.getQueryCache().notify({type:"observerOptionsUpdated",query:this.#i,observer:this});let i=this.hasListeners();i&&p(this.#i,r,this.options,t)&&this.#g(),this.updateResult(),i&&(this.#i!==r||(0,o.resolveEnabled)(this.options.enabled,this.#i)!==(0,o.resolveEnabled)(t.enabled,this.#i)||(0,o.resolveStaleTime)(this.options.staleTime,this.#i)!==(0,o.resolveStaleTime)(t.staleTime,this.#i))&&this.#O();let n=this.#$();i&&(this.#i!==r||(0,o.resolveEnabled)(this.options.enabled,this.#i)!==(0,o.resolveEnabled)(t.enabled,this.#i)||n!==this.#p)&&this.#R(n)}getOptimisticResult(e){var t,r;let i=this.#e.getQueryCache().build(this.#e,e),n=this.createResult(i,e);return t=this,r=n,(0,o.shallowEqualObjects)(t.getCurrentResult(),r)||(this.#s=n,this.#l=this.options,this.#a=this.#i.state),n}getCurrentResult(){return this.#s}trackResult(e,t){return new Proxy(e,{get:(e,r)=>(this.trackProp(r),t?.(r),"promise"===r&&(this.trackProp("data"),this.options.experimental_prefetchInRender||"pending"!==this.#r.status||this.#r.reject(Error("experimental_prefetchInRender feature flag is not enabled"))),Reflect.get(e,r))})}trackProp(e){this.#f.add(e)}getCurrentQuery(){return this.#i}refetch({...e}={}){return this.fetch({...e})}fetchOptimistic(e){let t=this.#e.defaultQueryOptions(e),r=this.#e.getQueryCache().build(this.#e,t);return r.fetch().then(()=>this.createResult(r,t))}fetch(e){return this.#g({...e,cancelRefetch:e.cancelRefetch??!0}).then(()=>(this.updateResult(),this.#s))}#g(e){this.#v();let t=this.#i.fetch(this.options,e);return e?.throwOnError||(t=t.catch(o.noop)),t}#O(){this.#b();let e=(0,o.resolveStaleTime)(this.options.staleTime,this.#i);if(o.isServer||this.#s.isStale||!(0,o.isValidTimeout)(e))return;let t=(0,o.timeUntilStale)(this.#s.dataUpdatedAt,e);this.#d=u.timeoutManager.setTimeout(()=>{this.#s.isStale||this.updateResult()},t+1)}#$(){return("function"==typeof this.options.refetchInterval?this.options.refetchInterval(this.#i):this.options.refetchInterval)??!1}#R(e){this.#y(),this.#p=e,!o.isServer&&!1!==(0,o.resolveEnabled)(this.options.enabled,this.#i)&&(0,o.isValidTimeout)(this.#p)&&0!==this.#p&&(this.#h=u.timeoutManager.setInterval(()=>{(this.options.refetchIntervalInBackground||i.focusManager.isFocused())&&this.#g()},this.#p))}#m(){this.#O(),this.#R(this.#$())}#b(){this.#d&&(u.timeoutManager.clearTimeout(this.#d),this.#d=void 0)}#y(){this.#h&&(u.timeoutManager.clearInterval(this.#h),this.#h=void 0)}createResult(e,t){let r,i=this.#i,n=this.options,a=this.#s,u=this.#a,c=this.#l,h=e!==i?e.state:this.#n,{state:g}=e,m={...g},b=!1;if(t._optimisticResults){let r=this.hasListeners(),a=!r&&d(e,t),l=r&&p(e,i,t,n);(a||l)&&(m={...m,...(0,s.fetchState)(g.data,e.options)}),"isRestoring"===t._optimisticResults&&(m.fetchStatus="idle")}let{error:y,errorUpdatedAt:v,status:O}=m;r=m.data;let $=!1;if(void 0!==t.placeholderData&&void 0===r&&"pending"===O){let e;a?.isPlaceholderData&&t.placeholderData===c?.placeholderData?(e=a.data,$=!0):e="function"==typeof t.placeholderData?t.placeholderData(this.#c?.state.data,this.#c):t.placeholderData,void 0!==e&&(O="success",r=(0,o.replaceData)(a?.data,e,t),b=!0)}if(t.select&&void 0!==r&&!$)if(a&&r===u?.data&&t.select===this.#o)r=this.#u;else try{this.#o=t.select,r=t.select(r),r=(0,o.replaceData)(a?.data,r,t),this.#u=r,this.#t=null}catch(e){this.#t=e}this.#t&&(y=this.#t,r=this.#u,v=Date.now(),O="error");let R="fetching"===m.fetchStatus,C="pending"===O,w="error"===O,x=C&&R,E=void 0!==r,k={status:O,fetchStatus:m.fetchStatus,isPending:C,isSuccess:"success"===O,isError:w,isInitialLoading:x,isLoading:x,data:r,dataUpdatedAt:m.dataUpdatedAt,error:y,errorUpdatedAt:v,failureCount:m.fetchFailureCount,failureReason:m.fetchFailureReason,errorUpdateCount:m.errorUpdateCount,isFetched:m.dataUpdateCount>0||m.errorUpdateCount>0,isFetchedAfterMount:m.dataUpdateCount>h.dataUpdateCount||m.errorUpdateCount>h.errorUpdateCount,isFetching:R,isRefetching:R&&!C,isLoadingError:w&&!E,isPaused:"paused"===m.fetchStatus,isPlaceholderData:b,isRefetchError:w&&E,isStale:f(e,t),refetch:this.refetch,promise:this.#r,isEnabled:!1!==(0,o.resolveEnabled)(t.enabled,e)};if(this.options.experimental_prefetchInRender){let t=void 0!==k.data,r="error"===k.status&&!t,n=e=>{r?e.reject(k.error):t&&e.resolve(k.data)},s=()=>{n(this.#r=k.promise=(0,l.pendingThenable)())},a=this.#r;switch(a.status){case"pending":e.queryHash===i.queryHash&&n(a);break;case"fulfilled":(r||k.data!==a.value)&&s();break;case"rejected":r&&k.error===a.reason||s()}}return k}updateResult(){let e=this.#s,t=this.createResult(this.#i,this.options);if(this.#a=this.#i.state,this.#l=this.options,void 0!==this.#a.data&&(this.#c=this.#i),(0,o.shallowEqualObjects)(t,e))return;this.#s=t;let r=()=>{if(!e)return!0;let{notifyOnChangeProps:t}=this.options,r="function"==typeof t?t():t;if("all"===r||!r&&!this.#f.size)return!0;let i=new Set(r??this.#f);return this.options.throwOnError&&i.add("error"),Object.keys(this.#s).some(t=>this.#s[t]!==e[t]&&i.has(t))};this.#C({listeners:r()})}#v(){let e=this.#e.getQueryCache().build(this.#e,this.options);if(e===this.#i)return;let t=this.#i;this.#i=e,this.#n=e.state,this.hasListeners()&&(t?.removeObserver(this),e.addObserver(this))}onQueryUpdate(){this.updateResult(),this.hasListeners()&&this.#m()}#C(e){n.notifyManager.batch(()=>{e.listeners&&this.listeners.forEach(e=>{e(this.#s)}),this.#e.getQueryCache().notify({query:this.#i,type:"observerResultsUpdated"})})}};function d(e,t){return!1!==(0,o.resolveEnabled)(t.enabled,e)&&void 0===e.state.data&&("error"!==e.state.status||!1!==t.retryOnMount)||void 0!==e.state.data&&h(e,t,t.refetchOnMount)}function h(e,t,r){if(!1!==(0,o.resolveEnabled)(t.enabled,e)&&"static"!==(0,o.resolveStaleTime)(t.staleTime,e)){let i="function"==typeof r?r(e):r;return"always"===i||!1!==i&&f(e,t)}return!1}function p(e,t,r,i){return(e!==t||!1===(0,o.resolveEnabled)(i.enabled,e))&&(!r.suspense||"error"!==e.state.status)&&f(e,r)}function f(e,t){return!1!==(0,o.resolveEnabled)(t.enabled,e)&&e.isStaleByTime((0,o.resolveStaleTime)(t.staleTime,e))}e.s(["QueryObserver",()=>c],869230),e.i(247167);var g=e.i(271645),m=e.i(912598);e.i(843476);var b=g.createContext((t=!1,{clearReset:()=>{t=!1},reset:()=>{t=!0},isReset:()=>t})),y=g.createContext(!1);y.Provider;var v=(e,t,r)=>t.fetchOptimistic(e).catch(()=>{r.clearReset()});function O(e,t,r){let i,s=g.useContext(y),a=g.useContext(b),l=(0,m.useQueryClient)(r),u=l.defaultQueryOptions(e);l.getDefaultOptions().queries?._experimental_beforeQuery?.(u);let c=l.getQueryCache().get(u.queryHash);if(u._optimisticResults=s?"isRestoring":"optimistic",u.suspense){let e=e=>"static"===e?e:Math.max(e??1e3,1e3),t=u.staleTime;u.staleTime="function"==typeof t?(...r)=>e(t(...r)):e(t),"number"==typeof u.gcTime&&(u.gcTime=Math.max(u.gcTime,1e3))}i=c?.state.error&&"function"==typeof u.throwOnError?(0,o.shouldThrowError)(u.throwOnError,[c.state.error,c]):u.throwOnError,(u.suspense||u.experimental_prefetchInRender||i)&&!a.isReset()&&(u.retryOnMount=!1),g.useEffect(()=>{a.clearReset()},[a]);let d=!l.getQueryCache().get(u.queryHash),[h]=g.useState(()=>new t(l,u)),p=h.getOptimisticResult(u),f=!s&&!1!==e.subscribed;if(g.useSyncExternalStore(g.useCallback(e=>{let t=f?h.subscribe(n.notifyManager.batchCalls(e)):o.noop;return h.updateResult(),t},[h,f]),()=>h.getCurrentResult(),()=>h.getCurrentResult()),g.useEffect(()=>{h.setOptions(u)},[u,h]),u?.suspense&&p.isPending)throw v(u,h,a);if((({result:e,errorResetBoundary:t,throwOnError:r,query:i,suspense:n})=>e.isError&&!t.isReset()&&!e.isFetching&&i&&(n&&void 0===e.data||(0,o.shouldThrowError)(r,[e.error,i])))({result:p,errorResetBoundary:a,throwOnError:u.throwOnError,query:c,suspense:u.suspense}))throw p.error;if(l.getDefaultOptions().queries?._experimental_afterQuery?.(u,p),u.experimental_prefetchInRender&&!o.isServer&&p.isLoading&&p.isFetching&&!s){let e=d?v(u,h,a):c?.promise;e?.catch(o.noop).finally(()=>{h.updateResult()})}return u.notifyOnChangeProps?p:h.trackResult(p)}function $(e,t){return O(e,c,t)}function R(e){let t=[e];return{all:t,lists:()=>[...t,"list"],list:e=>[...t,"list",{params:e}],details:()=>[...t,"detail"],detail:e=>[...t,"detail",e]}}e.s(["useBaseQuery",()=>O],469637),e.s(["useQuery",()=>$],266027),e.s(["createQueryKeys",()=>R],243652);let C=R("uiConfig");e.s(["useUIConfig",0,()=>$({queryKey:C.list({}),queryFn:async()=>await (0,r.getUiConfig)(),staleTime:864e5,gcTime:864e5})],612256)},321836,e=>{"use strict";let t="litellm_return_url",r="redirect_to";function i(){return window.location.href}function n(){let e=i();e&&function(e,t,r=300){if("u"typeof document&&(document.cookie=`${t}=; path=/; max-age=0`)}catch(e){console.error("Failed to clear return URL cookie:",e)}}function l(){return new URLSearchParams(window.location.search).get(r)}function o(e,t){let n=t||i();if(!n||n.includes("/login"))return e;let s=e.includes("?")?"&":"?";return`${e}${s}${r}=${encodeURIComponent(n)}`}function u(){let e=l();if(e)return e;let t=s();return t||null}function c(){let e=window.location.hostname;return"localhost"===e||"127.0.0.1"===e||"::1"===e||e.startsWith("127.")||e.endsWith(".local")}function d(e){if(!e)return!1;if(e.startsWith("/")&&!e.startsWith("//"))return!0;try{let t=new URL(e),r=window.location.hostname;if(t.hostname!==r)return!1;if(c())return!0;return t.origin===window.location.origin}catch{return!1}}function h(e){try{let t=new URL(e,window.location.origin),r=t.pathname;r.length>1&&r.endsWith("/")&&(r=r.slice(0,-1));let i=new URLSearchParams(t.search),n=new URLSearchParams;Array.from(i.entries()).sort(([e],[t])=>e.localeCompare(t)).forEach(([e,t])=>{n.append(e,t)});let s=n.toString(),a=t.hash||"";return`${t.origin}${r}${s?`?${s}`:""}${a}`}catch{return e}}function p(){let e=l();if(e){if(d(e))return a(),e;c()&&console.warn("[returnUrlUtils] Invalid return URL in params rejected:",e)}let t=s();if(t){if(d(t))return a(),t;c()&&console.warn("[returnUrlUtils] Invalid return URL in cookie rejected:",t)}return null}e.s(["buildLoginUrlWithReturn",()=>o,"clearStoredReturnUrl",()=>a,"consumeReturnUrl",()=>p,"getReturnUrl",()=>u,"isValidReturnUrl",()=>d,"normalizeUrlForCompare",()=>h,"storeReturnUrl",()=>n])},185793,e=>{"use strict";e.i(247167);var t=e.i(271645),r=e.i(343794),i=e.i(242064),n=e.i(529681);let s=e=>{let{prefixCls:i,className:n,style:s,size:a,shape:l}=e,o=(0,r.default)({[`${i}-lg`]:"large"===a,[`${i}-sm`]:"small"===a}),u=(0,r.default)({[`${i}-circle`]:"circle"===l,[`${i}-square`]:"square"===l,[`${i}-round`]:"round"===l}),c=t.useMemo(()=>"number"==typeof a?{width:a,height:a,lineHeight:`${a}px`}:{},[a]);return t.createElement("span",{className:(0,r.default)(i,o,u,n),style:Object.assign(Object.assign({},c),s)})};e.i(296059);var a=e.i(694758),l=e.i(915654),o=e.i(246422),u=e.i(838378);let c=new a.Keyframes("ant-skeleton-loading",{"0%":{backgroundPosition:"100% 50%"},"100%":{backgroundPosition:"0 50%"}}),d=e=>({height:e,lineHeight:(0,l.unit)(e)}),h=e=>Object.assign({width:e},d(e)),p=(e,t)=>Object.assign({width:t(e).mul(5).equal(),minWidth:t(e).mul(5).equal()},d(e)),f=e=>Object.assign({width:e},d(e)),g=(e,t,r)=>{let{skeletonButtonCls:i}=e;return{[`${r}${i}-circle`]:{width:t,minWidth:t,borderRadius:"50%"},[`${r}${i}-round`]:{borderRadius:t}}},m=(e,t)=>Object.assign({width:t(e).mul(2).equal(),minWidth:t(e).mul(2).equal()},d(e)),b=(0,o.genStyleHooks)("Skeleton",e=>{let{componentCls:t,calc:r}=e;return(e=>{let{componentCls:t,skeletonAvatarCls:r,skeletonTitleCls:i,skeletonParagraphCls:n,skeletonButtonCls:s,skeletonInputCls:a,skeletonImageCls:l,controlHeight:o,controlHeightLG:u,controlHeightSM:d,gradientFromColor:b,padding:y,marginSM:v,borderRadius:O,titleHeight:$,blockRadius:R,paragraphLiHeight:C,controlHeightXS:w,paragraphMarginTop:x}=e;return{[t]:{display:"table",width:"100%",[`${t}-header`]:{display:"table-cell",paddingInlineEnd:y,verticalAlign:"top",[r]:Object.assign({display:"inline-block",verticalAlign:"top",background:b},h(o)),[`${r}-circle`]:{borderRadius:"50%"},[`${r}-lg`]:Object.assign({},h(u)),[`${r}-sm`]:Object.assign({},h(d))},[`${t}-content`]:{display:"table-cell",width:"100%",verticalAlign:"top",[i]:{width:"100%",height:$,background:b,borderRadius:R,[`+ ${n}`]:{marginBlockStart:d}},[n]:{padding:0,"> li":{width:"100%",height:C,listStyle:"none",background:b,borderRadius:R,"+ li":{marginBlockStart:w}}},[`${n}> li:last-child:not(:first-child):not(:nth-child(2))`]:{width:"61%"}},[`&-round ${t}-content`]:{[`${i}, ${n} > li`]:{borderRadius:O}}},[`${t}-with-avatar ${t}-content`]:{[i]:{marginBlockStart:v,[`+ ${n}`]:{marginBlockStart:x}}},[`${t}${t}-element`]:Object.assign(Object.assign(Object.assign(Object.assign({display:"inline-block",width:"auto"},(e=>{let{borderRadiusSM:t,skeletonButtonCls:r,controlHeight:i,controlHeightLG:n,controlHeightSM:s,gradientFromColor:a,calc:l}=e;return Object.assign(Object.assign(Object.assign(Object.assign(Object.assign({[r]:Object.assign({display:"inline-block",verticalAlign:"top",background:a,borderRadius:t,width:l(i).mul(2).equal(),minWidth:l(i).mul(2).equal()},m(i,l))},g(e,i,r)),{[`${r}-lg`]:Object.assign({},m(n,l))}),g(e,n,`${r}-lg`)),{[`${r}-sm`]:Object.assign({},m(s,l))}),g(e,s,`${r}-sm`))})(e)),(e=>{let{skeletonAvatarCls:t,gradientFromColor:r,controlHeight:i,controlHeightLG:n,controlHeightSM:s}=e;return{[t]:Object.assign({display:"inline-block",verticalAlign:"top",background:r},h(i)),[`${t}${t}-circle`]:{borderRadius:"50%"},[`${t}${t}-lg`]:Object.assign({},h(n)),[`${t}${t}-sm`]:Object.assign({},h(s))}})(e)),(e=>{let{controlHeight:t,borderRadiusSM:r,skeletonInputCls:i,controlHeightLG:n,controlHeightSM:s,gradientFromColor:a,calc:l}=e;return{[i]:Object.assign({display:"inline-block",verticalAlign:"top",background:a,borderRadius:r},p(t,l)),[`${i}-lg`]:Object.assign({},p(n,l)),[`${i}-sm`]:Object.assign({},p(s,l))}})(e)),(e=>{let{skeletonImageCls:t,imageSizeBase:r,gradientFromColor:i,borderRadiusSM:n,calc:s}=e;return{[t]:Object.assign(Object.assign({display:"inline-flex",alignItems:"center",justifyContent:"center",verticalAlign:"middle",background:i,borderRadius:n},f(s(r).mul(2).equal())),{[`${t}-path`]:{fill:"#bfbfbf"},[`${t}-svg`]:Object.assign(Object.assign({},f(r)),{maxWidth:s(r).mul(4).equal(),maxHeight:s(r).mul(4).equal()}),[`${t}-svg${t}-svg-circle`]:{borderRadius:"50%"}}),[`${t}${t}-circle`]:{borderRadius:"50%"}}})(e)),[`${t}${t}-block`]:{width:"100%",[s]:{width:"100%"},[a]:{width:"100%"}},[`${t}${t}-active`]:{[` + ${i}, + ${n} > li, + ${r}, + ${s}, + ${a}, + ${l} + `]:Object.assign({},{background:e.skeletonLoadingBackground,backgroundSize:"400% 100%",animationName:c,animationDuration:e.skeletonLoadingMotionDuration,animationTimingFunction:"ease",animationIterationCount:"infinite"})}}})((0,u.mergeToken)(e,{skeletonAvatarCls:`${t}-avatar`,skeletonTitleCls:`${t}-title`,skeletonParagraphCls:`${t}-paragraph`,skeletonButtonCls:`${t}-button`,skeletonInputCls:`${t}-input`,skeletonImageCls:`${t}-image`,imageSizeBase:r(e.controlHeight).mul(1.5).equal(),borderRadius:100,skeletonLoadingBackground:`linear-gradient(90deg, ${e.gradientFromColor} 25%, ${e.gradientToColor} 37%, ${e.gradientFromColor} 63%)`,skeletonLoadingMotionDuration:"1.4s"}))},e=>{let{colorFillContent:t,colorFill:r}=e;return{color:t,colorGradientEnd:r,gradientFromColor:t,gradientToColor:r,titleHeight:e.controlHeight/2,blockRadius:e.borderRadiusSM,paragraphMarginTop:e.marginLG+e.marginXXS,paragraphLiHeight:e.controlHeight/2}},{deprecatedTokens:[["color","gradientFromColor"],["colorGradientEnd","gradientToColor"]]}),y=e=>{let{prefixCls:i,className:n,style:s,rows:a=0}=e,l=Array.from({length:a}).map((r,i)=>t.createElement("li",{key:i,style:{width:((e,t)=>{let{width:r,rows:i=2}=t;return Array.isArray(r)?r[e]:i-1===e?r:void 0})(i,e)}}));return t.createElement("ul",{className:(0,r.default)(i,n),style:s},l)},v=({prefixCls:e,className:i,width:n,style:s})=>t.createElement("h3",{className:(0,r.default)(e,i),style:Object.assign({width:n},s)});function O(e){return e&&"object"==typeof e?e:{}}let $=e=>{let{prefixCls:n,loading:a,className:l,rootClassName:o,style:u,children:c,avatar:d=!1,title:h=!0,paragraph:p=!0,active:f,round:g}=e,{getPrefixCls:m,direction:$,className:R,style:C}=(0,i.useComponentConfig)("skeleton"),w=m("skeleton",n),[x,E,k]=b(w);if(a||!("loading"in e)){let e,i,n=!!d,a=!!h,c=!!p;if(n){let r=Object.assign(Object.assign({prefixCls:`${w}-avatar`},a&&!c?{size:"large",shape:"square"}:{size:"large",shape:"circle"}),O(d));e=t.createElement("div",{className:`${w}-header`},t.createElement(s,Object.assign({},r)))}if(a||c){let e,r;if(a){let r=Object.assign(Object.assign({prefixCls:`${w}-title`},!n&&c?{width:"38%"}:n&&c?{width:"50%"}:{}),O(h));e=t.createElement(v,Object.assign({},r))}if(c){let e,i=Object.assign(Object.assign({prefixCls:`${w}-paragraph`},(e={},n&&a||(e.width="61%"),!n&&a?e.rows=3:e.rows=2,e)),O(p));r=t.createElement(y,Object.assign({},i))}i=t.createElement("div",{className:`${w}-content`},e,r)}let m=(0,r.default)(w,{[`${w}-with-avatar`]:n,[`${w}-active`]:f,[`${w}-rtl`]:"rtl"===$,[`${w}-round`]:g},R,l,o,E,k);return x(t.createElement("div",{className:m,style:Object.assign(Object.assign({},C),u)},e,i))}return null!=c?c:null};$.Button=e=>{let{prefixCls:a,className:l,rootClassName:o,active:u,block:c=!1,size:d="default"}=e,{getPrefixCls:h}=t.useContext(i.ConfigContext),p=h("skeleton",a),[f,g,m]=b(p),y=(0,n.default)(e,["prefixCls"]),v=(0,r.default)(p,`${p}-element`,{[`${p}-active`]:u,[`${p}-block`]:c},l,o,g,m);return f(t.createElement("div",{className:v},t.createElement(s,Object.assign({prefixCls:`${p}-button`,size:d},y))))},$.Avatar=e=>{let{prefixCls:a,className:l,rootClassName:o,active:u,shape:c="circle",size:d="default"}=e,{getPrefixCls:h}=t.useContext(i.ConfigContext),p=h("skeleton",a),[f,g,m]=b(p),y=(0,n.default)(e,["prefixCls","className"]),v=(0,r.default)(p,`${p}-element`,{[`${p}-active`]:u},l,o,g,m);return f(t.createElement("div",{className:v},t.createElement(s,Object.assign({prefixCls:`${p}-avatar`,shape:c,size:d},y))))},$.Input=e=>{let{prefixCls:a,className:l,rootClassName:o,active:u,block:c,size:d="default"}=e,{getPrefixCls:h}=t.useContext(i.ConfigContext),p=h("skeleton",a),[f,g,m]=b(p),y=(0,n.default)(e,["prefixCls"]),v=(0,r.default)(p,`${p}-element`,{[`${p}-active`]:u,[`${p}-block`]:c},l,o,g,m);return f(t.createElement("div",{className:v},t.createElement(s,Object.assign({prefixCls:`${p}-input`,size:d},y))))},$.Image=e=>{let{prefixCls:n,className:s,rootClassName:a,style:l,active:o}=e,{getPrefixCls:u}=t.useContext(i.ConfigContext),c=u("skeleton",n),[d,h,p]=b(c),f=(0,r.default)(c,`${c}-element`,{[`${c}-active`]:o},s,a,h,p);return d(t.createElement("div",{className:f},t.createElement("div",{className:(0,r.default)(`${c}-image`,s),style:l},t.createElement("svg",{viewBox:"0 0 1098 1024",xmlns:"http://www.w3.org/2000/svg",className:`${c}-image-svg`},t.createElement("title",null,"Image placeholder"),t.createElement("path",{d:"M365.714286 329.142857q0 45.714286-32.036571 77.677714t-77.677714 32.036571-77.677714-32.036571-32.036571-77.677714 32.036571-77.677714 77.677714-32.036571 77.677714 32.036571 32.036571 77.677714zM950.857143 548.571429l0 256-804.571429 0 0-109.714286 182.857143-182.857143 91.428571 91.428571 292.571429-292.571429zM1005.714286 146.285714l-914.285714 0q-7.460571 0-12.873143 5.412571t-5.412571 12.873143l0 694.857143q0 7.460571 5.412571 12.873143t12.873143 5.412571l914.285714 0q7.460571 0 12.873143-5.412571t5.412571-12.873143l0-694.857143q0-7.460571-5.412571-12.873143t-12.873143-5.412571zM1097.142857 164.571429l0 694.857143q0 37.741714-26.843429 64.585143t-64.585143 26.843429l-914.285714 0q-37.741714 0-64.585143-26.843429t-26.843429-64.585143l0-694.857143q0-37.741714 26.843429-64.585143t64.585143-26.843429l914.285714 0q37.741714 0 64.585143 26.843429t26.843429 64.585143z",className:`${c}-image-path`})))))},$.Node=e=>{let{prefixCls:n,className:s,rootClassName:a,style:l,active:o,children:u}=e,{getPrefixCls:c}=t.useContext(i.ConfigContext),d=c("skeleton",n),[h,p,f]=b(d),g=(0,r.default)(d,`${d}-element`,{[`${d}-active`]:o},p,s,a,f);return h(t.createElement("div",{className:g},t.createElement("div",{className:(0,r.default)(`${d}-image`,s),style:l},u)))},e.s(["default",0,$],185793)},959013,e=>{"use strict";e.i(247167);var t=e.i(931067),r=e.i(271645);let i={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M482 152h60q8 0 8 8v704q0 8-8 8h-60q-8 0-8-8V160q0-8 8-8z"}},{tag:"path",attrs:{d:"M192 474h672q8 0 8 8v60q0 8-8 8H160q-8 0-8-8v-60q0-8 8-8z"}}]},name:"plus",theme:"outlined"};var n=e.i(9583),s=r.forwardRef(function(e,s){return r.createElement(n.default,(0,t.default)({},e,{ref:s,icon:i}))});e.s(["default",0,s],959013)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/0377ae18aae60c57.js b/litellm/proxy/_experimental/out/_next/static/chunks/0377ae18aae60c57.js deleted file mode 100644 index 66e4d15294f..00000000000 --- a/litellm/proxy/_experimental/out/_next/static/chunks/0377ae18aae60c57.js +++ /dev/null @@ -1 +0,0 @@ -(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,362133,457202,439061,182399,234779,374615,330995,592143,372943,899268,87316,655900,299023,25652,882293,e=>{"use strict";e.i(247167);var t=e.i(931067),a=e.i(271645);let s={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M908 640H804V488c0-4.4-3.6-8-8-8H548v-96h108c8.8 0 16-7.2 16-16V80c0-8.8-7.2-16-16-16H368c-8.8 0-16 7.2-16 16v288c0 8.8 7.2 16 16 16h108v96H228c-4.4 0-8 3.6-8 8v152H116c-8.8 0-16 7.2-16 16v288c0 8.8 7.2 16 16 16h288c8.8 0 16-7.2 16-16V656c0-8.8-7.2-16-16-16H292v-88h440v88H620c-8.8 0-16 7.2-16 16v288c0 8.8 7.2 16 16 16h288c8.8 0 16-7.2 16-16V656c0-8.8-7.2-16-16-16zm-564 76v168H176V716h168zm84-408V140h168v168H428zm420 576H680V716h168v168z"}}]},name:"apartment",theme:"outlined"};var l=e.i(9583),r=a.forwardRef(function(e,r){return a.createElement(l.default,(0,t.default)({},e,{ref:r,icon:s}))});e.s(["ApartmentOutlined",0,r],362133);let i={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M296 250c-4.4 0-8 3.6-8 8v48c0 4.4 3.6 8 8 8h384c4.4 0 8-3.6 8-8v-48c0-4.4-3.6-8-8-8H296zm184 144H296c-4.4 0-8 3.6-8 8v48c0 4.4 3.6 8 8 8h184c4.4 0 8-3.6 8-8v-48c0-4.4-3.6-8-8-8zm-48 458H208V148h560v320c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8V108c0-17.7-14.3-32-32-32H168c-17.7 0-32 14.3-32 32v784c0 17.7 14.3 32 32 32h264c4.4 0 8-3.6 8-8v-56c0-4.4-3.6-8-8-8zm440-88H728v-36.6c46.3-13.8 80-56.6 80-107.4 0-61.9-50.1-112-112-112s-112 50.1-112 112c0 50.7 33.7 93.6 80 107.4V764H520c-8.8 0-16 7.2-16 16v152c0 8.8 7.2 16 16 16h352c8.8 0 16-7.2 16-16V780c0-8.8-7.2-16-16-16zM646 620c0-27.6 22.4-50 50-50s50 22.4 50 50-22.4 50-50 50-50-22.4-50-50zm180 266H566v-60h260v60z"}}]},name:"audit",theme:"outlined"};var n=a.forwardRef(function(e,s){return a.createElement(l.default,(0,t.default)({},e,{ref:s,icon:i}))});e.s(["AuditOutlined",0,n],457202);let o={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M766.4 744.3c43.7 0 79.4-36.2 79.4-80.5 0-53.5-79.4-140.8-79.4-140.8S687 610.3 687 663.8c0 44.3 35.7 80.5 79.4 80.5zm-377.1-44.1c7.1 7.1 18.6 7.1 25.6 0l256.1-256c7.1-7.1 7.1-18.6 0-25.6l-256-256c-.6-.6-1.3-1.2-2-1.7l-78.2-78.2a9.11 9.11 0 00-12.8 0l-48 48a9.11 9.11 0 000 12.8l67.2 67.2-207.8 207.9c-7.1 7.1-7.1 18.6 0 25.6l255.9 256zm12.9-448.6l178.9 178.9H223.4l178.8-178.9zM904 816H120c-4.4 0-8 3.6-8 8v80c0 4.4 3.6 8 8 8h784c4.4 0 8-3.6 8-8v-80c0-4.4-3.6-8-8-8z"}}]},name:"bg-colors",theme:"outlined"};var d=a.forwardRef(function(e,s){return a.createElement(l.default,(0,t.default)({},e,{ref:s,icon:o}))});e.s(["BgColorsOutlined",0,d],439061);let c={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M856 376H648V168c0-8.8-7.2-16-16-16H168c-8.8 0-16 7.2-16 16v464c0 8.8 7.2 16 16 16h208v208c0 8.8 7.2 16 16 16h464c8.8 0 16-7.2 16-16V392c0-8.8-7.2-16-16-16zm-480 16v188H220V220h360v156H392c-8.8 0-16 7.2-16 16zm204 52v136H444V444h136zm224 360H444V648h188c8.8 0 16-7.2 16-16V444h156v360z"}}]},name:"block",theme:"outlined"};var m=a.forwardRef(function(e,s){return a.createElement(l.default,(0,t.default)({},e,{ref:s,icon:c}))});e.s(["BlockOutlined",0,m],182399);let u={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M832 64H192c-17.7 0-32 14.3-32 32v832c0 17.7 14.3 32 32 32h640c17.7 0 32-14.3 32-32V96c0-17.7-14.3-32-32-32zm-260 72h96v209.9L621.5 312 572 347.4V136zm220 752H232V136h280v296.9c0 3.3 1 6.6 3 9.3a15.9 15.9 0 0022.3 3.7l83.8-59.9 81.4 59.4c2.7 2 6 3.1 9.4 3.1 8.8 0 16-7.2 16-16V136h64v752z"}}]},name:"book",theme:"outlined"};var g=a.forwardRef(function(e,s){return a.createElement(l.default,(0,t.default)({},e,{ref:s,icon:u}))});e.s(["BookOutlined",0,g],234779);let x={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M928 160H96c-17.7 0-32 14.3-32 32v640c0 17.7 14.3 32 32 32h832c17.7 0 32-14.3 32-32V192c0-17.7-14.3-32-32-32zm-792 72h752v120H136V232zm752 560H136V440h752v352zm-237-64h165c4.4 0 8-3.6 8-8v-72c0-4.4-3.6-8-8-8H651c-4.4 0-8 3.6-8 8v72c0 4.4 3.6 8 8 8z"}}]},name:"credit-card",theme:"outlined"};var p=a.forwardRef(function(e,s){return a.createElement(l.default,(0,t.default)({},e,{ref:s,icon:x}))});e.s(["CreditCardOutlined",0,p],374615);var h=e.i(366845);e.s(["FolderOutlined",()=>h.default],330995);var f=e.i(609587);e.s(["ConfigProvider",()=>f.default],592143);var y=e.i(8211),b=e.i(343794),v=e.i(529681),j=e.i(242064),N=e.i(704914),k=e.i(876556),w=e.i(290224),O=e.i(251224),_=function(e,t){var a={};for(var s in e)Object.prototype.hasOwnProperty.call(e,s)&&0>t.indexOf(s)&&(a[s]=e[s]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var l=0,s=Object.getOwnPropertySymbols(e);lt.indexOf(s[l])&&Object.prototype.propertyIsEnumerable.call(e,s[l])&&(a[s[l]]=e[s[l]]);return a};function L({suffixCls:e,tagName:t,displayName:s}){return s=>a.forwardRef((l,r)=>a.createElement(s,Object.assign({ref:r,suffixCls:e,tagName:t},l)))}let C=a.forwardRef((e,t)=>{let{prefixCls:s,suffixCls:l,className:r,tagName:i}=e,n=_(e,["prefixCls","suffixCls","className","tagName"]),{getPrefixCls:o}=a.useContext(j.ConfigContext),d=o("layout",s),[c,m,u]=(0,O.default)(d),g=l?`${d}-${l}`:d;return c(a.createElement(i,Object.assign({className:(0,b.default)(s||g,r,m,u),ref:t},n)))}),S=a.forwardRef((e,t)=>{let{direction:s}=a.useContext(j.ConfigContext),[l,r]=a.useState([]),{prefixCls:i,className:n,rootClassName:o,children:d,hasSider:c,tagName:m,style:u}=e,g=_(e,["prefixCls","className","rootClassName","children","hasSider","tagName","style"]),x=(0,v.default)(g,["suffixCls"]),{getPrefixCls:p,className:h,style:f}=(0,j.useComponentConfig)("layout"),L=p("layout",i),C="boolean"==typeof c?c:!!l.length||(0,k.default)(d).some(e=>e.type===w.default),[S,M,P]=(0,O.default)(L),H=(0,b.default)(L,{[`${L}-has-sider`]:C,[`${L}-rtl`]:"rtl"===s},h,n,o,M,P),z=a.useMemo(()=>({siderHook:{addSider:e=>{r(t=>[].concat((0,y.default)(t),[e]))},removeSider:e=>{r(t=>t.filter(t=>t!==e))}}}),[]);return S(a.createElement(N.LayoutContext.Provider,{value:z},a.createElement(m,Object.assign({ref:t,className:H,style:Object.assign(Object.assign({},f),u)},x),d)))}),M=L({tagName:"div",displayName:"Layout"})(S),P=L({suffixCls:"header",tagName:"header",displayName:"Header"})(C),H=L({suffixCls:"footer",tagName:"footer",displayName:"Footer"})(C),z=L({suffixCls:"content",tagName:"main",displayName:"Content"})(C);M.Header=P,M.Footer=H,M.Content=z,M.Sider=w.default,M._InternalSiderContext=w.SiderContext,e.s(["Layout",0,M],372943);var T=e.i(60699);e.s(["Menu",()=>T.default],899268);var R=e.i(475254);let E=(0,R.default)("calendar",[["path",{d:"M8 2v4",key:"1cmpym"}],["path",{d:"M16 2v4",key:"4m81vk"}],["rect",{width:"18",height:"18",x:"3",y:"4",rx:"2",key:"1hopcy"}],["path",{d:"M3 10h18",key:"8toen8"}]]);e.s(["Calendar",()=>E],87316);var U=e.i(399219);e.s(["ChevronUp",()=>U.default],655900);let V=(0,R.default)("minus",[["path",{d:"M5 12h14",key:"1ays0h"}]]);e.s(["Minus",()=>V],299023);let A=(0,R.default)("trending-up",[["path",{d:"M16 7h6v6",key:"box55l"}],["path",{d:"m22 7-8.5 8.5-5-5L2 17",key:"1t1m79"}]]);e.s(["TrendingUp",()=>A],25652);let B=(0,R.default)("user-check",[["path",{d:"m16 11 2 2 4-4",key:"9rsbq5"}],["path",{d:"M16 21v-2a4 4 0 0 0-4-4H6a4 4 0 0 0-4 4v2",key:"1yyitq"}],["circle",{cx:"9",cy:"7",r:"4",key:"nufk8"}]]);e.s(["UserCheck",()=>B],882293)},761911,98740,e=>{"use strict";let t=(0,e.i(475254).default)("users",[["path",{d:"M16 21v-2a4 4 0 0 0-4-4H6a4 4 0 0 0-4 4v2",key:"1yyitq"}],["path",{d:"M16 3.128a4 4 0 0 1 0 7.744",key:"16gr8j"}],["path",{d:"M22 21v-2a4 4 0 0 0-3-3.87",key:"kshegd"}],["circle",{cx:"9",cy:"7",r:"4",key:"nufk8"}]]);e.s(["default",()=>t],98740),e.s(["Users",()=>t],761911)},111672,e=>{"use strict";var t=e.i(247167),a=e.i(843476),s=e.i(109799),l=e.i(785242),r=e.i(135214),i=e.i(218129),n=e.i(362133),o=e.i(477189),d=e.i(457202),c=e.i(299251),m=e.i(153702),u=e.i(439061),g=e.i(182399),x=e.i(234779),p=e.i(374615),h=e.i(210612),f=e.i(19732),y=e.i(872934),b=e.i(993914),v=e.i(330995),j=e.i(438957),N=e.i(777579),k=e.i(788191),w=e.i(983561),O=e.i(602073),_=e.i(928685),L=e.i(313603),C=e.i(232164),S=e.i(645526),M=e.i(366308),P=e.i(771674),H=e.i(592143),z=e.i(372943),T=e.i(899268),R=e.i(271645),E=e.i(708347),U=e.i(844444),V=e.i(371401);e.i(389083);var A=e.i(878894),B=e.i(87316);e.i(664659),e.i(655900);var $=e.i(531278),I=e.i(299023),D=e.i(25652),K=e.i(882293),F=e.i(761911),W=e.i(764205);let G=(...e)=>e.filter(Boolean).join(" ");function q({accessToken:e,width:t=220}){let s=(0,V.useDisableUsageIndicator)(),[l,r]=(0,R.useState)(!1),[i,n]=(0,R.useState)(!1),[o,d]=(0,R.useState)(null),[c,m]=(0,R.useState)(null),[u,g]=(0,R.useState)(!1),[x,p]=(0,R.useState)(null);(0,R.useEffect)(()=>{(async()=>{if(e){g(!0),p(null);try{let[t,a]=await Promise.all([(0,W.getRemainingUsers)(e),(0,W.getLicenseInfo)(e).catch(()=>null)]);d(t),m(a)}catch(e){console.error("Failed to fetch usage data:",e),p("Failed to load usage data")}finally{g(!1)}}})()},[e]);let h=c?.expiration_date?(e=>{if(!e)return null;let t=new Date(e+"T00:00:00Z"),a=new Date;return a.setHours(0,0,0,0),Math.ceil((t.getTime()-a.getTime())/864e5)})(c.expiration_date):null,f=null!==h&&h<0,y=null!==h&&h>=0&&h<30,{isOverLimit:b,isNearLimit:v,usagePercentage:j,userMetrics:N,teamMetrics:k}=(e=>{if(!e)return{isOverLimit:!1,isNearLimit:!1,usagePercentage:0,userMetrics:{isOverLimit:!1,isNearLimit:!1,usagePercentage:0},teamMetrics:{isOverLimit:!1,isNearLimit:!1,usagePercentage:0}};let t=e.total_users?e.total_users_used/e.total_users*100:0,a=t>100,s=t>=80&&t<=100,l=e.total_teams?e.total_teams_used/e.total_teams*100:0,r=l>100,i=l>=80&&l<=100,n=a||r;return{isOverLimit:n,isNearLimit:(s||i)&&!n,usagePercentage:Math.max(t,l),userMetrics:{isOverLimit:a,isNearLimit:s,usagePercentage:t},teamMetrics:{isOverLimit:r,isNearLimit:i,usagePercentage:l}}})(o),w=b||v||f||y,O=b||f,_=(v||y)&&!O;return s||!e||o?.total_users===null&&o?.total_teams===null?null:(0,a.jsx)("div",{className:"fixed bottom-4 left-4 z-50",style:{width:`${Math.min(t,220)}px`},children:(0,a.jsx)(()=>i?(0,a.jsx)("button",{onClick:()=>n(!1),className:G("bg-white border border-gray-200 rounded-lg shadow-sm p-3 hover:shadow-md transition-all w-full"),title:"Show usage details",children:(0,a.jsxs)("div",{className:"flex items-center gap-2",children:[(0,a.jsx)(F.Users,{className:"h-4 w-4 flex-shrink-0"}),w&&(0,a.jsx)("span",{className:"flex-shrink-0",children:O?(0,a.jsx)(A.AlertTriangle,{className:"h-3 w-3"}):_?(0,a.jsx)(D.TrendingUp,{className:"h-3 w-3"}):null}),(0,a.jsxs)("div",{className:"flex items-center gap-2 text-sm font-medium truncate",children:[o&&null!==o.total_users&&(0,a.jsxs)("span",{className:G("flex-shrink-0 px-1.5 py-0.5 rounded text-xs border",N.isOverLimit&&"bg-red-50 text-red-700 border-red-200",N.isNearLimit&&"bg-yellow-50 text-yellow-700 border-yellow-200",!N.isOverLimit&&!N.isNearLimit&&"bg-gray-50 text-gray-700 border-gray-200"),children:["U: ",o.total_users_used,"/",o.total_users]}),o&&null!==o.total_teams&&(0,a.jsxs)("span",{className:G("flex-shrink-0 px-1.5 py-0.5 rounded text-xs border",k.isOverLimit&&"bg-red-50 text-red-700 border-red-200",k.isNearLimit&&"bg-yellow-50 text-yellow-700 border-yellow-200",!k.isOverLimit&&!k.isNearLimit&&"bg-gray-50 text-gray-700 border-gray-200"),children:["T: ",o.total_teams_used,"/",o.total_teams]}),c?.expiration_date&&null!==h&&(0,a.jsx)("span",{className:G("flex-shrink-0 px-1.5 py-0.5 rounded text-xs border",f&&"bg-red-50 text-red-700 border-red-200",y&&"bg-yellow-50 text-yellow-700 border-yellow-200",!f&&!y&&"bg-gray-50 text-gray-700 border-gray-200"),children:h<0?"Exp!":`${h}d`}),!o||null===o.total_users&&null===o.total_teams&&!c&&(0,a.jsx)("span",{className:"truncate",children:"Usage"})]})]})}):u?(0,a.jsx)("div",{className:"bg-white border border-gray-200 rounded-lg shadow-sm p-4 w-full",children:(0,a.jsxs)("div",{className:"flex items-center justify-center gap-2 py-2",children:[(0,a.jsx)($.Loader2,{className:"h-4 w-4 animate-spin"}),(0,a.jsx)("span",{className:"text-sm text-gray-500 truncate",children:"Loading..."})]})}):x||!o?(0,a.jsx)("div",{className:"bg-white border border-gray-200 rounded-lg shadow-sm p-4 group w-full",children:(0,a.jsxs)("div",{className:"flex items-center justify-between gap-2",children:[(0,a.jsx)("div",{className:"flex-1 min-w-0",children:(0,a.jsx)("span",{className:"text-sm text-gray-500 truncate block",children:x||"No data"})}),(0,a.jsx)("button",{onClick:()=>n(!0),className:"opacity-0 group-hover:opacity-100 p-1 hover:bg-gray-100 rounded transition-all flex-shrink-0",title:"Minimize",children:(0,a.jsx)(I.Minus,{className:"h-3 w-3 text-gray-400"})})]})}):(0,a.jsxs)("div",{className:G("bg-white border rounded-lg shadow-sm p-3 transition-all duration-200 group w-full"),children:[(0,a.jsxs)("div",{className:"flex items-center justify-between gap-2 mb-3",children:[(0,a.jsxs)("div",{className:"flex items-center gap-2 min-w-0 flex-1",children:[(0,a.jsx)(F.Users,{className:"h-4 w-4 flex-shrink-0"}),(0,a.jsx)("span",{className:"font-medium text-sm truncate",children:"Usage"})]}),(0,a.jsx)("button",{onClick:()=>n(!0),className:"opacity-0 group-hover:opacity-100 p-1 hover:bg-gray-100 rounded transition-all flex-shrink-0",title:"Minimize",children:(0,a.jsx)(I.Minus,{className:"h-3 w-3 text-gray-400"})})]}),(0,a.jsxs)("div",{className:"space-y-3 text-sm",children:[c?.has_license&&c.expiration_date&&(0,a.jsxs)("div",{className:G("space-y-1 border rounded-md p-2",f&&"border-red-200 bg-red-50",y&&"border-yellow-200 bg-yellow-50"),children:[(0,a.jsxs)("div",{className:"flex items-center gap-2 text-xs text-gray-600 mb-1",children:[(0,a.jsx)(B.Calendar,{className:"h-3 w-3"}),(0,a.jsx)("span",{className:"font-medium",children:"License"}),(0,a.jsx)("span",{className:G("ml-1 px-1.5 py-0.5 rounded border",f&&"bg-red-50 text-red-700 border-red-200",y&&"bg-yellow-50 text-yellow-700 border-yellow-200",!f&&!y&&"bg-gray-50 text-gray-600 border-gray-200"),children:f?"Expired":y?"Expiring soon":"OK"})]}),(0,a.jsxs)("div",{className:"flex justify-between items-center",children:[(0,a.jsx)("span",{className:"text-gray-600 text-xs",children:"Status:"}),(0,a.jsx)("span",{className:G("font-medium text-right",f&&"text-red-600",y&&"text-yellow-600"),children:(e=>{if(null===e)return"No expiration";if(e<0)return"Expired";if(0===e)return"Expires today";if(1===e)return"1 day remaining";if(e<30)return`${e} days remaining`;if(e<60)return"1 month remaining";let t=Math.floor(e/30);return`${t} months remaining`})(h)})]}),c.license_type&&(0,a.jsxs)("div",{className:"flex justify-between items-center",children:[(0,a.jsx)("span",{className:"text-gray-600 text-xs",children:"Type:"}),(0,a.jsx)("span",{className:"font-medium text-right capitalize",children:c.license_type})]})]}),null!==o.total_users&&(0,a.jsxs)("div",{className:G("space-y-1 border rounded-md p-2",N.isOverLimit&&"border-red-200 bg-red-50",N.isNearLimit&&"border-yellow-200 bg-yellow-50"),children:[(0,a.jsxs)("div",{className:"flex items-center gap-2 text-xs text-gray-600 mb-1",children:[(0,a.jsx)(F.Users,{className:"h-3 w-3"}),(0,a.jsx)("span",{className:"font-medium",children:"Users"}),(0,a.jsx)("span",{className:G("ml-1 px-1.5 py-0.5 rounded border",N.isOverLimit&&"bg-red-50 text-red-700 border-red-200",N.isNearLimit&&"bg-yellow-50 text-yellow-700 border-yellow-200",!N.isOverLimit&&!N.isNearLimit&&"bg-gray-50 text-gray-600 border-gray-200"),children:N.isOverLimit?"Over limit":N.isNearLimit?"Near limit":"OK"})]}),(0,a.jsxs)("div",{className:"flex justify-between items-center",children:[(0,a.jsx)("span",{className:"text-gray-600 text-xs",children:"Used:"}),(0,a.jsxs)("span",{className:"font-medium text-right",children:[o.total_users_used,"/",o.total_users]})]}),(0,a.jsxs)("div",{className:"flex justify-between items-center",children:[(0,a.jsx)("span",{className:"text-gray-600 text-xs",children:"Remaining:"}),(0,a.jsx)("span",{className:G("font-medium text-right",N.isOverLimit&&"text-red-600",N.isNearLimit&&"text-yellow-600"),children:o.total_users_remaining})]}),(0,a.jsxs)("div",{className:"flex justify-between items-center",children:[(0,a.jsx)("span",{className:"text-gray-600 text-xs",children:"Usage:"}),(0,a.jsxs)("span",{className:"font-medium text-right",children:[Math.round(N.usagePercentage),"%"]})]}),(0,a.jsx)("div",{className:"w-full bg-gray-200 rounded-full h-2",children:(0,a.jsx)("div",{className:G("h-2 rounded-full transition-all duration-300",N.isOverLimit&&"bg-red-500",N.isNearLimit&&"bg-yellow-500",!N.isOverLimit&&!N.isNearLimit&&"bg-green-500"),style:{width:`${Math.min(N.usagePercentage,100)}%`}})})]}),null!==o.total_teams&&(0,a.jsxs)("div",{className:G("space-y-1 border rounded-md p-2",k.isOverLimit&&"border-red-200 bg-red-50",k.isNearLimit&&"border-yellow-200 bg-yellow-50"),children:[(0,a.jsxs)("div",{className:"flex items-center gap-2 text-xs text-gray-600 mb-1",children:[(0,a.jsx)(K.UserCheck,{className:"h-3 w-3"}),(0,a.jsx)("span",{className:"font-medium",children:"Teams"}),(0,a.jsx)("span",{className:G("ml-1 px-1.5 py-0.5 rounded border",k.isOverLimit&&"bg-red-50 text-red-700 border-red-200",k.isNearLimit&&"bg-yellow-50 text-yellow-700 border-yellow-200",!k.isOverLimit&&!k.isNearLimit&&"bg-gray-50 text-gray-600 border-gray-200"),children:k.isOverLimit?"Over limit":k.isNearLimit?"Near limit":"OK"})]}),(0,a.jsxs)("div",{className:"flex justify-between items-center",children:[(0,a.jsx)("span",{className:"text-gray-600 text-xs",children:"Used:"}),(0,a.jsxs)("span",{className:"font-medium text-right",children:[o.total_teams_used,"/",o.total_teams]})]}),(0,a.jsxs)("div",{className:"flex justify-between items-center",children:[(0,a.jsx)("span",{className:"text-gray-600 text-xs",children:"Remaining:"}),(0,a.jsx)("span",{className:G("font-medium text-right",k.isOverLimit&&"text-red-600",k.isNearLimit&&"text-yellow-600"),children:o.total_teams_remaining})]}),(0,a.jsxs)("div",{className:"flex justify-between items-center",children:[(0,a.jsx)("span",{className:"text-gray-600 text-xs",children:"Usage:"}),(0,a.jsxs)("span",{className:"font-medium text-right",children:[Math.round(k.usagePercentage),"%"]})]}),(0,a.jsx)("div",{className:"w-full bg-gray-200 rounded-full h-2",children:(0,a.jsx)("div",{className:G("h-2 rounded-full transition-all duration-300",k.isOverLimit&&"bg-red-500",k.isNearLimit&&"bg-yellow-500",!k.isOverLimit&&!k.isNearLimit&&"bg-green-500"),style:{width:`${Math.min(k.usagePercentage,100)}%`}})})]})]})]}),{})})}let{Sider:Y}=z.Layout,X={"api-reference":"api-reference"},Z=[{groupLabel:"AI GATEWAY",items:[{key:"api-keys",page:"api-keys",label:"Virtual Keys",icon:(0,a.jsx)(j.KeyOutlined,{})},{key:"llm-playground",page:"llm-playground",label:"Playground",icon:(0,a.jsx)(k.PlayCircleOutlined,{}),roles:E.rolesWithWriteAccess},{key:"models",page:"models",label:"Models + Endpoints",icon:(0,a.jsx)(g.BlockOutlined,{}),roles:E.rolesAllowedToViewWriteScopedPages},{key:"agentic",page:"agentic",label:"Agentic",icon:(0,a.jsx)(w.RobotOutlined,{}),children:[{key:"agents",page:"agents",label:"Agents",icon:(0,a.jsx)(w.RobotOutlined,{}),roles:E.rolesAllowedToViewWriteScopedPages},{key:"workflows",page:"workflows",label:"Workflow Runs",icon:(0,a.jsx)(n.ApartmentOutlined,{})},{key:"memory",page:"memory",label:"Memory",icon:(0,a.jsx)(x.BookOutlined,{})}]},{key:"mcp-servers",page:"mcp-servers",label:"MCP Servers",icon:(0,a.jsx)(M.ToolOutlined,{})},{key:"skills",page:"skills",label:"Skills",icon:(0,a.jsx)(i.ApiOutlined,{}),roles:E.all_admin_roles},{key:"guardrails",page:"guardrails",label:"Guardrails",icon:(0,a.jsx)(O.SafetyOutlined,{})},{key:"policies",page:"policies",label:(0,a.jsx)("span",{className:"flex items-center gap-4",children:"Policies"}),icon:(0,a.jsx)(d.AuditOutlined,{}),roles:E.all_admin_roles},{key:"tools",page:"tools",label:"Tools",icon:(0,a.jsx)(M.ToolOutlined,{}),children:[{key:"search-tools",page:"search-tools",label:"Search Tools",icon:(0,a.jsx)(_.SearchOutlined,{})},{key:"vector-stores",page:"vector-stores",label:"Vector Stores",icon:(0,a.jsx)(h.DatabaseOutlined,{})},{key:"tool-policies",page:"tool-policies",label:"Tool Policies",icon:(0,a.jsx)(O.SafetyOutlined,{})}]}]},{groupLabel:"OBSERVABILITY",items:[{key:"new_usage",page:"new_usage",icon:(0,a.jsx)(m.BarChartOutlined,{}),roles:[...E.all_admin_roles,...E.internalUserRoles],label:"Usage"},{key:"logs",page:"logs",label:"Logs",icon:(0,a.jsx)(N.LineChartOutlined,{})},{key:"guardrails-monitor",page:"guardrails-monitor",label:"Guardrails Monitor",icon:(0,a.jsx)(O.SafetyOutlined,{}),roles:[...E.all_admin_roles,...E.internalUserRoles]}]},{groupLabel:"ACCESS CONTROL",items:[{key:"teams",page:"teams",label:"Teams",icon:(0,a.jsx)(S.TeamOutlined,{})},{key:"projects",page:"projects",label:(0,a.jsxs)("span",{className:"flex items-center gap-2",children:["Projects ",(0,a.jsx)(U.default,{})]}),icon:(0,a.jsx)(v.FolderOutlined,{}),roles:E.all_admin_roles},{key:"users",page:"users",label:"Internal Users",icon:(0,a.jsx)(P.UserOutlined,{}),roles:E.all_admin_roles},{key:"organizations",page:"organizations",label:"Organizations",icon:(0,a.jsx)(c.BankOutlined,{}),roles:E.all_admin_roles},{key:"access-groups",page:"access-groups",label:"Access Groups",icon:(0,a.jsx)(g.BlockOutlined,{}),roles:E.all_admin_roles},{key:"budgets",page:"budgets",label:"Budgets",icon:(0,a.jsx)(p.CreditCardOutlined,{}),roles:E.all_admin_roles}]},{groupLabel:"DEVELOPER TOOLS",items:[{key:"api-reference",page:"api-reference",label:"API Reference",icon:(0,a.jsx)(i.ApiOutlined,{})},{key:"model-hub-table",page:"model-hub-table",label:"AI Hub",icon:(0,a.jsx)(o.AppstoreOutlined,{})},{key:"learning-resources",page:"learning-resources",label:"Learning Resources",icon:(0,a.jsx)(x.BookOutlined,{}),external_url:"https://models.litellm.ai/cookbook"},{key:"experimental",page:"experimental",label:"Experimental",icon:(0,a.jsx)(f.ExperimentOutlined,{}),children:[{key:"caching",page:"caching",label:"Caching",icon:(0,a.jsx)(h.DatabaseOutlined,{}),roles:E.all_admin_roles},{key:"prompts",page:"prompts",label:"Prompts",icon:(0,a.jsx)(b.FileTextOutlined,{}),roles:E.all_admin_roles},{key:"transform-request",page:"transform-request",label:"API Playground",icon:(0,a.jsx)(i.ApiOutlined,{}),roles:[...E.all_admin_roles,...E.internalUserRoles]},{key:"tag-management",page:"tag-management",label:"Tag Management",icon:(0,a.jsx)(C.TagsOutlined,{}),roles:E.all_admin_roles},{key:"4",page:"usage",label:"Old Usage",icon:(0,a.jsx)(m.BarChartOutlined,{})}]}]},{groupLabel:"SETTINGS",roles:E.all_admin_roles,items:[{key:"settings",page:"settings",label:(0,a.jsxs)("span",{className:"flex items-center gap-2",children:["Settings ",(0,a.jsx)(U.default,{})]}),icon:(0,a.jsx)(L.SettingOutlined,{}),roles:E.all_admin_roles,children:[{key:"router-settings",page:"router-settings",label:"Router Settings",icon:(0,a.jsx)(L.SettingOutlined,{}),roles:E.all_admin_roles},{key:"logging-and-alerts",page:"logging-and-alerts",label:"Logging & Alerts",icon:(0,a.jsx)(L.SettingOutlined,{}),roles:E.all_admin_roles},{key:"admin-panel",page:"admin-panel",label:(0,a.jsxs)("span",{className:"flex items-center gap-2",children:["Admin Settings ",(0,a.jsx)(U.default,{dot:!0,children:(0,a.jsx)("span",{})})]}),icon:(0,a.jsx)(L.SettingOutlined,{}),roles:E.all_admin_roles},{key:"cost-tracking",page:"cost-tracking",label:"Cost Tracking",icon:(0,a.jsx)(m.BarChartOutlined,{}),roles:E.all_admin_roles},{key:"ui-theme",page:"ui-theme",label:"UI Theme",icon:(0,a.jsx)(u.BgColorsOutlined,{}),roles:E.all_admin_roles}]}]}];e.s(["default",0,({setPage:e,defaultSelectedKey:i,collapsed:n=!1,enabledPagesInternalUsers:o,enableProjectsUI:d,disableAgentsForInternalUsers:c,allowAgentsForTeamAdmins:m,disableVectorStoresForInternalUsers:u,allowVectorStoresForTeamAdmins:g})=>{let x,{userId:p,accessToken:h,userRole:f}=(0,r.default)(),{data:b}=(0,s.useOrganizations)(),{data:v}=(0,l.useTeams)(),j=(0,R.useMemo)(()=>!!p&&!!b&&b.some(e=>e.members?.some(e=>e.user_id===p&&"org_admin"===e.user_role)),[p,b]),N=(0,R.useMemo)(()=>(0,E.isUserTeamAdminForAnyTeam)(v??null,p??""),[v,p]),k=t=>{if(X[t])return void e(t);let a=new URLSearchParams(window.location.search);a.set("page",t),window.history.pushState(null,"",`?${a.toString()}`),e(t)},w=(e,s,l)=>{let r;if(l)return(0,a.jsxs)("a",{href:l,target:"_blank",rel:"noopener noreferrer",onClick:e=>e.stopPropagation(),style:{color:"inherit",textDecoration:"none"},children:[e," ",(0,a.jsx)(y.ExportOutlined,{style:{fontSize:10,marginLeft:4}})]});let i=X[s],n=i?function(e){let a=(t.default.env.NEXT_PUBLIC_BASE_URL??"").replace(/^\/+|\/+$/g,""),s=a?`/${a}/`:"/";if(W.serverRootPath&&"/"!==W.serverRootPath){let e=W.serverRootPath.replace(/\/+$/,""),t=s.replace(/^\/+/,"");s=`${e}/${t}`}return`${s}${e}`}(i):((r=new URLSearchParams(window.location.search)).set("page",s),`?${r.toString()}`);return(0,a.jsx)("a",{href:n,onClick:e=>{e.metaKey||e.ctrlKey||e.shiftKey||1===e.button?e.stopPropagation():e.preventDefault()},style:{color:"inherit",textDecoration:"none"},children:e})},O=e=>{let t=(0,E.isAdminRole)(f);return null!=o&&console.log("[LeftNav] Filtering with enabled pages:",{userRole:f,isAdmin:t,enabledPagesInternalUsers:o}),e.map(e=>({...e,children:e.children?O(e.children):void 0})).filter(e=>{if("organizations"===e.key||"users"===e.key){if(!(!e.roles||e.roles.includes(f)||j))return!1;if(!t&&null!=o){let t=o.includes(e.page);return console.log(`[LeftNav] Page "${e.page}" (${e.key}): ${t?"VISIBLE":"HIDDEN"}`),t}return!0}if("projects"===e.key&&!d||!t&&"agents"===e.key&&c&&!(m&&N)||!t&&"vector-stores"===e.key&&u&&!(g&&N)||e.roles&&!e.roles.includes(f))return!1;if(!t&&null!=o){if(e.children&&e.children.length>0&&e.children.some(e=>o.includes(e.page)))return console.log(`[LeftNav] Parent "${e.page}" (${e.key}): VISIBLE (has visible children)`),!0;let t=o.includes(e.page);return console.log(`[LeftNav] Page "${e.page}" (${e.key}): ${t?"VISIBLE":"HIDDEN"}`),t}return!0})},_=(e=>{for(let t of Z)for(let a of t.items){if(a.page===e)return a.key;if(a.children){let t=a.children.find(t=>t.page===e);if(t)return t.key}}return"api-keys"})(i);return(0,a.jsx)(z.Layout,{children:(0,a.jsxs)(Y,{theme:"light",width:220,collapsed:n,collapsedWidth:80,collapsible:!0,trigger:null,style:{transition:"all 0.3s cubic-bezier(0.4, 0, 0.2, 1)",position:"relative"},children:[(0,a.jsx)(H.ConfigProvider,{theme:{components:{Menu:{iconSize:15,fontSize:13,itemMarginInline:4,itemPaddingInline:8,itemHeight:30,itemBorderRadius:6,subMenuItemBorderRadius:6,groupTitleFontSize:10,groupTitleLineHeight:1.5}}},children:(0,a.jsx)(T.Menu,{mode:"inline",selectedKeys:[_],defaultOpenKeys:[],inlineCollapsed:n,className:"custom-sidebar-menu",style:{borderRight:0,backgroundColor:"transparent",fontSize:"13px",paddingTop:"4px"},items:(x=[],Z.forEach(e=>{if(e.roles&&!e.roles.includes(f))return;let t=O(e.items);0!==t.length&&x.push({type:"group",label:n?null:(0,a.jsx)("span",{style:{fontSize:"10px",fontWeight:600,color:"#6b7280",letterSpacing:"0.05em",padding:"12px 0 4px 12px",display:"block",marginBottom:"2px"},children:e.groupLabel}),children:t.map(e=>({key:e.key,icon:e.icon,label:w(e.label,e.page,e.external_url),children:e.children?.map(e=>({key:e.key,icon:e.icon,label:w(e.label,e.page,e.external_url),onClick:()=>{e.external_url?window.open(e.external_url,"_blank"):k(e.page)}})),onClick:e.children?void 0:()=>{e.external_url?window.open(e.external_url,"_blank"):k(e.page)}}))})}),x)})}),(0,E.isAdminRole)(f)&&!n&&(0,a.jsx)(q,{accessToken:h,width:220})]})})},"menuGroups",()=>Z],111672)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/03c715f5e0f1425c.js b/litellm/proxy/_experimental/out/_next/static/chunks/03c715f5e0f1425c.js deleted file mode 100644 index 963ac87c939..00000000000 --- a/litellm/proxy/_experimental/out/_next/static/chunks/03c715f5e0f1425c.js +++ /dev/null @@ -1 +0,0 @@ -(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,56567,838932,471145,e=>{"use strict";var t=e.i(843476),l=e.i(135214),a=e.i(109799),s=e.i(912598),i=e.i(907308),r=e.i(764205),n=e.i(266027);let o=(0,e.i(243652).createQueryKeys)("guardrails"),d=()=>{let{accessToken:e,userId:t,userRole:a}=(0,l.default)();return(0,n.useQuery)({queryKey:o.list({}),queryFn:async()=>(0,r.getGuardrailsList)(e),enabled:!!(e&&t&&a),select:e=>{let t=e?.guardrails??[],l=new Set,a=new Set;for(let e of t)e.litellm_params?.default_on?l.add(e.guardrail_name):a.add(e.guardrail_name);return{guardrails:t,globalGuardrailNames:l,optionalGuardrailNames:a}}})};e.s(["useGuardrails",0,d],838932);var m=e.i(500330),c=e.i(11751),u=e.i(708347),g=e.i(751904),h=e.i(160818),p=e.i(827252),x=e.i(564897),_=e.i(646563),b=e.i(987432),y=e.i(530212),j=e.i(677667),f=e.i(130643),v=e.i(898667),T=e.i(389083),S=e.i(304967),w=e.i(350967),N=e.i(599724),C=e.i(779241),k=e.i(629569),I=e.i(464571),M=e.i(808613),z=e.i(311451),A=e.i(28651),F=e.i(199133),O=e.i(770914),P=e.i(790848),D=e.i(653496),L=e.i(262218),R=e.i(592968),B=e.i(888259),V=e.i(678784),U=e.i(118366),E=e.i(271645),$=e.i(9314),K=e.i(552130),G=e.i(127952);function W({className:e,value:l,onChange:a}){return(0,t.jsxs)(F.Select,{className:e,value:l,onChange:a,children:[(0,t.jsx)(F.Select.Option,{value:"24h",children:"Daily"}),(0,t.jsx)(F.Select.Option,{value:"7d",children:"Weekly"}),(0,t.jsx)(F.Select.Option,{value:"30d",children:"Monthly"})]})}var q=e.i(844565),H=e.i(355619);let J=function({globalGuardrailNames:e,teamGuardrails:l=[],optedOutGlobalGuardrails:a=[],killSwitchOn:s=!1,variant:i="card",className:r=""}){let n=new Set(a),o=Array.from(e).filter(e=>!n.has(e)),d=l.filter(t=>!e.has(t)),m=s||0!==o.length||0!==d.length?(0,t.jsxs)("div",{className:"flex flex-col gap-4",children:[(0,t.jsxs)("div",{children:[(0,t.jsxs)("span",{className:"block text-sm font-medium text-gray-700 mb-2",children:[(0,t.jsx)(h.GlobalOutlined,{style:{marginInlineEnd:4},"aria-label":"Global guardrail"}),"Global"]}),s?(0,t.jsx)(L.Tag,{color:"gold",children:"Bypassed for this team"}):o.length>0?(0,t.jsx)("div",{className:"flex flex-wrap gap-2",children:o.map(e=>(0,t.jsx)(L.Tag,{color:"blue",children:e},e))}):(0,t.jsx)("span",{className:"block text-sm text-gray-500",children:"None configured"})]}),(0,t.jsxs)("div",{children:[(0,t.jsx)("span",{className:"block text-sm font-medium text-gray-700 mb-2",children:"Team-specific"}),d.length>0?(0,t.jsx)("div",{className:"flex flex-wrap gap-2",children:d.map(e=>(0,t.jsx)(L.Tag,{color:"blue",children:e},e))}):(0,t.jsx)("span",{className:"block text-sm text-gray-500",children:"None configured"})]})]}):(0,t.jsx)("span",{className:"block text-gray-500",children:"No guardrails configured"});return"card"===i?(0,t.jsxs)("div",{className:`bg-white border border-gray-200 rounded-lg p-6 ${r}`,children:[(0,t.jsx)("div",{className:"flex items-center gap-2 mb-6",children:(0,t.jsxs)("div",{children:[(0,t.jsx)("span",{className:"block font-semibold text-gray-900",children:"Guardrails Settings"}),(0,t.jsx)("span",{className:"block text-xs text-gray-500",children:"Global and team-specific guardrails applied to this team"})]})}),m]}):(0,t.jsxs)("div",{className:`${r}`,children:[(0,t.jsx)("span",{className:"block font-medium text-gray-900 mb-3",children:"Guardrails Settings"}),m]})};var Q=e.i(643449),Y=e.i(75921),X=e.i(390605),Z=e.i(162386),ee=e.i(727749),et=e.i(384767),el=e.i(435451),ea=e.i(916940);let es=({onChange:e,value:l,className:a,accessToken:s,placeholder:i="Select search tools (optional)",disabled:n=!1})=>{let[o,d]=(0,E.useState)([]),[m,c]=(0,E.useState)(!1);return(0,E.useEffect)(()=>{(async()=>{if(s){c(!0);try{let e=await (0,r.fetchSearchTools)(s),t=Array.isArray(e?.search_tools)?e.search_tools:Array.isArray(e?.data)?e.data:[];d(t.map(e=>e?.search_tool_name).filter(e=>"string"==typeof e&&e.length>0).map(e=>({label:e,value:e})))}catch(e){console.error("Failed to load search tools:",e)}finally{c(!1)}}})()},[s]),(0,t.jsx)(F.Select,{mode:"multiple",allowClear:!0,showSearch:!0,optionFilterProp:"label",placeholder:i,onChange:e,value:l,loading:m,className:a,options:o,style:{width:"100%"},disabled:n})};e.s(["default",0,es],471145);var ei=e.i(183588),er=e.i(460285),en=e.i(276173),eo=e.i(91979),ed=e.i(269200),em=e.i(942232),ec=e.i(977572),eu=e.i(427612),eg=e.i(64848),eh=e.i(496020),ep=e.i(536916),ex=e.i(21548);let e_={"/key/generate":"Member can generate a virtual key for this team","/key/service-account/generate":"Member can generate a service account key (not belonging to any user) for this team","/key/update":"Member can update a virtual key belonging to this team","/key/delete":"Member can delete a virtual key belonging to this team","/key/info":"Member can get info about a virtual key belonging to this team","/key/regenerate":"Member can regenerate a virtual key belonging to this team","/key/{key_id}/regenerate":"Member can regenerate a virtual key belonging to this team","/key/list":"Member can list virtual keys belonging to this team","/key/block":"Member can block a virtual key belonging to this team","/key/unblock":"Member can unblock a virtual key belonging to this team","/team/daily/activity":"Member can view all team usage data (not just their own)","/spend/logs":"Member can view spend logs for the entire team (not just their own)"},eb=({teamId:e,accessToken:l,canEditTeam:a})=>{let[s,i]=(0,E.useState)([]),[n,o]=(0,E.useState)([]),[d,m]=(0,E.useState)(!0),[c,u]=(0,E.useState)(!1),[g,h]=(0,E.useState)(!1),p=async()=>{try{if(m(!0),!l)return;let t=await (0,r.getTeamPermissionsCall)(l,e),a=t.all_available_permissions||[];i(a);let s=t.team_member_permissions||[];o(s),h(!1)}catch(e){ee.default.fromBackend("Failed to load permissions"),console.error("Error fetching permissions:",e)}finally{m(!1)}};(0,E.useEffect)(()=>{p()},[e,l]);let x=async()=>{try{if(!l)return;u(!0),await (0,r.teamPermissionsUpdateCall)(l,e,n),ee.default.success("Permissions updated successfully"),h(!1)}catch(e){ee.default.fromBackend("Failed to update permissions"),console.error("Error updating permissions:",e)}finally{u(!1)}};if(d)return(0,t.jsx)("div",{className:"p-6 text-center",children:"Loading permissions..."});let _=s.length>0;return(0,t.jsxs)(S.Card,{className:"bg-white shadow-md rounded-md p-6",children:[(0,t.jsxs)("div",{className:"flex flex-col sm:flex-row justify-between items-start sm:items-center border-b pb-4 mb-6",children:[(0,t.jsx)(k.Title,{className:"mb-2 sm:mb-0",children:"Member Permissions"}),a&&g&&(0,t.jsxs)("div",{className:"flex gap-3",children:[(0,t.jsx)(I.Button,{icon:(0,t.jsx)(eo.ReloadOutlined,{}),onClick:()=>{p()},children:"Reset"}),(0,t.jsx)(I.Button,{onClick:x,loading:c,type:"primary",icon:(0,t.jsx)(b.SaveOutlined,{}),children:"Save Changes"})]})]}),(0,t.jsx)(N.Text,{className:"mb-6 text-gray-600",children:"Control what team members can do when they are not team admins."}),_?(0,t.jsx)("div",{className:"overflow-x-auto",children:(0,t.jsxs)(ed.Table,{className:" min-w-full",children:[(0,t.jsx)(eu.TableHead,{children:(0,t.jsxs)(eh.TableRow,{children:[(0,t.jsx)(eg.TableHeaderCell,{children:"Method"}),(0,t.jsx)(eg.TableHeaderCell,{children:"Endpoint"}),(0,t.jsx)(eg.TableHeaderCell,{children:"Description"}),(0,t.jsx)(eg.TableHeaderCell,{className:"sticky right-0 bg-white shadow-[-4px_0_4px_-4px_rgba(0,0,0,0.1)] text-center",children:"Allow Access"})]})}),(0,t.jsx)(em.TableBody,{children:s.map(e=>{let l=(e=>{let t=e.includes("/info")||e.includes("/list")||e.includes("/activity")||"/spend/logs"===e?"GET":"POST",l=e_[e];if(!l){for(let[t,a]of Object.entries(e_))if(e.includes(t)){l=a;break}}return l||(l=`Access ${e}`),{method:t,endpoint:e,description:l,route:e}})(e);return(0,t.jsxs)(eh.TableRow,{className:"hover:bg-gray-50 transition-colors",children:[(0,t.jsx)(ec.TableCell,{children:(0,t.jsx)("span",{className:`px-2 py-1 rounded text-xs font-medium ${"GET"===l.method?"bg-blue-100 text-blue-800":"bg-green-100 text-green-800"}`,children:l.method})}),(0,t.jsx)(ec.TableCell,{children:(0,t.jsx)("span",{className:"font-mono text-sm text-gray-800",children:l.endpoint})}),(0,t.jsx)(ec.TableCell,{className:"text-gray-700",children:l.description}),(0,t.jsx)(ec.TableCell,{className:"sticky right-0 bg-white shadow-[-4px_0_4px_-4px_rgba(0,0,0,0.1)] text-center",children:(0,t.jsx)(ep.Checkbox,{checked:n.includes(e),onChange:t=>{o(t.target.checked?[...n,e]:n.filter(t=>t!==e)),h(!0)},disabled:!a})})]},e)})})]})}):(0,t.jsx)("div",{className:"py-12",children:(0,t.jsx)(ex.Empty,{description:"No permissions available"})})]})};var ey=e.i(822315);function ej(e){if(!e)return null;let t=(0,ey.default)(e);return t.isValid()?t.format("MMM D, YYYY"):null}var ef=e.i(175712),ev=e.i(178654),eT=e.i(621192),eS=e.i(898586);let ew=async(e,t)=>{let l=(0,r.getProxyBaseUrl)(),a=l?`${l}/team/${encodeURIComponent(t)}/members/me`:`/team/${encodeURIComponent(t)}/members/me`,s=await fetch(a,{method:"GET",headers:{[(0,r.getGlobalLitellmHeaderName)()]:`Bearer ${e}`,"Content-Type":"application/json"}});if(404===s.status)return null;if(!s.ok){let e=await s.json().catch(()=>({}));throw Error((0,r.deriveErrorMessage)(e))}return await s.json()},eN=(e,l)=>(0,t.jsxs)(O.Space,{size:4,children:[(0,t.jsx)(eS.Typography.Text,{type:"secondary",children:e}),(0,t.jsx)(R.Tooltip,{title:l,children:(0,t.jsx)(p.InfoCircleOutlined,{style:{color:"#8c8c8c"}})})]}),eC=(e,t=4)=>null==e?"0":(0,m.formatNumberWithCommas)(e,t),ek=e=>null==e?"Unlimited":(0,m.formatNumberWithCommas)(e,0);function eI({teamId:e}){let{data:a,isLoading:s,error:i}=(e=>{let{accessToken:t}=(0,l.default)();return(0,n.useQuery)({queryKey:["team",e,"members","me"],queryFn:()=>ew(t,e),enabled:!!(t&&e)})})(e);if(s)return(0,t.jsx)(ef.Card,{children:(0,t.jsx)(eS.Typography.Text,{type:"secondary",children:"Loading your membership info…"})});if(i)return(0,t.jsx)(ef.Card,{children:(0,t.jsx)(eS.Typography.Text,{type:"danger",children:i instanceof Error?i.message:"Failed to load your membership info for this team."})});if(!a)return(0,t.jsx)(ef.Card,{children:(0,t.jsx)(eS.Typography.Text,{type:"secondary",children:"No membership info available for the current user in this team."})});let r=a.litellm_budget_table??null,o=r?.max_budget??null,d=a.spend??0,m=a.total_spend??0,c=r?.tpm_limit??null,u=r?.rpm_limit??null,g=ej(r?.budget_reset_at),h=r?.allowed_models??null;return(0,t.jsxs)(O.Space,{direction:"vertical",size:"middle",style:{width:"100%"},children:[(0,t.jsx)(ef.Card,{children:(0,t.jsxs)(eT.Row,{gutter:[24,16],children:[(0,t.jsxs)(ev.Col,{xs:24,sm:12,md:8,children:[(0,t.jsx)(eS.Typography.Text,{type:"secondary",children:"User"}),(0,t.jsx)("div",{style:{marginTop:4},children:(0,t.jsx)(eS.Typography.Text,{strong:!0,children:a.user_email||a.user_id})}),(0,t.jsx)(eS.Typography.Text,{type:"secondary",style:{fontSize:12,fontFamily:"monospace"},children:a.user_id})]}),(0,t.jsxs)(ev.Col,{xs:24,sm:12,md:8,children:[(0,t.jsx)(eS.Typography.Text,{type:"secondary",children:"Team Role"}),(0,t.jsx)("div",{style:{marginTop:4},children:(0,t.jsx)(L.Tag,{color:"admin"===a.role?"blue":"default",children:a.role||"user"})})]})]})}),(0,t.jsxs)(eT.Row,{gutter:[16,16],children:[(0,t.jsx)(ev.Col,{xs:24,md:12,children:(0,t.jsxs)(ef.Card,{children:[eN("Current Cycle Spend (USD)","Spend for the current budget cycle. Resets to $0 when the budget window rolls over."),(0,t.jsxs)("div",{style:{marginTop:8},children:[(0,t.jsxs)(eS.Typography.Title,{level:3,style:{margin:0},children:["$",eC(d,4)]}),(0,t.jsxs)(eS.Typography.Text,{type:"secondary",children:["of ",null===o?"Unlimited":`$${eC(o,4)}`]})]}),g&&(0,t.jsx)("div",{style:{marginTop:4},children:(0,t.jsxs)(eS.Typography.Text,{type:"secondary",children:["Resets ",g]})})]})}),(0,t.jsx)(ev.Col,{xs:24,md:12,children:(0,t.jsxs)(ef.Card,{children:[eN("Rate Limits","Your per-member rate limits within this team."),(0,t.jsxs)("div",{style:{marginTop:8},children:[(0,t.jsxs)(eS.Typography.Text,{children:["TPM: ",ek(c)]}),(0,t.jsx)("br",{}),(0,t.jsxs)(eS.Typography.Text,{children:["RPM: ",ek(u)]})]})]})}),(0,t.jsx)(ev.Col,{xs:24,md:12,children:(0,t.jsxs)(ef.Card,{children:[eN("Total Spend (USD)","Cumulative spend across all budget cycles within this team."),(0,t.jsx)("div",{style:{marginTop:8},children:(0,t.jsxs)(eS.Typography.Title,{level:4,style:{margin:0},children:["$",eC(m,4)]})})]})}),(0,t.jsx)(ev.Col,{xs:24,md:12,children:(0,t.jsxs)(ef.Card,{children:[eN("Model Scope","Models you can access within this team."),(0,t.jsx)("div",{style:{marginTop:8},children:h&&h.length>0?(0,t.jsx)(O.Space,{wrap:!0,children:h.map(e=>(0,t.jsx)(L.Tag,{children:e},e))}):(0,t.jsx)(eS.Typography.Text,{children:"All Team Models"})})]})})]})]})}let eM="overview",ez="my-user",eA="virtual-keys",eF="members",eO="member-permissions",eP="settings",eD={[eM]:"Overview",[ez]:"My User",[eA]:"Virtual Keys",[eF]:"Members",[eO]:"Member Permissions",[eP]:"Settings"};var eL=e.i(292639),eR=e.i(294612);function eB({teamData:e,canEditTeam:a,handleMemberDelete:s,setSelectedEditMember:i,setIsEditMemberModalVisible:r,setIsAddMemberModalVisible:n}){let o=e=>{if(null==e)return"0";if("number"==typeof e){let t=Number(e);return t===Math.floor(t)?t.toString():(0,m.formatNumberWithCommas)(t,8).replace(/\.?0+$/,"")}return"0"},{data:d}=(0,eL.useUISettings)(),{userId:c,userRole:g}=(0,l.default)(),h=!!d?.values?.disable_team_admin_delete_team_user,x=(0,u.isUserTeamAdminForSingleTeam)(e.team_info.members_with_roles,c||""),_=(0,u.isProxyAdminRole)(g||""),b=[{title:(0,t.jsxs)(O.Space,{direction:"horizontal",children:["Model Scope",(0,t.jsx)(R.Tooltip,{title:"Models this member can access. Empty means they inherit all team models.",children:(0,t.jsx)(p.InfoCircleOutlined,{})})]}),key:"model_scope",render:(l,a)=>{let s=(t=>{if(!t)return null;let l=e.team_memberships.find(e=>e.user_id===t),a=l?.litellm_budget_table?.allowed_models;return a&&a.length>0?a:null})(a.user_id);if(!s)return(0,t.jsx)(eS.Typography.Text,{type:"secondary",children:"(all team models)"});let i=s.slice(0,2),r=s.length-i.length;return(0,t.jsxs)(O.Space,{wrap:!0,children:[i.map(e=>(0,t.jsx)(eS.Typography.Text,{code:!0,style:{fontSize:"12px"},children:e},e)),r>0&&(0,t.jsx)(R.Tooltip,{title:s.slice(2).join(", "),children:(0,t.jsxs)(eS.Typography.Text,{type:"secondary",children:["+",r," more"]})})]})}},{title:(0,t.jsxs)(O.Space,{direction:"horizontal",children:["Current Cycle Spend (USD)",(0,t.jsx)(R.Tooltip,{title:"Spend for the current budget cycle. Resets to $0 when the member's budget window rolls over. This is the value checked against the member's budget.",children:(0,t.jsx)(p.InfoCircleOutlined,{})})]}),key:"spend",render:(l,a)=>(0,t.jsxs)(eS.Typography.Text,{children:["$",(0,m.formatNumberWithCommas)((t=>{if(!t)return 0;let l=e.team_memberships.find(e=>e.user_id===t);return l?.spend??0})(a.user_id),4)]})},{title:(0,t.jsxs)(O.Space,{direction:"horizontal",children:["Total Spend (USD)",(0,t.jsx)(R.Tooltip,{title:"Cumulative spend by this member within this team, across all budget cycles. Tracking began 2026-04-21; spend from before that date is not included.",children:(0,t.jsx)(p.InfoCircleOutlined,{})})]}),key:"total_spend",render:(l,a)=>(0,t.jsxs)(eS.Typography.Text,{children:["$",(0,m.formatNumberWithCommas)((t=>{if(!t)return 0;let l=e.team_memberships.find(e=>e.user_id===t);return l?.total_spend??0})(a.user_id),4)]})},{title:"Team Member Budget (USD)",key:"budget",render:(l,a)=>{let s=(t=>{if(!t)return null;let l=e.team_memberships.find(e=>e.user_id===t),a=l?.litellm_budget_table?.max_budget;return null==a?null:o(a)})(a.user_id);return(0,t.jsx)(eS.Typography.Text,{children:s?`$${(0,m.formatNumberWithCommas)(Number(s),4)}`:"No Limit"})}},{title:"Budget Reset",key:"budget_reset",render:(l,a)=>{let s=(t=>{if(!t)return null;let l=e.team_memberships.find(e=>e.user_id===t);return ej(l?.litellm_budget_table?.budget_reset_at)})(a.user_id);return s?(0,t.jsx)(eS.Typography.Text,{children:s}):(0,t.jsx)(eS.Typography.Text,{type:"secondary",children:"—"})}},{title:(0,t.jsxs)(O.Space,{direction:"horizontal",children:["Team Member Rate Limits",(0,t.jsx)(R.Tooltip,{title:"Rate limits for this member's usage within this team.",children:(0,t.jsx)(p.InfoCircleOutlined,{})})]}),key:"rate_limits",render:(l,a)=>(0,t.jsx)(eS.Typography.Text,{children:(t=>{if(!t)return"No Limits";let l=e.team_memberships.find(e=>e.user_id===t),a=l?.litellm_budget_table?.rpm_limit,s=l?.litellm_budget_table?.tpm_limit,i=[a?`${o(a)} RPM`:null,s?`${o(s)} TPM`:null].filter(Boolean);return i.length>0?i.join(" / "):"No Limits"})(a.user_id)})}];return(0,t.jsx)(eR.default,{members:e.team_info.members_with_roles,canEdit:a,onEdit:t=>{let l=e.team_memberships.find(e=>e.user_id===t.user_id);i({...t,max_budget_in_team:l?.litellm_budget_table?.max_budget||null,tpm_limit:l?.litellm_budget_table?.tpm_limit||null,rpm_limit:l?.litellm_budget_table?.rpm_limit||null,allowed_models:l?.litellm_budget_table?.allowed_models||[]}),r(!0)},onDelete:s,onAddMember:()=>n(!0),roleColumnTitle:"Team Role",roleTooltip:"This role applies only to this team and is independent from the user's proxy-level role.",extraColumns:b,showDeleteForMember:()=>_||a&&!x||x&&!h})}var eV=e.i(207082),eU=e.i(871943),eE=e.i(502547),e$=e.i(360820),eK=e.i(94629),eG=e.i(152990),eW=e.i(682830),eq=e.i(994388),eH=e.i(752978),eJ=e.i(282786),eQ=e.i(981339),eY=e.i(969550),eX=e.i(20147),eZ=e.i(633627);function e0({teamId:e,teamAlias:a,organization:s}){let{accessToken:i}=(0,l.default)(),[r,o]=(0,E.useState)(null),[d,c]=(0,E.useState)([{id:"created_at",desc:!0}]),[u,g]=(0,E.useState)({pageIndex:0,pageSize:50}),[h,x]=(0,E.useState)({"Organization ID":"","Key Alias":"","User ID":"","Sort By":"created_at","Sort Order":"desc"}),_=d.length>0?d[0].id:"created_at",b=d.length>0?d[0].desc?"desc":"asc":"desc",y=u.pageIndex,j=u.pageSize,{data:f,isPending:v,isFetching:S,refetch:w}=(0,eV.useKeys)(y+1,j,{teamID:e,organizationID:h["Organization ID"]?.trim()||void 0,selectedKeyAlias:h["Key Alias"]?.trim()||void 0,userID:h["User ID"]?.trim()||void 0,sortBy:_||void 0,sortOrder:b||void 0,expand:"user"}),C=(0,E.useMemo)(()=>{let e=f?.keys||[],t=s?.organization_id;return t?e.map(e=>({...e,organization_id:(e.organization_id??e.org_id)||t})):e},[f?.keys,s?.organization_id]),k=f?.total_pages??0,[I,M]=(0,E.useState)({}),z=(0,E.useMemo)(()=>({team_id:e,team_alias:a||e,models:[],max_budget:null,budget_duration:null,tpm_limit:null,rpm_limit:null,organization_id:s?.organization_id||"",created_at:"",keys:[],members_with_roles:[],spend:0}),[e,a,s]),A=(0,n.useQuery)({queryKey:["teamFilterOptions",e,i],queryFn:async()=>(0,eZ.fetchTeamFilterOptions)(i,e),enabled:!!i&&!!e,staleTime:3e4}).data||{keyAliases:[],organizationIds:[],userIds:[]},F=(0,E.useCallback)(()=>{w?.()},[w]);(0,E.useEffect)(()=>(window.addEventListener("storage",F),()=>window.removeEventListener("storage",F)),[F]);let O=(0,E.useCallback)((e,t=!1)=>{x(t=>({...t,"Organization ID":e["Organization ID"]??t["Organization ID"],"Key Alias":e["Key Alias"]??t["Key Alias"],"User ID":e["User ID"]??t["User ID"],"Sort By":e["Sort By"]??t["Sort By"]??"created_at","Sort Order":e["Sort Order"]??t["Sort Order"]??"desc"})),t||g(e=>({...e,pageIndex:0}))},[]),P=(0,E.useCallback)(()=>{x({"Organization ID":"","Key Alias":"","User ID":"","Sort By":"created_at","Sort Order":"desc"}),g(e=>({...e,pageIndex:0}))},[]),D=(0,E.useMemo)(()=>[{name:"Organization ID",label:"Organization ID",isSearchable:!0,searchFn:async e=>{let{organizationIds:t}=A;if(!t.length)return[];let l=e.toLowerCase();return(l?t.filter(e=>e.toLowerCase().includes(l)):t).map(e=>({label:e,value:e}))}},{name:"Key Alias",label:"Key Alias",isSearchable:!0,searchFn:async e=>{let{keyAliases:t}=A,l=e.toLowerCase();return(l?t.filter(e=>e.toLowerCase().includes(l)):t).map(e=>({label:e,value:e}))}},{name:"User ID",label:"User ID",isSearchable:!0,searchFn:async e=>{let{userIds:t}=A,l=e.toLowerCase();return(l?t.filter(e=>e.id.toLowerCase().includes(l)||e.email.toLowerCase().includes(l)):t).map(e=>({label:e.email?`${e.id} (${e.email})`:e.id,value:e.id}))}}],[A]),L=(0,E.useMemo)(()=>[{id:"token",accessorKey:"token",header:"Key ID",size:100,enableSorting:!0,cell:e=>{let l=e.getValue(),a=e.cell.column.getSize();return(0,t.jsx)(R.Tooltip,{title:l,children:(0,t.jsx)(eq.Button,{size:"xs",variant:"light",className:"font-mono text-blue-500 bg-blue-50 hover:bg-blue-100 text-xs font-normal px-2 py-0.5 text-left overflow-hidden truncate block",style:{maxWidth:a,overflow:"hidden"},onClick:()=>o(e.row.original),children:l??"-"})})}},{id:"key_alias",accessorKey:"key_alias",header:"Key Alias",size:150,enableSorting:!0,cell:e=>{let l=e.getValue(),a=e.cell.column.getSize();return(0,t.jsx)(R.Tooltip,{title:l,children:(0,t.jsx)("span",{className:"font-mono text-xs truncate block",style:{maxWidth:a,overflow:"hidden"},children:l??"-"})})}},{id:"key_name",accessorKey:"key_name",header:"Secret Key",size:120,enableSorting:!1,cell:e=>(0,t.jsx)("span",{className:"font-mono text-xs",children:e.getValue()})},{id:"organization_id",accessorKey:"organization_id",header:"Organization ID",size:140,enableSorting:!1,cell:e=>e.getValue()?e.renderValue():"-"},{id:"user_email",accessorKey:"user",header:"User Email",size:160,enableSorting:!1,cell:e=>{let l=e.getValue(),a=l?.user_email,s=e.cell.column.getSize();return(0,t.jsx)(R.Tooltip,{title:a,children:(0,t.jsx)("span",{className:"font-mono text-xs truncate block",style:{maxWidth:s,overflow:"hidden"},children:a??"-"})})}},{id:"user_id",accessorKey:"user_id",header:"User ID",size:70,enableSorting:!1,cell:e=>{let l=e.getValue(),a="default_user_id"===l?"Default Proxy Admin":l,s=e.cell.column.getSize();return(0,t.jsx)(R.Tooltip,{title:a,children:(0,t.jsx)("span",{className:"font-mono text-xs truncate block",style:{maxWidth:s,overflow:"hidden"},children:a??"-"})})}},{id:"created_at",accessorKey:"created_at",header:"Created At",size:120,enableSorting:!0,cell:e=>{let t=e.getValue();return t?new Date(t).toLocaleDateString():"-"}},{id:"created_by",accessorKey:"created_by",header:"Created By",size:70,enableSorting:!1,cell:e=>{let l=e.getValue(),a="default_user_id"===l?"Default Proxy Admin":l,s=e.cell.column.getSize();return(0,t.jsx)(R.Tooltip,{title:a,children:(0,t.jsx)("span",{className:"font-mono text-xs truncate block",style:{maxWidth:s,overflow:"hidden"},children:a??"-"})})}},{id:"updated_at",accessorKey:"updated_at",header:"Updated At",size:120,enableSorting:!0,cell:e=>{let t=e.getValue();return t?new Date(t).toLocaleDateString():"Never"}},{id:"last_active",accessorKey:"last_active",header:()=>(0,t.jsxs)("span",{className:"flex items-center gap-1",children:["Last Active",(0,t.jsx)(eJ.Popover,{content:"This is a new field and is not backfilled. Only new key usage will update this value.",trigger:"hover",children:(0,t.jsx)(p.InfoCircleOutlined,{className:"text-gray-400 text-xs cursor-help"})})]}),size:130,enableSorting:!1,cell:e=>{let l=e.getValue();if(!l)return"Unknown";let a=new Date(l);return(0,t.jsx)(R.Tooltip,{title:a.toLocaleString(void 0,{dateStyle:"medium",timeStyle:"long"}),children:(0,t.jsx)("span",{children:a.toLocaleDateString()})})}},{id:"expires",accessorKey:"expires",header:"Expires",size:120,enableSorting:!1,cell:e=>{let t=e.getValue();return t?new Date(t).toLocaleDateString():"Never"}},{id:"spend",accessorKey:"spend",header:"Spend (USD)",size:100,enableSorting:!0,cell:e=>(0,m.formatNumberWithCommas)(e.getValue(),4)},{id:"max_budget",accessorKey:"max_budget",header:"Budget (USD)",size:110,enableSorting:!0,cell:e=>{let t=e.getValue();return null===t?"Unlimited":`$${(0,m.formatNumberWithCommas)(t)}`}},{id:"budget_reset_at",accessorKey:"budget_reset_at",header:"Budget Reset",size:130,enableSorting:!1,cell:e=>{let t=e.getValue();return t?new Date(t).toLocaleString():"Never"}},{id:"models",accessorKey:"models",header:"Models",size:200,enableSorting:!1,cell:e=>{let l=e.getValue();return(0,t.jsx)("div",{className:"flex flex-col py-2",children:Array.isArray(l)?(0,t.jsx)("div",{className:"flex flex-col",children:0===l.length?(0,t.jsx)(T.Badge,{size:"xs",className:"mb-1",color:"red",children:(0,t.jsx)(N.Text,{children:"All Proxy Models"})}):(0,t.jsx)(t.Fragment,{children:(0,t.jsxs)("div",{className:"flex items-start",children:[l.length>3&&(0,t.jsx)("div",{children:(0,t.jsx)(eH.Icon,{icon:I[e.row.id]?eU.ChevronDownIcon:eE.ChevronRightIcon,className:"cursor-pointer",size:"xs",onClick:()=>M(t=>({...t,[e.row.id]:!t[e.row.id]}))})}),(0,t.jsxs)("div",{className:"flex flex-wrap gap-1",children:[l.slice(0,3).map((e,l)=>"all-proxy-models"===e?(0,t.jsx)(T.Badge,{size:"xs",color:"red",children:(0,t.jsx)(N.Text,{children:"All Proxy Models"})},l):(0,t.jsx)(T.Badge,{size:"xs",color:"blue",children:(0,t.jsx)(N.Text,{children:e.length>30?`${(0,H.getModelDisplayName)(e).slice(0,30)}...`:(0,H.getModelDisplayName)(e)})},l)),l.length>3&&!I[e.row.id]&&(0,t.jsx)(T.Badge,{size:"xs",color:"gray",className:"cursor-pointer",children:(0,t.jsxs)(N.Text,{children:["+",l.length-3," ",l.length-3==1?"more model":"more models"]})}),I[e.row.id]&&(0,t.jsx)("div",{className:"flex flex-wrap gap-1",children:l.slice(3).map((e,l)=>"all-proxy-models"===e?(0,t.jsx)(T.Badge,{size:"xs",color:"red",children:(0,t.jsx)(N.Text,{children:"All Proxy Models"})},l+3):(0,t.jsx)(T.Badge,{size:"xs",color:"blue",children:(0,t.jsx)(N.Text,{children:e.length>30?`${(0,H.getModelDisplayName)(e).slice(0,30)}...`:(0,H.getModelDisplayName)(e)})},l+3))})]})]})})}):null})}},{id:"rate_limits",header:"Rate Limits",size:140,enableSorting:!1,cell:({row:e})=>{let l=e.original;return(0,t.jsxs)("div",{children:[(0,t.jsxs)("div",{children:["TPM: ",null!==l.tpm_limit?l.tpm_limit:"Unlimited"]}),(0,t.jsxs)("div",{children:["RPM: ",null!==l.rpm_limit?l.rpm_limit:"Unlimited"]})]})}}],[I]),B=(0,E.useCallback)(e=>{let t="function"==typeof e?e(d):e;if(c(t),t?.length>0){let e=t[0];O({"Sort By":e.id,"Sort Order":e.desc?"desc":"asc"},!0)}},[d,O]),V=(0,eG.useReactTable)({data:C,columns:L,columnResizeMode:"onChange",columnResizeDirection:"ltr",state:{sorting:d,pagination:u},onSortingChange:B,onPaginationChange:g,getCoreRowModel:(0,eW.getCoreRowModel)(),enableSorting:!0,manualSorting:!0,manualPagination:!0,pageCount:k});return(0,t.jsx)("div",{className:"w-full h-full overflow-hidden",children:r?(0,t.jsx)(eX.default,{keyId:r.token,onClose:()=>o(null),keyData:r,teams:[z],onDelete:w}):(0,t.jsxs)("div",{className:"border-b py-4 flex-1 overflow-hidden",children:[(0,t.jsx)("div",{className:"w-full mb-6",children:(0,t.jsx)(eY.default,{options:D,onApplyFilters:O,initialValues:h,onResetFilters:P})}),(0,t.jsx)("div",{className:"flex items-center justify-end w-full mb-4",children:(0,t.jsxs)("div",{className:"inline-flex items-center gap-2",children:[v||S?(0,t.jsx)(eQ.Skeleton.Node,{active:!0,style:{width:74,height:20}}):(0,t.jsxs)("span",{className:"text-sm text-gray-700",children:["Page ",y+1," of ",V.getPageCount()]}),v||S?(0,t.jsx)(eQ.Skeleton.Button,{active:!0,size:"small",style:{width:84,height:30}}):(0,t.jsx)("button",{onClick:()=>V.previousPage(),disabled:v||S||!V.getCanPreviousPage(),className:"px-3 py-1 text-sm border rounded-md hover:bg-gray-50 disabled:opacity-50 disabled:cursor-not-allowed",children:"Previous"}),v||S?(0,t.jsx)(eQ.Skeleton.Button,{active:!0,size:"small",style:{width:58,height:30}}):(0,t.jsx)("button",{onClick:()=>V.nextPage(),disabled:v||S||!V.getCanNextPage(),className:"px-3 py-1 text-sm border rounded-md hover:bg-gray-50 disabled:opacity-50 disabled:cursor-not-allowed",children:"Next"})]})}),(0,t.jsx)("div",{className:"h-[75vh] overflow-auto",children:(0,t.jsx)("div",{className:"rounded-lg custom-border relative",children:(0,t.jsx)("div",{className:"overflow-x-auto",children:(0,t.jsxs)(ed.Table,{className:"[&_td]:py-0.5 [&_th]:py-1",style:{width:V.getCenterTotalSize()},children:[(0,t.jsx)(eu.TableHead,{children:V.getHeaderGroups().map(e=>(0,t.jsx)(eh.TableRow,{children:e.headers.map(e=>(0,t.jsx)(eg.TableHeaderCell,{"data-header-id":e.id,className:`py-1 h-8 relative hover:bg-gray-50 ${"actions"===e.id?"sticky right-0 bg-white shadow-[-4px_0_8px_-6px_rgba(0,0,0,0.1)]":""}`,style:{width:e.getSize(),position:"relative",cursor:e.column.getCanSort()?"pointer":"default"},onMouseEnter:()=>{let t=document.querySelector(`[data-header-id="${e.id}"] .resizer`);t&&(t.style.opacity="0.5")},onMouseLeave:()=>{let t=document.querySelector(`[data-header-id="${e.id}"] .resizer`);t&&!e.column.getIsResizing()&&(t.style.opacity="0")},onClick:e.column.getCanSort()?e.column.getToggleSortingHandler():void 0,children:(0,t.jsxs)("div",{className:"flex items-center justify-between gap-2",children:[(0,t.jsx)("div",{className:"flex items-center",children:e.isPlaceholder?null:(0,eG.flexRender)(e.column.columnDef.header,e.getContext())}),"actions"!==e.id&&e.column.getCanSort()&&(0,t.jsx)("div",{className:"w-4",children:e.column.getIsSorted()?({asc:(0,t.jsx)(e$.ChevronUpIcon,{className:"h-4 w-4 text-blue-500"}),desc:(0,t.jsx)(eU.ChevronDownIcon,{className:"h-4 w-4 text-blue-500"})})[e.column.getIsSorted()]:(0,t.jsx)(eK.SwitchVerticalIcon,{className:"h-4 w-4 text-gray-400"})}),(0,t.jsx)("div",{onDoubleClick:()=>e.column.resetSize(),onMouseDown:e.getResizeHandler(),onTouchStart:e.getResizeHandler(),className:`resizer ${V.options.columnResizeDirection} ${e.column.getIsResizing()?"isResizing":""}`,style:{position:"absolute",right:0,top:0,height:"100%",width:"5px",background:e.column.getIsResizing()?"#3b82f6":"transparent",cursor:"col-resize",userSelect:"none",touchAction:"none",opacity:+!!e.column.getIsResizing()}})]})},e.id))},e.id))}),(0,t.jsx)(em.TableBody,{children:v||S?(0,t.jsx)(eh.TableRow,{children:(0,t.jsx)(ec.TableCell,{colSpan:L.length,className:"h-8 text-center",children:(0,t.jsx)("div",{className:"text-center text-gray-500",children:(0,t.jsx)("p",{children:"Loading keys..."})})})}):C.length>0?V.getRowModel().rows.map(e=>(0,t.jsx)(eh.TableRow,{className:"h-8",children:e.getVisibleCells().map(e=>(0,t.jsx)(ec.TableCell,{style:{width:e.column.getSize(),maxWidth:"8-x",whiteSpace:"pre-wrap",overflow:"hidden"},className:`py-0.5 max-h-8 overflow-hidden text-ellipsis whitespace-nowrap ${"models"===e.column.id&&Array.isArray(e.getValue())&&e.getValue().length>3?"px-0":""}`,children:(0,eG.flexRender)(e.column.columnDef.cell,e.getContext())},e.id))},e.id)):(0,t.jsx)(eh.TableRow,{children:(0,t.jsx)(ec.TableCell,{colSpan:L.length,className:"h-8 text-center",children:(0,t.jsx)("div",{className:"text-center text-gray-500",children:(0,t.jsx)("p",{children:"No keys found"})})})})})]})})})})]})})}e.s(["default",0,({teamId:e,onClose:n,accessToken:o,is_team_admin:eo,is_proxy_admin:ed,is_org_admin:em=!1,userModels:ec,editTeam:eu,premiumUser:eg=!1,onUpdate:eh})=>{let ep,ex,e_,ey,ej,ef,[ev,eT]=(0,E.useState)(null),[eS,ew]=(0,E.useState)(!0),[eN,eC]=(0,E.useState)(!1),[ek]=M.Form.useForm(),[eL,eR]=(0,E.useState)(!1),[eV,eU]=(0,E.useState)(null),[eE,e$]=(0,E.useState)(!1),[eK,eG]=(0,E.useState)([]),[eW,eq]=(0,E.useState)(!1),[eH,eJ]=(0,E.useState)({}),{data:eQ,isLoading:eY}=d(),eX=eQ?.globalGuardrailNames??new Set,[eZ,e1]=(0,E.useState)([]),[e4,e2]=(0,E.useState)({}),[e5,e3]=(0,E.useState)(!1),[e6,e8]=(0,E.useState)(null),[e7,e9]=(0,E.useState)(!1),[te,tt]=(0,E.useState)(!1),[tl,ta]=(0,E.useState)(!1),ts=E.default.useRef(null),[ti,tr]=(0,E.useState)(null),{userRole:tn,userId:to}=(0,l.default)(),{data:td=[]}=(0,a.useOrganizations)(),tm=(0,s.useQueryClient)(),tc=(0,E.useMemo)(()=>{let e=ev?.team_info?.organization_id;if(!e||!to)return!1;let t=td.find(t=>t.organization_id===e);return t?.members?.some(e=>e.user_id===to&&"org_admin"===e.user_role)??!1},[ev,td,to]),tu=M.Form.useWatch("models",ek),tg=M.Form.useWatch("disable_global_guardrails",ek),th=(0,E.useMemo)(()=>{let e=tu??ev?.team_info?.models??[];return e.includes("all-proxy-models")||e.includes("all-team-models")?ec:(0,H.unfurlWildcardModelsInList)(e,ec)},[tu,ev,ec]),tp=eo||ed||em||tc,tx=(0,E.useMemo)(()=>{let e;return e=[eM,ez,eA],tp?[...e,eF,eO,eP]:e},[tp]),t_=(0,E.useMemo)(()=>eu&&tp?eP:eM,[eu,tp]),tb=async()=>{try{if(ew(!0),!o)return;let t=await (0,r.teamInfoCall)(o,e);eT(t)}catch(e){ee.default.fromBackend("Failed to load team information"),console.error("Error fetching team info:",e)}finally{ew(!1)}};(0,E.useEffect)(()=>{tb()},[e,o]),(0,E.useEffect)(()=>{(async()=>{if(!o||!ev?.team_info?.organization_id)return tr(null);try{let e=await (0,r.organizationInfoCall)(o,ev.team_info.organization_id);tr(e)}catch(e){console.error("Error fetching organization info:",e),tr(null)}})()},[o,ev?.team_info?.organization_id]),(0,E.useMemo)(()=>{let e;return e=[],e=ti?ti.models.includes("all-proxy-models")?ec:ti.models.length>0?ti.models:ec:ec,(0,H.unfurlWildcardModelsInList)(e,ec)},[ti,ec]),(0,E.useEffect)(()=>{(async()=>{try{if(!o)return;let e=(await (0,r.getPoliciesList)(o)).policies.map(e=>e.policy_name);e1(e)}catch(e){console.error("Failed to fetch policies:",e)}})()},[o]),(0,E.useEffect)(()=>{(async()=>{if(!o||!ev?.team_info?.policies||0===ev.team_info.policies.length)return;e3(!0);let e={};try{await Promise.all(ev.team_info.policies.map(async t=>{try{let l=await (0,r.getPolicyInfoWithGuardrails)(o,t);e[t]=l.resolved_guardrails||[]}catch(l){console.error(`Failed to fetch guardrails for policy ${t}:`,l),e[t]=[]}})),e2(e)}catch(e){console.error("Failed to fetch policy guardrails:",e)}finally{e3(!1)}})()},[o,ev?.team_info?.policies]);let ty=async t=>{try{if(null==o)return;let l={user_email:t.user_email,user_id:t.user_id,role:t.role};await (0,r.teamMemberAddCall)(o,e,l),ee.default.success("Team member added successfully"),eC(!1),ek.resetFields();let a=await (0,r.teamInfoCall)(o,e);eT(a),eh(a)}catch(t){let e="Failed to add team member";t?.raw?.detail?.error?.includes("Assigning team admins is a premium feature")?e="Assigning admins is an enterprise-only feature. Please upgrade your LiteLLM plan to enable this.":t?.message&&(e=t.message),ee.default.fromBackend(e),console.error("Error adding team member:",t)}},tj=async t=>{try{if(null==o)return;let l={user_email:t.user_email,user_id:t.user_id,role:t.role,max_budget_in_team:t.max_budget_in_team,tpm_limit:t.tpm_limit,rpm_limit:t.rpm_limit,allowed_models:t.allowed_models};B.default.destroy(),await (0,r.teamMemberUpdateCall)(o,e,l),ee.default.success("Team member updated successfully"),eR(!1);let a=await (0,r.teamInfoCall)(o,e);eT(a),eh(a)}catch(t){let e="Failed to update team member";t?.raw?.detail?.includes("Assigning team admins is a premium feature")?e="Assigning admins is an enterprise-only feature. Please upgrade your LiteLLM plan to enable this.":t?.message&&(e=t.message),eR(!1),B.default.destroy(),ee.default.fromBackend(e),console.error("Error updating team member:",t)}},tf=async()=>{if(e6&&o){tt(!0);try{await (0,r.teamMemberDeleteCall)(o,e,e6),ee.default.success("Team member removed successfully");let t=await (0,r.teamInfoCall)(o,e);eT(t),eh(t)}catch(e){ee.default.fromBackend("Failed to remove team member"),console.error("Error removing team member:",e)}finally{tt(!1),e9(!1),e8(null)}}},tv=async t=>{try{let l;if(!o)return;ta(!0);let s={};try{let{soft_budget_alerting_emails:e,...l}=t.metadata?JSON.parse(t.metadata):{};s=l}catch(e){ee.default.fromBackend("Invalid JSON in metadata field");return}if("string"==typeof t.secret_manager_settings&&t.secret_manager_settings.trim().length>0)try{l=JSON.parse(t.secret_manager_settings)}catch(e){ee.default.fromBackend("Invalid JSON in secret manager settings");return}let i=e=>null==e||"string"==typeof e&&""===e.trim()||"number"==typeof e&&Number.isNaN(e)?null:e,n={},d={};for(let e of t.modelLimits??[])e?.model&&(null!=e.tpm&&(n[e.model]=e.tpm),null!=e.rpm&&(d[e.model]=e.rpm));let m=!0===t.disable_global_guardrails,u=m?Array.from(eX):Array.from(eX).filter(e=>!(t.guardrails||[]).includes(e)),g={team_id:e,team_alias:t.team_alias,models:t.models,tpm_limit:i(t.tpm_limit),rpm_limit:i(t.rpm_limit),model_tpm_limit:n,model_rpm_limit:d,max_budget:t.max_budget,soft_budget:i(t.soft_budget),budget_duration:t.budget_duration,metadata:{...s,guardrails:(t.guardrails||[]).filter(e=>!eX.has(e)),opted_out_global_guardrails:u,...t.logging_settings?.length>0?{logging:t.logging_settings}:{},disable_global_guardrails:m,soft_budget_alerting_emails:"string"==typeof t.soft_budget_alerting_emails?t.soft_budget_alerting_emails.split(",").map(e=>e.trim()).filter(e=>e.length>0):t.soft_budget_alerting_emails||[],...void 0!==l?{secret_manager_settings:l}:{}},...t.policies?.length>0?{policies:t.policies}:{},...t.organization_id!==tT.organization_id?{organization_id:t.organization_id??null}:{}};g.max_budget=(0,c.mapEmptyStringToNull)(g.max_budget),g.team_member_budget_duration=t.team_member_budget_duration,void 0!==t.team_member_budget&&(g.team_member_budget=Number(t.team_member_budget)),void 0!==t.team_member_key_duration&&(g.team_member_key_duration=t.team_member_key_duration),(void 0!==t.team_member_tpm_limit||void 0!==t.team_member_rpm_limit)&&(g.team_member_tpm_limit=i(t.team_member_tpm_limit),g.team_member_rpm_limit=i(t.team_member_rpm_limit));let{servers:h,accessGroups:p,toolsets:x}=t.mcp_servers_and_groups||{servers:[],accessGroups:[],toolsets:[]},_=new Set(h||[]),b=Object.fromEntries(Object.entries(t.mcp_tool_permissions||{}).filter(([e])=>_.has(e)));g.object_permission={},h&&(g.object_permission.mcp_servers=h),p&&(g.object_permission.mcp_access_groups=p),b&&(g.object_permission.mcp_tool_permissions=b),x&&(g.object_permission.mcp_toolsets=x),delete t.mcp_servers_and_groups,delete t.mcp_tool_permissions;let{agents:y,accessGroups:j}=t.agents_and_groups||{agents:[],accessGroups:[]};y&&y.length>0&&(g.object_permission.agents=y),j&&j.length>0&&(g.object_permission.agent_access_groups=j),delete t.agents_and_groups,t.vector_stores&&t.vector_stores.length>0&&(g.object_permission.vector_stores=t.vector_stores),Array.isArray(t.object_permission_search_tools)&&(g.object_permission.search_tools=t.object_permission_search_tools),void 0!==t.access_group_ids&&(g.access_group_ids=t.access_group_ids),void 0!==t.default_team_member_models&&(g.default_team_member_models=t.default_team_member_models);let f=ts.current?.getValue();if(f?.router_settings){let e=e=>null!=e&&""!==e&&!1!==e&&!(Array.isArray(e)&&0===e.length),t=Object.values(f.router_settings).some(e),l=tT.router_settings&&Object.values(tT.router_settings).some(e);(t||l)&&(g.router_settings=f.router_settings)}await (0,r.teamUpdateCall)(o,g),tm.invalidateQueries({queryKey:a.organizationKeys.all}),ee.default.success("Team settings updated successfully"),e$(!1),tb()}catch(e){console.error("Error updating team:",e)}finally{ta(!1)}};if(eS)return(0,t.jsx)("div",{className:"p-4",children:"Loading..."});if(!ev?.team_info)return(0,t.jsx)("div",{className:"p-4",children:"Team not found"});let{team_info:tT}=ev,tS=tT.metadata?.disable_global_guardrails===!0,tw=new Set(Array.isArray(tT.metadata?.opted_out_global_guardrails)?tT.metadata.opted_out_global_guardrails:[]),tN=(Array.isArray(tT.metadata?.guardrails)?tT.metadata.guardrails:[]).filter(e=>!eX.has(e)),tC=tS?tN:[...Array.from(eX).filter(e=>!tw.has(e)),...tN],tk=e=>{e.preventDefault(),e.stopPropagation()},tI=async(e,t)=>{await (0,m.copyToClipboard)(e)&&(eJ(e=>({...e,[t]:!0})),setTimeout(()=>{eJ(e=>({...e,[t]:!1}))},2e3))};return(0,t.jsxs)("div",{className:"p-4",children:[(0,t.jsx)("div",{className:"flex justify-between items-center mb-6",children:(0,t.jsxs)("div",{children:[(0,t.jsx)(I.Button,{type:"text",icon:(0,t.jsx)(y.ArrowLeftIcon,{className:"h-4 w-4"}),onClick:n,className:"mb-4",children:"Back to Teams"}),(0,t.jsx)(k.Title,{children:tT.team_alias}),(0,t.jsxs)("div",{className:"flex items-center",children:[(0,t.jsx)(N.Text,{className:"text-gray-500 font-mono",children:tT.team_id}),(0,t.jsx)(I.Button,{type:"text",size:"small",icon:eH["team-id"]?(0,t.jsx)(V.CheckIcon,{size:12}):(0,t.jsx)(U.CopyIcon,{size:12}),onClick:()=>tI(tT.team_id,"team-id"),className:`left-2 z-10 transition-all duration-200 ${eH["team-id"]?"text-green-600 bg-green-50 border-green-200":"text-gray-500 hover:text-gray-700 hover:bg-gray-100"}`})]})]})}),(0,t.jsx)(D.Tabs,{defaultActiveKey:t_,className:"mb-4",items:[{key:eM,label:eD[eM],children:(0,t.jsxs)(w.Grid,{numItems:1,numItemsSm:2,numItemsLg:3,className:"gap-6",children:[(0,t.jsxs)(S.Card,{children:[(0,t.jsx)(N.Text,{children:"Budget Status"}),(0,t.jsxs)("div",{className:"mt-2",children:[(0,t.jsxs)(k.Title,{children:["$",(0,m.formatNumberWithCommas)(tT.spend,4)]}),(0,t.jsxs)(N.Text,{children:["of ",null===tT.max_budget?"Unlimited":`$${(0,m.formatNumberWithCommas)(tT.max_budget,4)}`]}),tT.budget_duration&&(0,t.jsxs)(N.Text,{className:"text-gray-500",children:["Reset: ",tT.budget_duration]}),(0,t.jsx)("br",{}),tT.team_member_budget_table&&(0,t.jsxs)(N.Text,{className:"text-gray-500",children:["Team Member Budget: $",(0,m.formatNumberWithCommas)(tT.team_member_budget_table.max_budget,4)]})]})]}),(0,t.jsxs)(S.Card,{children:[(0,t.jsx)(N.Text,{children:"Rate Limits"}),(0,t.jsxs)("div",{className:"mt-2",children:[(0,t.jsxs)(N.Text,{children:["TPM: ",tT.tpm_limit||"Unlimited"]}),(0,t.jsxs)(N.Text,{children:["RPM: ",tT.rpm_limit||"Unlimited"]}),tT.max_parallel_requests&&(0,t.jsxs)(N.Text,{children:["Max Parallel Requests: ",tT.max_parallel_requests]}),(ep=tT.metadata?.model_tpm_limit??{},ex=tT.metadata?.model_rpm_limit??{},0===(e_=Array.from(new Set([...Object.keys(ep),...Object.keys(ex)]))).length?null:(0,t.jsxs)("div",{className:"mt-3",children:[(0,t.jsx)(N.Text,{className:"text-gray-500",children:"Per-model limits:"}),e_.map(e=>(0,t.jsxs)(N.Text,{className:"text-xs",children:[e,": TPM ",ep[e]??"—",", RPM ",ex[e]??"—"]},e))]}))]})]}),(0,t.jsxs)(S.Card,{children:[(0,t.jsx)(N.Text,{children:"Models"}),(0,t.jsx)("div",{className:"mt-2 flex flex-wrap gap-2",children:0===tT.models.length||tT.models.includes("all-proxy-models")?(0,t.jsx)(T.Badge,{color:"red",children:"All proxy models"}):(0,t.jsxs)(t.Fragment,{children:[tT.models.map((e,l)=>(0,t.jsx)(T.Badge,{color:"blue",children:e},`direct-${l}`)),(tT.access_group_models||[]).map((e,l)=>(0,t.jsx)(T.Badge,{color:"green",title:"From access group",children:e},`ag-${l}`))]})})]}),(0,t.jsxs)(S.Card,{children:[(0,t.jsx)(N.Text,{className:"font-semibold text-gray-900",children:"Virtual Keys"}),(0,t.jsxs)("div",{className:"mt-2",children:[(0,t.jsxs)(N.Text,{children:["User Keys: ",ev.keys.filter(e=>e.user_id).length]}),(0,t.jsxs)(N.Text,{children:["Service Account Keys: ",ev.keys.filter(e=>!e.user_id).length]}),(0,t.jsxs)(N.Text,{className:"text-gray-500",children:["Total: ",ev.keys.length]})]})]}),(0,t.jsx)(et.default,{objectPermission:tT.object_permission,variant:"card",accessToken:o}),(0,t.jsx)(S.Card,{children:(0,t.jsx)(J,{globalGuardrailNames:eX,teamGuardrails:Array.isArray(tT.metadata?.guardrails)?tT.metadata.guardrails:[],optedOutGlobalGuardrails:Array.isArray(tT.metadata?.opted_out_global_guardrails)?tT.metadata.opted_out_global_guardrails:[],killSwitchOn:tS,variant:"inline"})}),(0,t.jsxs)(S.Card,{children:[(0,t.jsx)(N.Text,{className:"font-semibold text-gray-900 mb-3",children:"Policies"}),tT.policies&&tT.policies.length>0?(0,t.jsx)("div",{className:"space-y-4",children:tT.policies.map((e,l)=>(0,t.jsxs)("div",{className:"space-y-2",children:[(0,t.jsxs)("div",{className:"flex items-center gap-2",children:[(0,t.jsx)(T.Badge,{color:"purple",children:e}),e5&&(0,t.jsx)(N.Text,{className:"text-xs text-gray-400",children:"Loading guardrails..."})]}),!e5&&e4[e]&&e4[e].length>0&&(0,t.jsxs)("div",{className:"ml-4 pl-3 border-l-2 border-gray-200",children:[(0,t.jsx)(N.Text,{className:"text-xs text-gray-500 mb-1",children:"Resolved Guardrails:"}),(0,t.jsx)("div",{className:"flex flex-wrap gap-1",children:e4[e].map((e,l)=>(0,t.jsx)(T.Badge,{color:"blue",size:"xs",children:e},l))})]})]},l))}):(0,t.jsx)(N.Text,{className:"text-gray-500",children:"No policies configured"})]}),(0,t.jsx)(Q.default,{loggingConfigs:tT.metadata?.logging||[],disabledCallbacks:[],variant:"card"})]})},{key:ez,label:eD[ez],children:(0,t.jsx)(eI,{teamId:e})},{key:eA,label:eD[eA],children:(0,t.jsx)(e0,{teamId:e,teamAlias:tT.team_alias,organization:ti})},{key:eF,label:eD[eF],children:(0,t.jsx)(eB,{teamData:ev,canEditTeam:tp,handleMemberDelete:e=>{e8(e),e9(!0)},setSelectedEditMember:eU,setIsEditMemberModalVisible:eR,setIsAddMemberModalVisible:eC})},{key:eO,label:eD[eO],children:(0,t.jsx)(eb,{teamId:e,accessToken:o,canEditTeam:tp})},{key:eP,label:eD[eP],children:(0,t.jsxs)(S.Card,{className:"overflow-y-auto max-h-[65vh]",children:[(0,t.jsxs)("div",{className:"flex justify-between items-center mb-4",children:[(0,t.jsx)(k.Title,{children:"Team Settings"}),tp&&!eE&&(0,t.jsx)(I.Button,{icon:(0,t.jsx)(g.EditOutlined,{className:"h-4 w-4"}),onClick:()=>e$(!0),children:"Edit Settings"})]}),eE&&eY?(0,t.jsx)("div",{className:"p-4",children:"Loading..."}):eE?(0,t.jsxs)(M.Form,{form:ek,onFinish:tv,onValuesChange:e=>{if("disable_global_guardrails"in e){let t=!0===e.disable_global_guardrails,l=(ek.getFieldValue("guardrails")||[]).filter(e=>!eX.has(e));ek.setFieldValue("guardrails",t?l:[...Array.from(eX),...l])}},initialValues:{...tT,team_alias:tT.team_alias,models:tT.models,tpm_limit:tT.tpm_limit,rpm_limit:tT.rpm_limit,object_permission_search_tools:tT.object_permission?.search_tools||[],modelLimits:Array.from(new Set([...Object.keys(tT.metadata?.model_tpm_limit??{}),...Object.keys(tT.metadata?.model_rpm_limit??{})])).map(e=>({model:e,tpm:tT.metadata?.model_tpm_limit?.[e],rpm:tT.metadata?.model_rpm_limit?.[e]})),max_budget:tT.max_budget,soft_budget:tT.soft_budget,budget_duration:tT.budget_duration,team_member_tpm_limit:tT.team_member_budget_table?.tpm_limit,team_member_rpm_limit:tT.team_member_budget_table?.rpm_limit,team_member_budget:tT.team_member_budget_table?.max_budget,team_member_budget_duration:tT.team_member_budget_table?.budget_duration,guardrails:tC,policies:tT.policies||[],disable_global_guardrails:tT.metadata?.disable_global_guardrails||!1,soft_budget_alerting_emails:Array.isArray(tT.metadata?.soft_budget_alerting_emails)?tT.metadata.soft_budget_alerting_emails.join(", "):"",metadata:tT.metadata?JSON.stringify((({logging:e,secret_manager_settings:t,soft_budget_alerting_emails:l,model_tpm_limit:a,model_rpm_limit:s,...i})=>i)(tT.metadata),null,2):"",logging_settings:tT.metadata?.logging||[],secret_manager_settings:tT.metadata?.secret_manager_settings?JSON.stringify(tT.metadata.secret_manager_settings,null,2):"",organization_id:tT.organization_id,vector_stores:tT.object_permission?.vector_stores||[],mcp_servers:tT.object_permission?.mcp_servers||[],mcp_access_groups:tT.object_permission?.mcp_access_groups||[],mcp_servers_and_groups:{servers:tT.object_permission?.mcp_servers||[],accessGroups:tT.object_permission?.mcp_access_groups||[],toolsets:tT.object_permission?.mcp_toolsets||[]},mcp_tool_permissions:tT.object_permission?.mcp_tool_permissions||{},agents_and_groups:{agents:tT.object_permission?.agents||[],accessGroups:tT.object_permission?.agent_access_groups||[]},access_group_ids:tT.access_group_ids||[],default_team_member_models:tT.default_team_member_models||[]},layout:"vertical",children:[(0,t.jsx)(M.Form.Item,{label:"Team Name",name:"team_alias",rules:[{required:!0,message:"Please input a team name"}],children:(0,t.jsx)(z.Input,{type:""})}),(0,t.jsx)(M.Form.Item,{label:"Models",name:"models",rules:[{required:!0,message:"Please select at least one model"}],children:(0,t.jsx)(Z.ModelSelect,{value:ek.getFieldValue("models")||[],onChange:e=>ek.setFieldValue("models",e),teamID:e,organizationID:ev?.team_info?.organization_id||void 0,options:{includeSpecialOptions:!0,includeUserModels:!ev?.team_info?.organization_id,showAllProxyModelsOverride:(0,u.isProxyAdminRole)(tn)&&!ev?.team_info?.organization_id},context:"team",dataTestId:"models-select"})}),(0,t.jsx)(M.Form.Item,{label:"Max Budget (USD)",name:"max_budget",children:(0,t.jsx)(el.default,{step:.01,precision:2,style:{width:"100%"}})}),(0,t.jsx)(M.Form.Item,{label:"Soft Budget (USD)",name:"soft_budget",children:(0,t.jsx)(el.default,{step:.01,precision:2,style:{width:"100%"}})}),(0,t.jsx)(M.Form.Item,{label:"Soft Budget Alerting Emails",name:"soft_budget_alerting_emails",tooltip:"Comma-separated email addresses to receive alerts when the soft budget is reached",children:(0,t.jsx)(z.Input,{placeholder:"example1@test.com, example2@test.com"})}),(0,t.jsxs)(j.Accordion,{className:"mt-4 mb-4",children:[(0,t.jsx)(v.AccordionHeader,{children:(0,t.jsx)("b",{children:"Team Member Settings"})}),(0,t.jsxs)(f.AccordionBody,{children:[(0,t.jsx)(N.Text,{className:"text-xs text-gray-500 mb-4",children:"Optional defaults applied when members join this team. All fields can be overridden per member."}),(0,t.jsx)(M.Form.Item,{label:(0,t.jsxs)("span",{children:["Default Model Access"," ",(0,t.jsx)(R.Tooltip,{title:"Optional. If set, new members can only access these models by default. Must be a subset of the team's models above. Leave empty to give all members access to all team models.",children:(0,t.jsx)(p.InfoCircleOutlined,{style:{marginLeft:"4px"}})})]}),name:"default_team_member_models",children:(0,t.jsx)(M.Form.Item,{noStyle:!0,shouldUpdate:(e,t)=>e.models!==t.models,children:({getFieldValue:e})=>{let l=e("models")||tT.models||[];return(0,t.jsx)(F.Select,{mode:"multiple",placeholder:"Leave empty — all team models accessible to every member",value:ek.getFieldValue("default_team_member_models")||[],onChange:e=>ek.setFieldValue("default_team_member_models",e),options:l.map(e=>({label:e,value:e}))})}})}),(0,t.jsx)(M.Form.Item,{label:"Default Budget (USD)",name:"team_member_budget",tooltip:"Default spend budget for each member in this team.",children:(0,t.jsx)(el.default,{step:.01,precision:2,style:{width:"100%"}})}),(0,t.jsx)(M.Form.Item,{label:"Default Budget Duration",name:"team_member_budget_duration",children:(0,t.jsx)(W,{onChange:e=>ek.setFieldValue("team_member_budget_duration",e),value:ek.getFieldValue("team_member_budget_duration")})}),(0,t.jsx)(M.Form.Item,{label:"Default Key Duration (eg: 1d, 1mo)",name:"team_member_key_duration",tooltip:"Set a limit to the duration of a team member's key. Format: 30s (seconds), 30m (minutes), 30h (hours), 30d (days), 1mo (month)",children:(0,t.jsx)(C.TextInput,{placeholder:"e.g., 30d"})}),(0,t.jsx)(M.Form.Item,{label:"Default TPM Limit",name:"team_member_tpm_limit",tooltip:"Default tokens per minute limit for each member. Can be overridden per member.",children:(0,t.jsx)(el.default,{step:1,style:{width:"100%"},placeholder:"e.g., 1000"})}),(0,t.jsx)(M.Form.Item,{label:"Default RPM Limit",name:"team_member_rpm_limit",tooltip:"Default requests per minute limit for each member. Can be overridden per member.",children:(0,t.jsx)(el.default,{step:1,style:{width:"100%"},placeholder:"e.g., 100"})})]})]}),(0,t.jsx)(M.Form.Item,{label:"Reset Budget",name:"budget_duration",children:(0,t.jsxs)(F.Select,{placeholder:"n/a",children:[(0,t.jsx)(F.Select.Option,{value:"24h",children:"daily"}),(0,t.jsx)(F.Select.Option,{value:"7d",children:"weekly"}),(0,t.jsx)(F.Select.Option,{value:"30d",children:"monthly"})]})}),(0,t.jsx)(M.Form.Item,{label:"Tokens per minute Limit (TPM)",name:"tpm_limit",children:(0,t.jsx)(el.default,{step:1,style:{width:"100%"}})}),(0,t.jsx)(M.Form.Item,{label:"Requests per minute Limit (RPM)",name:"rpm_limit",children:(0,t.jsx)(el.default,{step:1,style:{width:"100%"}})}),(0,t.jsx)(M.Form.Item,{label:"Model-Specific Rate Limits",tooltip:"Set per-model TPM/RPM limits that apply across the whole team.",children:(0,t.jsx)(M.Form.List,{name:"modelLimits",children:(e,{add:l,remove:a})=>(0,t.jsxs)(t.Fragment,{children:[e.map(({key:e,name:l,...s})=>(0,t.jsxs)(O.Space,{style:{display:"flex",marginBottom:8},align:"baseline",children:[(0,t.jsx)(M.Form.Item,{...s,name:[l,"model"],rules:[{required:!0,message:"Missing model"},{validator:(e,t)=>t&&(ek.getFieldValue("modelLimits")??[]).filter(e=>e?.model===t).length>1?Promise.reject(Error("Duplicate model")):Promise.resolve()}],style:{minWidth:240},children:(0,t.jsx)(F.Select,{showSearch:!0,placeholder:"Select model",allowClear:!0,options:th.map(e=>({value:e,label:e}))})}),(0,t.jsx)(M.Form.Item,{...s,name:[l,"tpm"],rules:[{validator:async(e,t)=>{let a=(ek.getFieldValue("modelLimits")??[])[l]??{};return a.model&&null==t&&null==a.rpm?Promise.reject(Error("Set at least one of TPM or RPM")):Promise.resolve()}}],children:(0,t.jsx)(A.InputNumber,{placeholder:"TPM Limit",min:0})}),(0,t.jsx)(M.Form.Item,{...s,name:[l,"rpm"],children:(0,t.jsx)(A.InputNumber,{placeholder:"RPM Limit",min:0})}),(0,t.jsx)(x.MinusCircleOutlined,{onClick:()=>a(l),style:{color:"#ef4444"}})]},e)),(0,t.jsx)(M.Form.Item,{children:(0,t.jsx)(I.Button,{type:"dashed",onClick:()=>l(),block:!0,icon:(0,t.jsx)(_.PlusOutlined,{}),children:"Add Model Limit"})})]})})}),(0,t.jsx)(M.Form.Item,{label:"Router Settings",children:(0,t.jsx)(er.default,{ref:ts,accessToken:o||"",value:tT.router_settings?{router_settings:tT.router_settings}:void 0})}),(0,t.jsx)(M.Form.Item,{label:(0,t.jsxs)("span",{children:["Guardrails"," ",(0,t.jsx)(R.Tooltip,{title:"Select which guardrails apply to this team. Global guardrails are enabled by default — uncheck to opt out. Other guardrails are opt-in.",children:(0,t.jsx)("a",{href:"https://docs.litellm.ai/docs/proxy/guardrails/quick_start",target:"_blank",rel:"noopener noreferrer",onClick:e=>e.stopPropagation(),children:(0,t.jsx)(p.InfoCircleOutlined,{style:{marginLeft:"4px"}})})})]}),name:"guardrails",children:(0,t.jsxs)(F.Select,{mode:"multiple",placeholder:"Select guardrails",optionLabelProp:"label",tagRender:({label:e,value:l,closable:a,onClose:s})=>{let i=eX.has(l);return(0,t.jsxs)(L.Tag,{color:"blue",closable:a,onClose:s,onMouseDown:tk,style:{marginInlineEnd:4},children:[i&&(0,t.jsx)(h.GlobalOutlined,{style:{marginInlineEnd:4},"aria-label":"Global guardrail"}),e]})},children:[(0,t.jsx)(F.Select.OptGroup,{label:(0,t.jsxs)(t.Fragment,{children:[(0,t.jsx)(h.GlobalOutlined,{style:{marginInlineEnd:4}}),"Global"]}),children:(eQ?.guardrails??[]).filter(e=>e.litellm_params?.default_on).map(e=>(0,t.jsx)(F.Select.Option,{value:e.guardrail_name,label:e.guardrail_name,disabled:tg,children:e.guardrail_name},e.guardrail_name))}),(0,t.jsx)(F.Select.OptGroup,{label:"Other",children:(eQ?.guardrails??[]).filter(e=>!e.litellm_params?.default_on).map(e=>(0,t.jsx)(F.Select.Option,{value:e.guardrail_name,label:e.guardrail_name,children:e.guardrail_name},e.guardrail_name))})]})}),(0,t.jsx)(M.Form.Item,{label:(0,t.jsxs)("span",{children:["Disable all global guardrails"," ",(0,t.jsx)(R.Tooltip,{title:"Kill switch: bypass every global guardrail for this team, including any added in the future. For per-guardrail opt-out instead, use the Guardrails dropdown above.",children:(0,t.jsx)(p.InfoCircleOutlined,{style:{marginLeft:"4px"}})})]}),name:"disable_global_guardrails",valuePropName:"checked",children:(0,t.jsx)(P.Switch,{checkedChildren:"Yes",unCheckedChildren:"No"})}),(0,t.jsx)(M.Form.Item,{label:(0,t.jsxs)("span",{children:["Policies"," ",(0,t.jsx)(R.Tooltip,{title:"Apply policies to this team to control guardrails and other settings",children:(0,t.jsx)("a",{href:"https://docs.litellm.ai/docs/proxy/guardrails/guardrail_policies",target:"_blank",rel:"noopener noreferrer",onClick:e=>e.stopPropagation(),children:(0,t.jsx)(p.InfoCircleOutlined,{style:{marginLeft:"4px"}})})})]}),name:"policies",children:(0,t.jsx)(F.Select,{mode:"tags",placeholder:"Select or enter policies",options:eZ.map(e=>({value:e,label:e}))})}),(0,t.jsx)(M.Form.Item,{label:(0,t.jsxs)("span",{children:["Access Groups"," ",(0,t.jsx)(R.Tooltip,{title:"Assign access groups to this team. Access groups control which models, MCP servers, and agents this team can use",children:(0,t.jsx)(p.InfoCircleOutlined,{style:{marginLeft:"4px"}})})]}),name:"access_group_ids",children:(0,t.jsx)($.default,{placeholder:"Select access groups (optional)"})}),(0,t.jsx)(M.Form.Item,{label:"Vector Stores",name:"vector_stores","aria-label":"Vector Stores",children:(0,t.jsx)(ea.default,{onChange:e=>ek.setFieldValue("vector_stores",e),value:ek.getFieldValue("vector_stores"),accessToken:o||"",placeholder:"Select vector stores"})}),(0,t.jsx)(M.Form.Item,{label:"Allowed Pass Through Routes",name:"allowed_passthrough_routes",children:(0,t.jsx)(q.default,{onChange:e=>ek.setFieldValue("allowed_passthrough_routes",e),value:ek.getFieldValue("allowed_passthrough_routes"),accessToken:o||"",placeholder:"Select pass through routes"})}),(0,t.jsx)(M.Form.Item,{label:"MCP Servers / Access Groups",name:"mcp_servers_and_groups",children:(0,t.jsx)(Y.default,{onChange:e=>ek.setFieldValue("mcp_servers_and_groups",e),value:ek.getFieldValue("mcp_servers_and_groups"),accessToken:o||"",placeholder:"Select MCP servers or access groups (optional)"})}),(0,t.jsx)(M.Form.Item,{name:"mcp_tool_permissions",initialValue:{},hidden:!0,children:(0,t.jsx)(z.Input,{type:"hidden"})}),(0,t.jsx)(M.Form.Item,{noStyle:!0,shouldUpdate:(e,t)=>e.mcp_servers_and_groups!==t.mcp_servers_and_groups||e.mcp_tool_permissions!==t.mcp_tool_permissions,children:()=>(0,t.jsx)("div",{className:"mb-6",children:(0,t.jsx)(X.default,{accessToken:o||"",selectedServers:ek.getFieldValue("mcp_servers_and_groups")?.servers||[],toolPermissions:ek.getFieldValue("mcp_tool_permissions")||{},onChange:e=>ek.setFieldsValue({mcp_tool_permissions:e})})})}),(0,t.jsx)(M.Form.Item,{label:"Agents / Access Groups",name:"agents_and_groups",children:(0,t.jsx)(K.default,{onChange:e=>ek.setFieldValue("agents_and_groups",e),value:ek.getFieldValue("agents_and_groups"),accessToken:o||"",placeholder:"Select agents or access groups (optional)"})}),(0,t.jsxs)(j.Accordion,{className:"mt-4 mb-4",children:[(0,t.jsx)(v.AccordionHeader,{children:(0,t.jsx)("b",{children:"Search Tool Settings"})}),(0,t.jsx)(f.AccordionBody,{children:(0,t.jsx)(M.Form.Item,{label:"Allowed Search Tools",name:"object_permission_search_tools",tooltip:"Select which search tools this team can access. Leave empty to allow all search tools.",children:(0,t.jsx)(es,{onChange:e=>ek.setFieldValue("object_permission_search_tools",e),value:ek.getFieldValue("object_permission_search_tools"),accessToken:o||"",placeholder:"Select search tools (optional, empty = all allowed)"})})})]}),(0,t.jsx)(M.Form.Item,{label:"Organization",name:"organization_id",children:(0,t.jsx)(F.Select,{allowClear:!0,placeholder:"Select an organization",showSearch:!0,optionFilterProp:"label",options:td.map(e=>({value:e.organization_id,label:e.organization_alias||e.organization_id}))})}),(0,t.jsx)(M.Form.Item,{label:"Logging Settings",name:"logging_settings",children:(0,t.jsx)(ei.default,{value:ek.getFieldValue("logging_settings"),onChange:e=>ek.setFieldValue("logging_settings",e)})}),(0,t.jsx)(M.Form.Item,{label:"Secret Manager Settings",name:"secret_manager_settings",help:eg?"Enter secret manager configuration as a JSON object.":"Premium feature - Upgrade to manage secret manager settings.",rules:[{validator:async(e,t)=>{if(!t)return Promise.resolve();try{return JSON.parse(t),Promise.resolve()}catch(e){return Promise.reject(Error("Please enter valid JSON"))}}}],children:(0,t.jsx)(z.Input.TextArea,{rows:6,placeholder:'{"namespace": "admin", "mount": "secret", "path_prefix": "litellm"}',disabled:!eg})}),(0,t.jsx)(M.Form.Item,{label:"Metadata",name:"metadata",children:(0,t.jsx)(z.Input.TextArea,{rows:10})}),(0,t.jsx)("div",{className:"sticky z-10 bg-white p-4 pr-0 border-t border-gray-200 bottom-[-1.5rem] inset-x-[-1.5rem]",children:(0,t.jsxs)("div",{className:"flex justify-end items-center gap-2",children:[(0,t.jsx)(I.Button,{onClick:()=>e$(!1),disabled:tl,children:"Cancel"}),(0,t.jsx)(I.Button,{icon:(0,t.jsx)(b.SaveOutlined,{className:"h-4 w-4"}),type:"primary",htmlType:"submit",loading:tl,children:"Save Changes"})]})})]}):(0,t.jsxs)("div",{className:"space-y-4",children:[(0,t.jsxs)("div",{children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Team Name"}),(0,t.jsx)("div",{children:tT.team_alias})]}),(0,t.jsxs)("div",{children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Team ID"}),(0,t.jsx)("div",{className:"font-mono",children:tT.team_id})]}),(0,t.jsxs)("div",{children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Created At"}),(0,t.jsx)("div",{children:new Date(tT.created_at).toLocaleString()})]}),(0,t.jsxs)("div",{children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Models"}),(0,t.jsx)("div",{className:"flex flex-wrap gap-2 mt-1",children:tT.models.map((e,l)=>(0,t.jsx)(T.Badge,{color:"red",children:e},l))})]}),tT.default_team_member_models&&tT.default_team_member_models.length>0&&(0,t.jsxs)("div",{children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Default Member Models"}),(0,t.jsx)("div",{className:"flex flex-wrap gap-2 mt-1",children:tT.default_team_member_models.map((e,l)=>(0,t.jsx)(T.Badge,{color:"blue",children:e},l))})]}),(0,t.jsxs)("div",{children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Rate Limits"}),(0,t.jsxs)("div",{children:["TPM: ",tT.tpm_limit||"Unlimited"]}),(0,t.jsxs)("div",{children:["RPM: ",tT.rpm_limit||"Unlimited"]}),(ey=tT.metadata?.model_tpm_limit??{},ej=tT.metadata?.model_rpm_limit??{},0===(ef=Array.from(new Set([...Object.keys(ey),...Object.keys(ej)]))).length?null:(0,t.jsxs)("div",{className:"mt-2",children:[(0,t.jsx)(N.Text,{className:"text-gray-500",children:"Per-model limits:"}),ef.map(e=>(0,t.jsxs)("div",{className:"text-xs ml-2",children:[e,": TPM ",ey[e]??"—",", RPM ",ej[e]??"—"]},e))]}))]}),(0,t.jsxs)("div",{children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Team Budget"}),(0,t.jsxs)("div",{children:["Max Budget:"," ",null!==tT.max_budget?`$${(0,m.formatNumberWithCommas)(tT.max_budget,4)}`:"No Limit"]}),(0,t.jsxs)("div",{children:["Soft Budget:"," ",null!==tT.soft_budget&&void 0!==tT.soft_budget?`$${(0,m.formatNumberWithCommas)(tT.soft_budget,4)}`:"No Limit"]}),(0,t.jsxs)("div",{children:["Budget Reset: ",tT.budget_duration||"Never"]}),tT.metadata?.soft_budget_alerting_emails&&Array.isArray(tT.metadata.soft_budget_alerting_emails)&&tT.metadata.soft_budget_alerting_emails.length>0&&(0,t.jsxs)("div",{children:["Soft Budget Alerting Emails: ",tT.metadata.soft_budget_alerting_emails.join(", ")]})]}),(0,t.jsxs)("div",{children:[(0,t.jsxs)(N.Text,{className:"font-medium",children:["Team Member Settings"," ",(0,t.jsx)(R.Tooltip,{title:"These are limits on individual team members",children:(0,t.jsx)(p.InfoCircleOutlined,{style:{marginLeft:"4px"}})})]}),(0,t.jsxs)("div",{children:["Max Budget: ",tT.team_member_budget_table?.max_budget||"No Limit"]}),(0,t.jsxs)("div",{children:["Budget Duration: ",tT.team_member_budget_table?.budget_duration||"No Limit"]}),(0,t.jsxs)("div",{children:["Key Duration: ",tT.metadata?.team_member_key_duration||"No Limit"]}),(0,t.jsxs)("div",{children:["TPM Limit: ",tT.team_member_budget_table?.tpm_limit||"No Limit"]}),(0,t.jsxs)("div",{children:["RPM Limit: ",tT.team_member_budget_table?.rpm_limit||"No Limit"]})]}),(0,t.jsxs)("div",{children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Router Settings"}),tT.router_settings&&Object.values(tT.router_settings).some(e=>null!=e&&""!==e&&!(Array.isArray(e)&&0===e.length))?(0,t.jsxs)("div",{className:"mt-1 space-y-1",children:[tT.router_settings.routing_strategy&&(0,t.jsxs)("div",{children:["Routing Strategy:"," ",(0,t.jsx)(T.Badge,{color:"blue",children:tT.router_settings.routing_strategy})]}),null!=tT.router_settings.num_retries&&(0,t.jsxs)("div",{children:["Number of Retries: ",tT.router_settings.num_retries]}),null!=tT.router_settings.allowed_fails&&(0,t.jsxs)("div",{children:["Allowed Failures: ",tT.router_settings.allowed_fails]}),null!=tT.router_settings.cooldown_time&&(0,t.jsxs)("div",{children:["Cooldown Time: ",tT.router_settings.cooldown_time,"s"]}),null!=tT.router_settings.timeout&&(0,t.jsxs)("div",{children:["Timeout: ",tT.router_settings.timeout,"s"]}),null!=tT.router_settings.retry_after&&(0,t.jsxs)("div",{children:["Retry After: ",tT.router_settings.retry_after,"s"]}),tT.router_settings.fallbacks&&Array.isArray(tT.router_settings.fallbacks)&&tT.router_settings.fallbacks.length>0&&(0,t.jsxs)("div",{children:["Fallbacks: ",tT.router_settings.fallbacks.length," configured"]}),tT.router_settings.enable_tag_filtering&&(0,t.jsx)("div",{children:"Tag Filtering: Enabled"})]}):(0,t.jsx)("div",{className:"text-gray-400",children:"No router settings configured"})]}),(0,t.jsxs)("div",{children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Organization ID"}),(0,t.jsx)("div",{children:tT.organization_id})]}),(0,t.jsxs)("div",{children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Status"}),(0,t.jsx)(T.Badge,{color:tT.blocked?"red":"green",children:tT.blocked?"Blocked":"Active"})]}),(0,t.jsx)(et.default,{objectPermission:tT.object_permission,variant:"inline",className:"pt-4 border-t border-gray-200",accessToken:o}),(0,t.jsx)(J,{globalGuardrailNames:eX,teamGuardrails:Array.isArray(tT.metadata?.guardrails)?tT.metadata.guardrails:[],optedOutGlobalGuardrails:Array.isArray(tT.metadata?.opted_out_global_guardrails)?tT.metadata.opted_out_global_guardrails:[],killSwitchOn:tS,variant:"inline",className:"pt-4 border-t border-gray-200"}),(0,t.jsx)(Q.default,{loggingConfigs:tT.metadata?.logging||[],disabledCallbacks:[],variant:"inline",className:"pt-4 border-t border-gray-200"}),tT.metadata?.secret_manager_settings&&(0,t.jsxs)("div",{className:"pt-4 border-t border-gray-200",children:[(0,t.jsx)(N.Text,{className:"font-medium",children:"Secret Manager Settings"}),(0,t.jsx)("pre",{className:"mt-2 bg-gray-50 p-3 rounded text-xs overflow-x-auto",children:JSON.stringify(tT.metadata.secret_manager_settings,null,2)})]})]})]})}].filter(e=>tx.includes(e.key))}),(0,t.jsx)(en.default,{visible:eL,onCancel:()=>eR(!1),onSubmit:tj,initialData:eV,mode:"edit",config:{title:"Edit Member",showEmail:!0,showUserId:!0,roleOptions:[{label:"Admin",value:"admin"},{label:"User",value:"user"}],additionalFields:[{name:"max_budget_in_team",label:(0,t.jsxs)("span",{children:["Team Member Budget (USD)"," ",(0,t.jsx)(R.Tooltip,{title:"Maximum amount in USD this member can spend within this team. This is separate from any global user budget limits",children:(0,t.jsx)(p.InfoCircleOutlined,{style:{marginLeft:"4px"}})})]}),type:"numerical",step:.01,min:0,placeholder:"Budget limit for this member within this team"},{name:"tpm_limit",label:(0,t.jsxs)("span",{children:["Team Member TPM Limit"," ",(0,t.jsx)(R.Tooltip,{title:"Maximum tokens per minute this member can use within this team. This is separate from any global user TPM limit",children:(0,t.jsx)(p.InfoCircleOutlined,{style:{marginLeft:"4px"}})})]}),type:"numerical",step:1,min:0,placeholder:"Tokens per minute limit for this member in this team"},{name:"rpm_limit",label:(0,t.jsxs)("span",{children:["Team Member RPM Limit"," ",(0,t.jsx)(R.Tooltip,{title:"Maximum requests per minute this member can make within this team. This is separate from any global user RPM limit",children:(0,t.jsx)(p.InfoCircleOutlined,{style:{marginLeft:"4px"}})})]}),type:"numerical",step:1,min:0,placeholder:"Requests per minute limit for this member in this team"},{name:"allowed_models",label:(0,t.jsxs)("span",{children:["Allowed Models"," ",(0,t.jsx)(R.Tooltip,{title:"Models this member can access within this team. Leave empty to inherit all team models.",children:(0,t.jsx)(p.InfoCircleOutlined,{style:{marginLeft:"4px"}})})]}),type:"multi-select",options:(tT.models||[]).map(e=>({label:e,value:e})),placeholder:"Leave empty to inherit all team models"}]}}),(0,t.jsx)(i.default,{isVisible:eN,onCancel:()=>eC(!1),onSubmit:ty,accessToken:o,teamId:e}),(0,t.jsx)(G.default,{isOpen:e7,title:"Delete Team Member",alertMessage:"Removing team members will also delete any keys created by or created for this member.",message:"Are you sure you want to remove this member from the team? This action cannot be undone.",resourceInformationTitle:"Team Member Information",resourceInformation:[{label:"User ID",value:e6?.user_id,code:!0},{label:"Email",value:e6?.user_email},{label:"Role",value:e6?.role}],onCancel:()=>{e9(!1),e8(null)},onOk:tf,confirmLoading:te})]})}],56567)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/04711b0f8ffa7bbd.js b/litellm/proxy/_experimental/out/_next/static/chunks/04711b0f8ffa7bbd.js new file mode 100644 index 00000000000..6cfa66f43a4 --- /dev/null +++ b/litellm/proxy/_experimental/out/_next/static/chunks/04711b0f8ffa7bbd.js @@ -0,0 +1,7 @@ +(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,309821,e=>{"use strict";e.i(247167);var t=e.i(271645);e.i(262370);var r=e.i(135551),n=e.i(201072),o=e.i(121229),i=e.i(726289),l=e.i(864517),a=e.i(343794),s=e.i(529681),c=e.i(242064),u=e.i(931067),d=e.i(209428),p=e.i(703923),f={percent:0,prefixCls:"rc-progress",strokeColor:"#2db7f5",strokeLinecap:"round",strokeWidth:1,trailColor:"#D9D9D9",trailWidth:1,gapPosition:"bottom"},g=function(){var e=(0,t.useRef)([]),r=(0,t.useRef)(null);return(0,t.useEffect)(function(){var t=Date.now(),n=!1;e.current.forEach(function(e){if(e){n=!0;var o=e.style;o.transitionDuration=".3s, .3s, .3s, .06s",r.current&&t-r.current<100&&(o.transitionDuration="0s, 0s")}}),n&&(r.current=Date.now())}),e.current},m=e.i(410160),b=e.i(392221),h=e.i(654310),v=0,y=(0,h.default)();let $=function(e){var r=t.useState(),n=(0,b.default)(r,2),o=n[0],i=n[1];return t.useEffect(function(){var e;i("rc_progress_".concat((y?(e=v,v+=1):e="TEST_OR_SSR",e)))},[]),e||o};var C=function(e){var r=e.bg,n=e.children;return t.createElement("div",{style:{width:"100%",height:"100%",background:r}},n)};function k(e,t){return Object.keys(e).map(function(r){var n=parseFloat(r),o="".concat(Math.floor(n*t),"%");return"".concat(e[r]," ").concat(o)})}var x=t.forwardRef(function(e,r){var n=e.prefixCls,o=e.color,i=e.gradientId,l=e.radius,a=e.style,s=e.ptg,c=e.strokeLinecap,u=e.strokeWidth,d=e.size,p=e.gapDegree,f=o&&"object"===(0,m.default)(o),g=d/2,b=t.createElement("circle",{className:"".concat(n,"-circle-path"),r:l,cx:g,cy:g,stroke:f?"#FFF":void 0,strokeLinecap:c,strokeWidth:u,opacity:+(0!==s),style:a,ref:r});if(!f)return b;var h="".concat(i,"-conic"),v=k(o,(360-p)/360),y=k(o,1),$="conic-gradient(from ".concat(p?"".concat(180+p/2,"deg"):"0deg",", ").concat(v.join(", "),")"),x="linear-gradient(to ".concat(p?"bottom":"top",", ").concat(y.join(", "),")");return t.createElement(t.Fragment,null,t.createElement("mask",{id:h},b),t.createElement("foreignObject",{x:0,y:0,width:d,height:d,mask:"url(#".concat(h,")")},t.createElement(C,{bg:x},t.createElement(C,{bg:$}))))}),S=function(e,t,r,n,o,i,l,a,s,c){var u=arguments.length>10&&void 0!==arguments[10]?arguments[10]:0,d=(100-n)/100*t;return"round"===s&&100!==n&&(d+=c/2)>=t&&(d=t-.01),{stroke:"string"==typeof a?a:void 0,strokeDasharray:"".concat(t,"px ").concat(e),strokeDashoffset:d+u,transform:"rotate(".concat(o+r/100*360*((360-i)/360)+(0===i?0:({bottom:0,top:180,left:90,right:-90})[l]),"deg)"),transformOrigin:"".concat(50,"px ").concat(50,"px"),transition:"stroke-dashoffset .3s ease 0s, stroke-dasharray .3s ease 0s, stroke .3s, stroke-width .06s ease .3s, opacity .3s ease 0s",fillOpacity:0}},O=["id","prefixCls","steps","strokeWidth","trailWidth","gapDegree","gapPosition","trailColor","strokeLinecap","style","className","strokeColor","percent"];function w(e){var t=null!=e?e:[];return Array.isArray(t)?t:[t]}let E=function(e){var r,n,o,i,l=(0,d.default)((0,d.default)({},f),e),s=l.id,c=l.prefixCls,b=l.steps,h=l.strokeWidth,v=l.trailWidth,y=l.gapDegree,C=void 0===y?0:y,k=l.gapPosition,E=l.trailColor,j=l.strokeLinecap,N=l.style,I=l.className,P=l.strokeColor,D=l.percent,R=(0,p.default)(l,O),z=$(s),A="".concat(z,"-gradient"),M=50-h/2,T=2*Math.PI*M,W=C>0?90+C/2:-90,B=(360-C)/360*T,F="object"===(0,m.default)(b)?b:{count:b,gap:2},X=F.count,L=F.gap,H=w(D),_=w(P),q=_.find(function(e){return e&&"object"===(0,m.default)(e)}),G=q&&"object"===(0,m.default)(q)?"butt":j,V=S(T,B,0,100,W,C,k,E,G,h),K=g();return t.createElement("svg",(0,u.default)({className:(0,a.default)("".concat(c,"-circle"),I),viewBox:"0 0 ".concat(100," ").concat(100),style:N,id:s,role:"presentation"},R),!X&&t.createElement("circle",{className:"".concat(c,"-circle-trail"),r:M,cx:50,cy:50,stroke:E,strokeLinecap:G,strokeWidth:v||h,style:V}),X?(r=Math.round(X*(H[0]/100)),n=100/X,o=0,Array(X).fill(null).map(function(e,i){var l=i<=r-1?_[0]:E,a=l&&"object"===(0,m.default)(l)?"url(#".concat(A,")"):void 0,s=S(T,B,o,n,W,C,k,l,"butt",h,L);return o+=(B-s.strokeDashoffset+L)*100/B,t.createElement("circle",{key:i,className:"".concat(c,"-circle-path"),r:M,cx:50,cy:50,stroke:a,strokeWidth:h,opacity:1,style:s,ref:function(e){K[i]=e}})})):(i=0,H.map(function(e,r){var n=_[r]||_[_.length-1],o=S(T,B,i,e,W,C,k,n,G,h);return i+=e,t.createElement(x,{key:r,color:n,ptg:e,radius:M,prefixCls:c,gradientId:A,style:o,strokeLinecap:G,strokeWidth:h,gapDegree:C,ref:function(e){K[r]=e},size:100})}).reverse()))};var j=e.i(491816);e.i(765846);var N=e.i(896091);function I(e){return!e||e<0?0:e>100?100:e}function P({success:e,successPercent:t}){let r=t;return e&&"progress"in e&&(r=e.progress),e&&"percent"in e&&(r=e.percent),r}let D=(e,t,r)=>{var n,o,i,l;let a=-1,s=-1;if("step"===t){let t=r.steps,n=r.strokeWidth;"string"==typeof e||void 0===e?(a="small"===e?2:14,s=null!=n?n:8):"number"==typeof e?[a,s]=[e,e]:[a=14,s=8]=Array.isArray(e)?e:[e.width,e.height],a*=t}else if("line"===t){let t=null==r?void 0:r.strokeWidth;"string"==typeof e||void 0===e?s=t||("small"===e?6:8):"number"==typeof e?[a,s]=[e,e]:[a=-1,s=8]=Array.isArray(e)?e:[e.width,e.height]}else("circle"===t||"dashboard"===t)&&("string"==typeof e||void 0===e?[a,s]="small"===e?[60,60]:[120,120]:"number"==typeof e?[a,s]=[e,e]:Array.isArray(e)&&(a=null!=(o=null!=(n=e[0])?n:e[1])?o:120,s=null!=(l=null!=(i=e[0])?i:e[1])?l:120));return[a,s]},R=e=>{let{prefixCls:r,trailColor:n=null,strokeLinecap:o="round",gapPosition:i,gapDegree:l,width:s=120,type:c,children:u,success:d,size:p=s,steps:f}=e,[g,m]=D(p,"circle"),{strokeWidth:b}=e;void 0===b&&(b=Math.max(3/g*100,6));let h=t.useMemo(()=>l||0===l?l:"dashboard"===c?75:void 0,[l,c]),v=(({percent:e,success:t,successPercent:r})=>{let n=I(P({success:t,successPercent:r}));return[n,I(I(e)-n)]})(e),y="[object Object]"===Object.prototype.toString.call(e.strokeColor),$=(({success:e={},strokeColor:t})=>{let{strokeColor:r}=e;return[r||N.presetPrimaryColors.green,t||null]})({success:d,strokeColor:e.strokeColor}),C=(0,a.default)(`${r}-inner`,{[`${r}-circle-gradient`]:y}),k=t.createElement(E,{steps:f,percent:f?v[1]:v,strokeWidth:b,trailWidth:b,strokeColor:f?$[1]:$,strokeLinecap:o,trailColor:n,prefixCls:r,gapDegree:h,gapPosition:i||"dashboard"===c&&"bottom"||void 0}),x=g<=20,S=t.createElement("div",{className:C,style:{width:g,height:m,fontSize:.15*g+6}},k,!x&&u);return x?t.createElement(j.default,{title:u},S):S};e.i(296059);var z=e.i(694758),A=e.i(915654),M=e.i(183293),T=e.i(246422),W=e.i(838378);let B="--progress-line-stroke-color",F="--progress-percent",X=e=>{let t=e?"100%":"-100%";return new z.Keyframes(`antProgress${e?"RTL":"LTR"}Active`,{"0%":{transform:`translateX(${t}) scaleX(0)`,opacity:.1},"20%":{transform:`translateX(${t}) scaleX(0)`,opacity:.5},to:{transform:"translateX(0) scaleX(1)",opacity:0}})},L=(0,T.genStyleHooks)("Progress",e=>{let t=e.calc(e.marginXXS).div(2).equal(),r=(0,W.mergeToken)(e,{progressStepMarginInlineEnd:t,progressStepMinWidth:t,progressActiveMotionDuration:"2.4s"});return[(e=>{let{componentCls:t,iconCls:r}=e;return{[t]:Object.assign(Object.assign({},(0,M.resetComponent)(e)),{display:"inline-block","&-rtl":{direction:"rtl"},"&-line":{position:"relative",width:"100%",fontSize:e.fontSize},[`${t}-outer`]:{display:"inline-flex",alignItems:"center",width:"100%"},[`${t}-inner`]:{position:"relative",display:"inline-block",width:"100%",flex:1,overflow:"hidden",verticalAlign:"middle",backgroundColor:e.remainingColor,borderRadius:e.lineBorderRadius},[`${t}-inner:not(${t}-circle-gradient)`]:{[`${t}-circle-path`]:{stroke:e.defaultColor}},[`${t}-success-bg, ${t}-bg`]:{position:"relative",background:e.defaultColor,borderRadius:e.lineBorderRadius,transition:`all ${e.motionDurationSlow} ${e.motionEaseInOutCirc}`},[`${t}-layout-bottom`]:{display:"flex",flexDirection:"column",alignItems:"center",justifyContent:"center",[`${t}-text`]:{width:"max-content",marginInlineStart:0,marginTop:e.marginXXS}},[`${t}-bg`]:{overflow:"hidden","&::after":{content:'""',background:{_multi_value_:!0,value:["inherit",`var(${B})`]},height:"100%",width:`calc(1 / var(${F}) * 100%)`,display:"block"},[`&${t}-bg-inner`]:{minWidth:"max-content","&::after":{content:"none"},[`${t}-text-inner`]:{color:e.colorWhite,[`&${t}-text-bright`]:{color:"rgba(0, 0, 0, 0.45)"}}}},[`${t}-success-bg`]:{position:"absolute",insetBlockStart:0,insetInlineStart:0,backgroundColor:e.colorSuccess},[`${t}-text`]:{display:"inline-block",marginInlineStart:e.marginXS,color:e.colorText,lineHeight:1,width:"2em",whiteSpace:"nowrap",textAlign:"start",verticalAlign:"middle",wordBreak:"normal",[r]:{fontSize:e.fontSize},[`&${t}-text-outer`]:{width:"max-content"},[`&${t}-text-outer${t}-text-start`]:{width:"max-content",marginInlineStart:0,marginInlineEnd:e.marginXS}},[`${t}-text-inner`]:{display:"flex",justifyContent:"center",alignItems:"center",width:"100%",height:"100%",marginInlineStart:0,padding:`0 ${(0,A.unit)(e.paddingXXS)}`,[`&${t}-text-start`]:{justifyContent:"start"},[`&${t}-text-end`]:{justifyContent:"end"}},[`&${t}-status-active`]:{[`${t}-bg::before`]:{position:"absolute",inset:0,backgroundColor:e.colorBgContainer,borderRadius:e.lineBorderRadius,opacity:0,animationName:X(),animationDuration:e.progressActiveMotionDuration,animationTimingFunction:e.motionEaseOutQuint,animationIterationCount:"infinite",content:'""'}},[`&${t}-rtl${t}-status-active`]:{[`${t}-bg::before`]:{animationName:X(!0)}},[`&${t}-status-exception`]:{[`${t}-bg`]:{backgroundColor:e.colorError},[`${t}-text`]:{color:e.colorError}},[`&${t}-status-exception ${t}-inner:not(${t}-circle-gradient)`]:{[`${t}-circle-path`]:{stroke:e.colorError}},[`&${t}-status-success`]:{[`${t}-bg`]:{backgroundColor:e.colorSuccess},[`${t}-text`]:{color:e.colorSuccess}},[`&${t}-status-success ${t}-inner:not(${t}-circle-gradient)`]:{[`${t}-circle-path`]:{stroke:e.colorSuccess}}})}})(r),(e=>{let{componentCls:t,iconCls:r}=e;return{[t]:{[`${t}-circle-trail`]:{stroke:e.remainingColor},[`&${t}-circle ${t}-inner`]:{position:"relative",lineHeight:1,backgroundColor:"transparent"},[`&${t}-circle ${t}-text`]:{position:"absolute",insetBlockStart:"50%",insetInlineStart:0,width:"100%",margin:0,padding:0,color:e.circleTextColor,fontSize:e.circleTextFontSize,lineHeight:1,whiteSpace:"normal",textAlign:"center",transform:"translateY(-50%)",[r]:{fontSize:e.circleIconFontSize}},[`${t}-circle&-status-exception`]:{[`${t}-text`]:{color:e.colorError}},[`${t}-circle&-status-success`]:{[`${t}-text`]:{color:e.colorSuccess}}},[`${t}-inline-circle`]:{lineHeight:1,[`${t}-inner`]:{verticalAlign:"bottom"}}}})(r),(e=>{let{componentCls:t}=e;return{[t]:{[`${t}-steps`]:{display:"inline-block","&-outer":{display:"flex",flexDirection:"row",alignItems:"center"},"&-item":{flexShrink:0,minWidth:e.progressStepMinWidth,marginInlineEnd:e.progressStepMarginInlineEnd,backgroundColor:e.remainingColor,transition:`all ${e.motionDurationSlow}`,"&-active":{backgroundColor:e.defaultColor}}}}}})(r),(e=>{let{componentCls:t,iconCls:r}=e;return{[t]:{[`${t}-small&-line, ${t}-small&-line ${t}-text ${r}`]:{fontSize:e.fontSizeSM}}}})(r)]},e=>({circleTextColor:e.colorText,defaultColor:e.colorInfo,remainingColor:e.colorFillSecondary,lineBorderRadius:100,circleTextFontSize:"1em",circleIconFontSize:`${e.fontSize/e.fontSizeSM}em`}));var H=function(e,t){var r={};for(var n in e)Object.prototype.hasOwnProperty.call(e,n)&&0>t.indexOf(n)&&(r[n]=e[n]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var o=0,n=Object.getOwnPropertySymbols(e);ot.indexOf(n[o])&&Object.prototype.propertyIsEnumerable.call(e,n[o])&&(r[n[o]]=e[n[o]]);return r};let _=e=>{let{prefixCls:r,direction:n,percent:o,size:i,strokeWidth:l,strokeColor:s,strokeLinecap:c="round",children:u,trailColor:d=null,percentPosition:p,success:f}=e,{align:g,type:m}=p,b=s&&"string"!=typeof s?((e,t)=>{let{from:r=N.presetPrimaryColors.blue,to:n=N.presetPrimaryColors.blue,direction:o="rtl"===t?"to left":"to right"}=e,i=H(e,["from","to","direction"]);if(0!==Object.keys(i).length){let e,t=(e=[],Object.keys(i).forEach(t=>{let r=Number.parseFloat(t.replace(/%/g,""));Number.isNaN(r)||e.push({key:r,value:i[t]})}),(e=e.sort((e,t)=>e.key-t.key)).map(({key:e,value:t})=>`${t} ${e}%`).join(", ")),r=`linear-gradient(${o}, ${t})`;return{background:r,[B]:r}}let l=`linear-gradient(${o}, ${r}, ${n})`;return{background:l,[B]:l}})(s,n):{[B]:s,background:s},h="square"===c||"butt"===c?0:void 0,[v,y]=D(null!=i?i:[-1,l||("small"===i?6:8)],"line",{strokeWidth:l}),$=Object.assign(Object.assign({width:`${I(o)}%`,height:y,borderRadius:h},b),{[F]:I(o)/100}),C=P(e),k={width:`${I(C)}%`,height:y,borderRadius:h,backgroundColor:null==f?void 0:f.strokeColor},x=t.createElement("div",{className:`${r}-inner`,style:{backgroundColor:d||void 0,borderRadius:h}},t.createElement("div",{className:(0,a.default)(`${r}-bg`,`${r}-bg-${m}`),style:$},"inner"===m&&u),void 0!==C&&t.createElement("div",{className:`${r}-success-bg`,style:k})),S="outer"===m&&"start"===g,O="outer"===m&&"end"===g;return"outer"===m&&"center"===g?t.createElement("div",{className:`${r}-layout-bottom`},x,u):t.createElement("div",{className:`${r}-outer`,style:{width:v<0?"100%":v}},S&&u,x,O&&u)},q=e=>{let{size:r,steps:n,rounding:o=Math.round,percent:i=0,strokeWidth:l=8,strokeColor:s,trailColor:c=null,prefixCls:u,children:d}=e,p=o(i/100*n),[f,g]=D(null!=r?r:["small"===r?2:14,l],"step",{steps:n,strokeWidth:l}),m=f/n,b=Array.from({length:n});for(let e=0;et.indexOf(n)&&(r[n]=e[n]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var o=0,n=Object.getOwnPropertySymbols(e);ot.indexOf(n[o])&&Object.prototype.propertyIsEnumerable.call(e,n[o])&&(r[n[o]]=e[n[o]]);return r};let V=["normal","exception","active","success"],K=t.forwardRef((e,u)=>{let d,{prefixCls:p,className:f,rootClassName:g,steps:m,strokeColor:b,percent:h=0,size:v="default",showInfo:y=!0,type:$="line",status:C,format:k,style:x,percentPosition:S={}}=e,O=G(e,["prefixCls","className","rootClassName","steps","strokeColor","percent","size","showInfo","type","status","format","style","percentPosition"]),{align:w="end",type:E="outer"}=S,j=Array.isArray(b)?b[0]:b,N="string"==typeof b||Array.isArray(b)?b:void 0,z=t.useMemo(()=>{if(j){let e="string"==typeof j?j:Object.values(j)[0];return new r.FastColor(e).isLight()}return!1},[b]),A=t.useMemo(()=>{var t,r;let n=P(e);return Number.parseInt(void 0!==n?null==(t=null!=n?n:0)?void 0:t.toString():null==(r=null!=h?h:0)?void 0:r.toString(),10)},[h,e.success,e.successPercent]),M=t.useMemo(()=>!V.includes(C)&&A>=100?"success":C||"normal",[C,A]),{getPrefixCls:T,direction:W,progress:B}=t.useContext(c.ConfigContext),F=T("progress",p),[X,H,K]=L(F),U="line"===$,Q=U&&!m,Y=t.useMemo(()=>{let r;if(!y)return null;let s=P(e),c=k||(e=>`${e}%`),u=U&&z&&"inner"===E;return"inner"===E||k||"exception"!==M&&"success"!==M?r=c(I(h),I(s)):"exception"===M?r=U?t.createElement(i.default,null):t.createElement(l.default,null):"success"===M&&(r=U?t.createElement(n.default,null):t.createElement(o.default,null)),t.createElement("span",{className:(0,a.default)(`${F}-text`,{[`${F}-text-bright`]:u,[`${F}-text-${w}`]:Q,[`${F}-text-${E}`]:Q}),title:"string"==typeof r?r:void 0},r)},[y,h,A,M,$,F,k]);"line"===$?d=m?t.createElement(q,Object.assign({},e,{strokeColor:N,prefixCls:F,steps:"object"==typeof m?m.count:m}),Y):t.createElement(_,Object.assign({},e,{strokeColor:j,prefixCls:F,direction:W,percentPosition:{align:w,type:E}}),Y):("circle"===$||"dashboard"===$)&&(d=t.createElement(R,Object.assign({},e,{strokeColor:j,prefixCls:F,progressStatus:M}),Y));let J=(0,a.default)(F,`${F}-status-${M}`,{[`${F}-${"dashboard"===$&&"circle"||$}`]:"line"!==$,[`${F}-inline-circle`]:"circle"===$&&D(v,"circle")[0]<=20,[`${F}-line`]:Q,[`${F}-line-align-${w}`]:Q,[`${F}-line-position-${E}`]:Q,[`${F}-steps`]:m,[`${F}-show-info`]:y,[`${F}-${v}`]:"string"==typeof v,[`${F}-rtl`]:"rtl"===W},null==B?void 0:B.className,f,g,H,K);return X(t.createElement("div",Object.assign({ref:u,style:Object.assign(Object.assign({},null==B?void 0:B.style),x),className:J,role:"progressbar","aria-valuenow":A,"aria-valuemin":0,"aria-valuemax":100},(0,s.default)(O,["trailColor","strokeWidth","width","gapDegree","gapPosition","strokeLinecap","success","successPercent"])),d))});e.s(["default",0,K],309821)},91874,e=>{"use strict";var t=e.i(931067),r=e.i(209428),n=e.i(211577),o=e.i(392221),i=e.i(703923),l=e.i(343794),a=e.i(914949),s=e.i(271645),c=["prefixCls","className","style","checked","disabled","defaultChecked","type","title","onChange"],u=(0,s.forwardRef)(function(e,u){var d=e.prefixCls,p=void 0===d?"rc-checkbox":d,f=e.className,g=e.style,m=e.checked,b=e.disabled,h=e.defaultChecked,v=e.type,y=void 0===v?"checkbox":v,$=e.title,C=e.onChange,k=(0,i.default)(e,c),x=(0,s.useRef)(null),S=(0,s.useRef)(null),O=(0,a.default)(void 0!==h&&h,{value:m}),w=(0,o.default)(O,2),E=w[0],j=w[1];(0,s.useImperativeHandle)(u,function(){return{focus:function(e){var t;null==(t=x.current)||t.focus(e)},blur:function(){var e;null==(e=x.current)||e.blur()},input:x.current,nativeElement:S.current}});var N=(0,l.default)(p,f,(0,n.default)((0,n.default)({},"".concat(p,"-checked"),E),"".concat(p,"-disabled"),b));return s.createElement("span",{className:N,title:$,style:g,ref:S},s.createElement("input",(0,t.default)({},k,{className:"".concat(p,"-input"),ref:x,onChange:function(t){b||("checked"in e||j(t.target.checked),null==C||C({target:(0,r.default)((0,r.default)({},e),{},{type:y,checked:t.target.checked}),stopPropagation:function(){t.stopPropagation()},preventDefault:function(){t.preventDefault()},nativeEvent:t.nativeEvent}))},disabled:b,checked:!!E,type:y})),s.createElement("span",{className:"".concat(p,"-inner")}))});e.s(["default",0,u])},681216,e=>{"use strict";var t=e.i(271645),r=e.i(963188);function n(e){let n=t.default.useRef(null),o=()=>{r.default.cancel(n.current),n.current=null};return[()=>{o(),n.current=(0,r.default)(()=>{n.current=null})},t=>{n.current&&(t.stopPropagation(),o()),null==e||e(t)}]}e.s(["default",()=>n])},421512,236836,e=>{"use strict";let t=e.i(271645).default.createContext(null);e.s(["default",0,t],421512),e.i(296059);var r=e.i(915654),n=e.i(183293),o=e.i(246422),i=e.i(838378);function l(e,t){return(e=>{let{checkboxCls:t}=e,o=`${t}-wrapper`;return[{[`${t}-group`]:Object.assign(Object.assign({},(0,n.resetComponent)(e)),{display:"inline-flex",flexWrap:"wrap",columnGap:e.marginXS,[`> ${e.antCls}-row`]:{flex:1}}),[o]:Object.assign(Object.assign({},(0,n.resetComponent)(e)),{display:"inline-flex",alignItems:"baseline",cursor:"pointer","&:after":{display:"inline-block",width:0,overflow:"hidden",content:"'\\a0'"},[`& + ${o}`]:{marginInlineStart:0},[`&${o}-in-form-item`]:{'input[type="checkbox"]':{width:14,height:14}}}),[t]:Object.assign(Object.assign({},(0,n.resetComponent)(e)),{position:"relative",whiteSpace:"nowrap",lineHeight:1,cursor:"pointer",borderRadius:e.borderRadiusSM,alignSelf:"center",[`${t}-input`]:{position:"absolute",inset:0,zIndex:1,cursor:"pointer",opacity:0,margin:0,[`&:focus-visible + ${t}-inner`]:(0,n.genFocusOutline)(e)},[`${t}-inner`]:{boxSizing:"border-box",display:"block",width:e.checkboxSize,height:e.checkboxSize,direction:"ltr",backgroundColor:e.colorBgContainer,border:`${(0,r.unit)(e.lineWidth)} ${e.lineType} ${e.colorBorder}`,borderRadius:e.borderRadiusSM,borderCollapse:"separate",transition:`all ${e.motionDurationSlow}`,"&:after":{boxSizing:"border-box",position:"absolute",top:"50%",insetInlineStart:"25%",display:"table",width:e.calc(e.checkboxSize).div(14).mul(5).equal(),height:e.calc(e.checkboxSize).div(14).mul(8).equal(),border:`${(0,r.unit)(e.lineWidthBold)} solid ${e.colorWhite}`,borderTop:0,borderInlineStart:0,transform:"rotate(45deg) scale(0) translate(-50%,-50%)",opacity:0,content:'""',transition:`all ${e.motionDurationFast} ${e.motionEaseInBack}, opacity ${e.motionDurationFast}`}},"& + span":{paddingInlineStart:e.paddingXS,paddingInlineEnd:e.paddingXS}})},{[` + ${o}:not(${o}-disabled), + ${t}:not(${t}-disabled) + `]:{[`&:hover ${t}-inner`]:{borderColor:e.colorPrimary}},[`${o}:not(${o}-disabled)`]:{[`&:hover ${t}-checked:not(${t}-disabled) ${t}-inner`]:{backgroundColor:e.colorPrimaryHover,borderColor:"transparent"},[`&:hover ${t}-checked:not(${t}-disabled):after`]:{borderColor:e.colorPrimaryHover}}},{[`${t}-checked`]:{[`${t}-inner`]:{backgroundColor:e.colorPrimary,borderColor:e.colorPrimary,"&:after":{opacity:1,transform:"rotate(45deg) scale(1) translate(-50%,-50%)",transition:`all ${e.motionDurationMid} ${e.motionEaseOutBack} ${e.motionDurationFast}`}}},[` + ${o}-checked:not(${o}-disabled), + ${t}-checked:not(${t}-disabled) + `]:{[`&:hover ${t}-inner`]:{backgroundColor:e.colorPrimaryHover,borderColor:"transparent"}}},{[t]:{"&-indeterminate":{"&":{[`${t}-inner`]:{backgroundColor:`${e.colorBgContainer}`,borderColor:`${e.colorBorder}`,"&:after":{top:"50%",insetInlineStart:"50%",width:e.calc(e.fontSizeLG).div(2).equal(),height:e.calc(e.fontSizeLG).div(2).equal(),backgroundColor:e.colorPrimary,border:0,transform:"translate(-50%, -50%) scale(1)",opacity:1,content:'""'}},[`&:hover ${t}-inner`]:{backgroundColor:`${e.colorBgContainer}`,borderColor:`${e.colorPrimary}`}}}}},{[`${o}-disabled`]:{cursor:"not-allowed"},[`${t}-disabled`]:{[`&, ${t}-input`]:{cursor:"not-allowed",pointerEvents:"none"},[`${t}-inner`]:{background:e.colorBgContainerDisabled,borderColor:e.colorBorder,"&:after":{borderColor:e.colorTextDisabled}},"&:after":{display:"none"},"& + span":{color:e.colorTextDisabled},[`&${t}-indeterminate ${t}-inner::after`]:{background:e.colorTextDisabled}}}]})((0,i.mergeToken)(t,{checkboxCls:`.${e}`,checkboxSize:t.controlInteractiveSize}))}let a=(0,o.genStyleHooks)("Checkbox",(e,{prefixCls:t})=>[l(t,e)]);e.s(["default",0,a,"getStyle",()=>l],236836)},536916,374276,e=>{"use strict";e.i(247167);var t=e.i(271645),r=e.i(343794),n=e.i(91874),o=e.i(611935),i=e.i(121872),l=e.i(26905),a=e.i(242064),s=e.i(937328),c=e.i(321883),u=e.i(62139),d=e.i(421512),p=e.i(236836),f=e.i(681216),g=function(e,t){var r={};for(var n in e)Object.prototype.hasOwnProperty.call(e,n)&&0>t.indexOf(n)&&(r[n]=e[n]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var o=0,n=Object.getOwnPropertySymbols(e);ot.indexOf(n[o])&&Object.prototype.propertyIsEnumerable.call(e,n[o])&&(r[n[o]]=e[n[o]]);return r};let m=t.forwardRef((e,m)=>{var b;let{prefixCls:h,className:v,rootClassName:y,children:$,indeterminate:C=!1,style:k,onMouseEnter:x,onMouseLeave:S,skipGroup:O=!1,disabled:w}=e,E=g(e,["prefixCls","className","rootClassName","children","indeterminate","style","onMouseEnter","onMouseLeave","skipGroup","disabled"]),{getPrefixCls:j,direction:N,checkbox:I}=t.useContext(a.ConfigContext),P=t.useContext(d.default),{isFormItemInput:D}=t.useContext(u.FormItemInputContext),R=t.useContext(s.default),z=null!=(b=(null==P?void 0:P.disabled)||w)?b:R,A=t.useRef(E.value),M=t.useRef(null),T=(0,o.composeRef)(m,M);t.useEffect(()=>{null==P||P.registerValue(E.value)},[]),t.useEffect(()=>{if(!O)return E.value!==A.current&&(null==P||P.cancelValue(A.current),null==P||P.registerValue(E.value),A.current=E.value),()=>null==P?void 0:P.cancelValue(E.value)},[E.value]),t.useEffect(()=>{var e;(null==(e=M.current)?void 0:e.input)&&(M.current.input.indeterminate=C)},[C]);let W=j("checkbox",h),B=(0,c.default)(W),[F,X,L]=(0,p.default)(W,B),H=Object.assign({},E);P&&!O&&(H.onChange=(...e)=>{E.onChange&&E.onChange.apply(E,e),P.toggleOption&&P.toggleOption({label:$,value:E.value})},H.name=P.name,H.checked=P.value.includes(E.value));let _=(0,r.default)(`${W}-wrapper`,{[`${W}-rtl`]:"rtl"===N,[`${W}-wrapper-checked`]:H.checked,[`${W}-wrapper-disabled`]:z,[`${W}-wrapper-in-form-item`]:D},null==I?void 0:I.className,v,y,L,B,X),q=(0,r.default)({[`${W}-indeterminate`]:C},l.TARGET_CLS,X),[G,V]=(0,f.default)(H.onClick);return F(t.createElement(i.default,{component:"Checkbox",disabled:z},t.createElement("label",{className:_,style:Object.assign(Object.assign({},null==I?void 0:I.style),k),onMouseEnter:x,onMouseLeave:S,onClick:G},t.createElement(n.default,Object.assign({},H,{onClick:V,prefixCls:W,className:q,disabled:z,ref:T})),null!=$&&t.createElement("span",{className:`${W}-label`},$))))});var b=e.i(8211),h=e.i(529681),v=function(e,t){var r={};for(var n in e)Object.prototype.hasOwnProperty.call(e,n)&&0>t.indexOf(n)&&(r[n]=e[n]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var o=0,n=Object.getOwnPropertySymbols(e);ot.indexOf(n[o])&&Object.prototype.propertyIsEnumerable.call(e,n[o])&&(r[n[o]]=e[n[o]]);return r};let y=t.forwardRef((e,n)=>{let{defaultValue:o,children:i,options:l=[],prefixCls:s,className:u,rootClassName:f,style:g,onChange:y}=e,$=v(e,["defaultValue","children","options","prefixCls","className","rootClassName","style","onChange"]),{getPrefixCls:C,direction:k}=t.useContext(a.ConfigContext),[x,S]=t.useState($.value||o||[]),[O,w]=t.useState([]);t.useEffect(()=>{"value"in $&&S($.value||[])},[$.value]);let E=t.useMemo(()=>l.map(e=>"string"==typeof e||"number"==typeof e?{label:e,value:e}:e),[l]),j=e=>{w(t=>t.filter(t=>t!==e))},N=e=>{w(t=>[].concat((0,b.default)(t),[e]))},I=e=>{let t=x.indexOf(e.value),r=(0,b.default)(x);-1===t?r.push(e.value):r.splice(t,1),"value"in $||S(r),null==y||y(r.filter(e=>O.includes(e)).sort((e,t)=>E.findIndex(t=>t.value===e)-E.findIndex(e=>e.value===t)))},P=C("checkbox",s),D=`${P}-group`,R=(0,c.default)(P),[z,A,M]=(0,p.default)(P,R),T=(0,h.default)($,["value","disabled"]),W=l.length?E.map(e=>t.createElement(m,{prefixCls:P,key:e.value.toString(),disabled:"disabled"in e?e.disabled:$.disabled,value:e.value,checked:x.includes(e.value),onChange:e.onChange,className:(0,r.default)(`${D}-item`,e.className),style:e.style,title:e.title,id:e.id,required:e.required},e.label)):i,B=t.useMemo(()=>({toggleOption:I,value:x,disabled:$.disabled,name:$.name,registerValue:N,cancelValue:j}),[I,x,$.disabled,$.name,N,j]),F=(0,r.default)(D,{[`${D}-rtl`]:"rtl"===k},u,f,M,R,A);return z(t.createElement("div",Object.assign({className:F,style:g},T,{ref:n}),t.createElement(d.default.Provider,{value:B},W)))});m.Group=y,m.__ANT_CHECKBOX=!0,e.s(["default",0,m],374276),e.s(["Checkbox",0,m],536916)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/0493aafc4891dd29.js b/litellm/proxy/_experimental/out/_next/static/chunks/0493aafc4891dd29.js deleted file mode 100644 index 90c97f4525a..00000000000 --- a/litellm/proxy/_experimental/out/_next/static/chunks/0493aafc4891dd29.js +++ /dev/null @@ -1 +0,0 @@ -(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,312361,e=>{"use strict";e.i(247167);var t=e.i(271645),n=e.i(343794),r=e.i(242064),i=e.i(517455);e.i(296059);var a=e.i(915654),l=e.i(183293),o=e.i(246422),c=e.i(838378);let s=(0,o.genStyleHooks)("Divider",e=>{let t=(0,c.mergeToken)(e,{dividerHorizontalWithTextGutterMargin:e.margin,sizePaddingEdgeHorizontal:0});return[(e=>{let{componentCls:t,sizePaddingEdgeHorizontal:n,colorSplit:r,lineWidth:i,textPaddingInline:o,orientationMargin:c,verticalMarginInline:s}=e;return{[t]:Object.assign(Object.assign({},(0,l.resetComponent)(e)),{borderBlockStart:`${(0,a.unit)(i)} solid ${r}`,"&-vertical":{position:"relative",top:"-0.06em",display:"inline-block",height:"0.9em",marginInline:s,marginBlock:0,verticalAlign:"middle",borderTop:0,borderInlineStart:`${(0,a.unit)(i)} solid ${r}`},"&-horizontal":{display:"flex",clear:"both",width:"100%",minWidth:"100%",margin:`${(0,a.unit)(e.marginLG)} 0`},[`&-horizontal${t}-with-text`]:{display:"flex",alignItems:"center",margin:`${(0,a.unit)(e.dividerHorizontalWithTextGutterMargin)} 0`,color:e.colorTextHeading,fontWeight:500,fontSize:e.fontSizeLG,whiteSpace:"nowrap",textAlign:"center",borderBlockStart:`0 ${r}`,"&::before, &::after":{position:"relative",width:"50%",borderBlockStart:`${(0,a.unit)(i)} solid transparent`,borderBlockStartColor:"inherit",borderBlockEnd:0,transform:"translateY(50%)",content:"''"}},[`&-horizontal${t}-with-text-start`]:{"&::before":{width:`calc(${c} * 100%)`},"&::after":{width:`calc(100% - ${c} * 100%)`}},[`&-horizontal${t}-with-text-end`]:{"&::before":{width:`calc(100% - ${c} * 100%)`},"&::after":{width:`calc(${c} * 100%)`}},[`${t}-inner-text`]:{display:"inline-block",paddingBlock:0,paddingInline:o},"&-dashed":{background:"none",borderColor:r,borderStyle:"dashed",borderWidth:`${(0,a.unit)(i)} 0 0`},[`&-horizontal${t}-with-text${t}-dashed`]:{"&::before, &::after":{borderStyle:"dashed none none"}},[`&-vertical${t}-dashed`]:{borderInlineStartWidth:i,borderInlineEnd:0,borderBlockStart:0,borderBlockEnd:0},"&-dotted":{background:"none",borderColor:r,borderStyle:"dotted",borderWidth:`${(0,a.unit)(i)} 0 0`},[`&-horizontal${t}-with-text${t}-dotted`]:{"&::before, &::after":{borderStyle:"dotted none none"}},[`&-vertical${t}-dotted`]:{borderInlineStartWidth:i,borderInlineEnd:0,borderBlockStart:0,borderBlockEnd:0},[`&-plain${t}-with-text`]:{color:e.colorText,fontWeight:"normal",fontSize:e.fontSize},[`&-horizontal${t}-with-text-start${t}-no-default-orientation-margin-start`]:{"&::before":{width:0},"&::after":{width:"100%"},[`${t}-inner-text`]:{paddingInlineStart:n}},[`&-horizontal${t}-with-text-end${t}-no-default-orientation-margin-end`]:{"&::before":{width:"100%"},"&::after":{width:0},[`${t}-inner-text`]:{paddingInlineEnd:n}}})}})(t),(e=>{let{componentCls:t}=e;return{[t]:{"&-horizontal":{[`&${t}`]:{"&-sm":{marginBlock:e.marginXS},"&-md":{marginBlock:e.margin}}}}}})(t)]},e=>({textPaddingInline:"1em",orientationMargin:.05,verticalMarginInline:e.marginXS}),{unitless:{orientationMargin:!0}});var d=function(e,t){var n={};for(var r in e)Object.prototype.hasOwnProperty.call(e,r)&&0>t.indexOf(r)&&(n[r]=e[r]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var i=0,r=Object.getOwnPropertySymbols(e);it.indexOf(r[i])&&Object.prototype.propertyIsEnumerable.call(e,r[i])&&(n[r[i]]=e[r[i]]);return n};let u={small:"sm",middle:"md"};e.s(["Divider",0,e=>{let{getPrefixCls:a,direction:l,className:o,style:c}=(0,r.useComponentConfig)("divider"),{prefixCls:g,type:m="horizontal",orientation:p="center",orientationMargin:h,className:f,rootClassName:b,children:$,dashed:y,variant:S="solid",plain:v,style:k,size:C}=e,w=d(e,["prefixCls","type","orientation","orientationMargin","className","rootClassName","children","dashed","variant","plain","style","size"]),I=a("divider",g),[x,O,E]=s(I),z=u[(0,i.default)(C)],j=!!$,N=t.useMemo(()=>"left"===p?"rtl"===l?"end":"start":"right"===p?"rtl"===l?"start":"end":p,[l,p]),P="start"===N&&null!=h,T="end"===N&&null!=h,M=(0,n.default)(I,o,O,E,`${I}-${m}`,{[`${I}-with-text`]:j,[`${I}-with-text-${N}`]:j,[`${I}-dashed`]:!!y,[`${I}-${S}`]:"solid"!==S,[`${I}-plain`]:!!v,[`${I}-rtl`]:"rtl"===l,[`${I}-no-default-orientation-margin-start`]:P,[`${I}-no-default-orientation-margin-end`]:T,[`${I}-${z}`]:!!z},f,b),B=t.useMemo(()=>"number"==typeof h?h:/^\d+$/.test(h)?Number(h):h,[h]);return x(t.createElement("div",Object.assign({className:M,style:Object.assign(Object.assign({},c),k)},w,{role:"separator"}),$&&"vertical"!==m&&t.createElement("span",{className:`${I}-inner-text`,style:{marginInlineStart:P?B:void 0,marginInlineEnd:T?B:void 0}},$)))}],312361)},801312,e=>{"use strict";e.i(247167);var t=e.i(931067),n=e.i(271645);let r={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M724 218.3V141c0-6.7-7.7-10.4-12.9-6.3L260.3 486.8a31.86 31.86 0 000 50.3l450.8 352.1c5.3 4.1 12.9.4 12.9-6.3v-77.3c0-4.9-2.3-9.6-6.1-12.6l-360-281 360-281.1c3.8-3 6.1-7.7 6.1-12.6z"}}]},name:"left",theme:"outlined"};var i=e.i(9583),a=n.forwardRef(function(e,a){return n.createElement(i.default,(0,t.default)({},e,{ref:a,icon:r}))});e.s(["default",0,a],801312)},475254,e=>{"use strict";var t=e.i(271645);let n=e=>{let t=e.replace(/^([A-Z])|[\s-_]+(\w)/g,(e,t,n)=>n?n.toUpperCase():t.toLowerCase());return t.charAt(0).toUpperCase()+t.slice(1)},r=(...e)=>e.filter((e,t,n)=>!!e&&""!==e.trim()&&n.indexOf(e)===t).join(" ").trim();var i={xmlns:"http://www.w3.org/2000/svg",width:24,height:24,viewBox:"0 0 24 24",fill:"none",stroke:"currentColor",strokeWidth:2,strokeLinecap:"round",strokeLinejoin:"round"};let a=(0,t.forwardRef)(({color:e="currentColor",size:n=24,strokeWidth:a=2,absoluteStrokeWidth:l,className:o="",children:c,iconNode:s,...d},u)=>(0,t.createElement)("svg",{ref:u,...i,width:n,height:n,stroke:e,strokeWidth:l?24*Number(a)/Number(n):a,className:r("lucide",o),...!c&&!(e=>{for(let t in e)if(t.startsWith("aria-")||"role"===t||"title"===t)return!0})(d)&&{"aria-hidden":"true"},...d},[...s.map(([e,n])=>(0,t.createElement)(e,n)),...Array.isArray(c)?c:[c]])),l=(e,i)=>{let l=(0,t.forwardRef)(({className:l,...o},c)=>(0,t.createElement)(a,{ref:c,iconNode:i,className:r(`lucide-${n(e).replace(/([a-z0-9])([A-Z])/g,"$1-$2").toLowerCase()}`,`lucide-${e}`,l),...o}));return l.displayName=n(e),l};e.s(["default",()=>l],475254)},262218,e=>{"use strict";e.i(247167);var t=e.i(271645),n=e.i(343794),r=e.i(529681),i=e.i(702779),a=e.i(563113),l=e.i(763731),o=e.i(121872),c=e.i(242064);e.i(296059);var s=e.i(915654);e.i(262370);var d=e.i(135551),u=e.i(183293),g=e.i(246422),m=e.i(838378);let p=e=>{let{lineWidth:t,fontSizeIcon:n,calc:r}=e,i=e.fontSizeSM;return(0,m.mergeToken)(e,{tagFontSize:i,tagLineHeight:(0,s.unit)(r(e.lineHeightSM).mul(i).equal()),tagIconSize:r(n).sub(r(t).mul(2)).equal(),tagPaddingHorizontal:8,tagBorderlessBg:e.defaultBg})},h=e=>({defaultBg:new d.FastColor(e.colorFillQuaternary).onBackground(e.colorBgContainer).toHexString(),defaultColor:e.colorText}),f=(0,g.genStyleHooks)("Tag",e=>(e=>{let{paddingXXS:t,lineWidth:n,tagPaddingHorizontal:r,componentCls:i,calc:a}=e,l=a(r).sub(n).equal(),o=a(t).sub(n).equal();return{[i]:Object.assign(Object.assign({},(0,u.resetComponent)(e)),{display:"inline-block",height:"auto",marginInlineEnd:e.marginXS,paddingInline:l,fontSize:e.tagFontSize,lineHeight:e.tagLineHeight,whiteSpace:"nowrap",background:e.defaultBg,border:`${(0,s.unit)(e.lineWidth)} ${e.lineType} ${e.colorBorder}`,borderRadius:e.borderRadiusSM,opacity:1,transition:`all ${e.motionDurationMid}`,textAlign:"start",position:"relative",[`&${i}-rtl`]:{direction:"rtl"},"&, a, a:hover":{color:e.defaultColor},[`${i}-close-icon`]:{marginInlineStart:o,fontSize:e.tagIconSize,color:e.colorIcon,cursor:"pointer",transition:`all ${e.motionDurationMid}`,"&:hover":{color:e.colorTextHeading}},[`&${i}-has-color`]:{borderColor:"transparent",[`&, a, a:hover, ${e.iconCls}-close, ${e.iconCls}-close:hover`]:{color:e.colorTextLightSolid}},"&-checkable":{backgroundColor:"transparent",borderColor:"transparent",cursor:"pointer",[`&:not(${i}-checkable-checked):hover`]:{color:e.colorPrimary,backgroundColor:e.colorFillSecondary},"&:active, &-checked":{color:e.colorTextLightSolid},"&-checked":{backgroundColor:e.colorPrimary,"&:hover":{backgroundColor:e.colorPrimaryHover}},"&:active":{backgroundColor:e.colorPrimaryActive}},"&-hidden":{display:"none"},[`> ${e.iconCls} + span, > span + ${e.iconCls}`]:{marginInlineStart:l}}),[`${i}-borderless`]:{borderColor:"transparent",background:e.tagBorderlessBg}}})(p(e)),h);var b=function(e,t){var n={};for(var r in e)Object.prototype.hasOwnProperty.call(e,r)&&0>t.indexOf(r)&&(n[r]=e[r]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var i=0,r=Object.getOwnPropertySymbols(e);it.indexOf(r[i])&&Object.prototype.propertyIsEnumerable.call(e,r[i])&&(n[r[i]]=e[r[i]]);return n};let $=t.forwardRef((e,r)=>{let{prefixCls:i,style:a,className:l,checked:o,children:s,icon:d,onChange:u,onClick:g}=e,m=b(e,["prefixCls","style","className","checked","children","icon","onChange","onClick"]),{getPrefixCls:p,tag:h}=t.useContext(c.ConfigContext),$=p("tag",i),[y,S,v]=f($),k=(0,n.default)($,`${$}-checkable`,{[`${$}-checkable-checked`]:o},null==h?void 0:h.className,l,S,v);return y(t.createElement("span",Object.assign({},m,{ref:r,style:Object.assign(Object.assign({},a),null==h?void 0:h.style),className:k,onClick:e=>{null==u||u(!o),null==g||g(e)}}),d,t.createElement("span",null,s)))});var y=e.i(403541);let S=(0,g.genSubStyleComponent)(["Tag","preset"],e=>{let t;return t=p(e),(0,y.genPresetColor)(t,(e,{textColor:n,lightBorderColor:r,lightColor:i,darkColor:a})=>({[`${t.componentCls}${t.componentCls}-${e}`]:{color:n,background:i,borderColor:r,"&-inverse":{color:t.colorTextLightSolid,background:a,borderColor:a},[`&${t.componentCls}-borderless`]:{borderColor:"transparent"}}}))},h),v=(e,t,n)=>{let r="string"!=typeof n?n:n.charAt(0).toUpperCase()+n.slice(1);return{[`${e.componentCls}${e.componentCls}-${t}`]:{color:e[`color${n}`],background:e[`color${r}Bg`],borderColor:e[`color${r}Border`],[`&${e.componentCls}-borderless`]:{borderColor:"transparent"}}}},k=(0,g.genSubStyleComponent)(["Tag","status"],e=>{let t=p(e);return[v(t,"success","Success"),v(t,"processing","Info"),v(t,"error","Error"),v(t,"warning","Warning")]},h);var C=function(e,t){var n={};for(var r in e)Object.prototype.hasOwnProperty.call(e,r)&&0>t.indexOf(r)&&(n[r]=e[r]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var i=0,r=Object.getOwnPropertySymbols(e);it.indexOf(r[i])&&Object.prototype.propertyIsEnumerable.call(e,r[i])&&(n[r[i]]=e[r[i]]);return n};let w=t.forwardRef((e,s)=>{let{prefixCls:d,className:u,rootClassName:g,style:m,children:p,icon:h,color:b,onClose:$,bordered:y=!0,visible:v}=e,w=C(e,["prefixCls","className","rootClassName","style","children","icon","color","onClose","bordered","visible"]),{getPrefixCls:I,direction:x,tag:O}=t.useContext(c.ConfigContext),[E,z]=t.useState(!0),j=(0,r.default)(w,["closeIcon","closable"]);t.useEffect(()=>{void 0!==v&&z(v)},[v]);let N=(0,i.isPresetColor)(b),P=(0,i.isPresetStatusColor)(b),T=N||P,M=Object.assign(Object.assign({backgroundColor:b&&!T?b:void 0},null==O?void 0:O.style),m),B=I("tag",d),[H,L,R]=f(B),q=(0,n.default)(B,null==O?void 0:O.className,{[`${B}-${b}`]:T,[`${B}-has-color`]:b&&!T,[`${B}-hidden`]:!E,[`${B}-rtl`]:"rtl"===x,[`${B}-borderless`]:!y},u,g,L,R),G=e=>{e.stopPropagation(),null==$||$(e),e.defaultPrevented||z(!1)},[,A]=(0,a.useClosable)((0,a.pickClosable)(e),(0,a.pickClosable)(O),{closable:!1,closeIconRender:e=>{let r=t.createElement("span",{className:`${B}-close-icon`,onClick:G},e);return(0,l.replaceElement)(e,r,e=>({onClick:t=>{var n;null==(n=null==e?void 0:e.onClick)||n.call(e,t),G(t)},className:(0,n.default)(null==e?void 0:e.className,`${B}-close-icon`)}))}}),W="function"==typeof w.onClick||p&&"a"===p.type,D=h||null,X=D?t.createElement(t.Fragment,null,D,p&&t.createElement("span",null,p)):p,F=t.createElement("span",Object.assign({},j,{ref:s,className:q,style:M}),X,A,N&&t.createElement(S,{key:"preset",prefixCls:B}),P&&t.createElement(k,{key:"status",prefixCls:B}));return H(W?t.createElement(o.default,{component:"Tag"},F):F)});w.CheckableTag=$,e.s(["Tag",0,w],262218)},653496,e=>{"use strict";var t=e.i(721369);e.s(["Tabs",()=>t.default])},790848,e=>{"use strict";e.i(247167);var t=e.i(271645),n=e.i(739295),r=e.i(343794),i=e.i(931067),a=e.i(211577),l=e.i(392221),o=e.i(703923),c=e.i(914949),s=e.i(404948),d=["prefixCls","className","checked","defaultChecked","disabled","loadingIcon","checkedChildren","unCheckedChildren","onClick","onChange","onKeyDown"],u=t.forwardRef(function(e,n){var u,g=e.prefixCls,m=void 0===g?"rc-switch":g,p=e.className,h=e.checked,f=e.defaultChecked,b=e.disabled,$=e.loadingIcon,y=e.checkedChildren,S=e.unCheckedChildren,v=e.onClick,k=e.onChange,C=e.onKeyDown,w=(0,o.default)(e,d),I=(0,c.default)(!1,{value:h,defaultValue:f}),x=(0,l.default)(I,2),O=x[0],E=x[1];function z(e,t){var n=O;return b||(E(n=e),null==k||k(n,t)),n}var j=(0,r.default)(m,p,(u={},(0,a.default)(u,"".concat(m,"-checked"),O),(0,a.default)(u,"".concat(m,"-disabled"),b),u));return t.createElement("button",(0,i.default)({},w,{type:"button",role:"switch","aria-checked":O,disabled:b,className:j,ref:n,onKeyDown:function(e){e.which===s.default.LEFT?z(!1,e):e.which===s.default.RIGHT&&z(!0,e),null==C||C(e)},onClick:function(e){var t=z(!O,e);null==v||v(t,e)}}),$,t.createElement("span",{className:"".concat(m,"-inner")},t.createElement("span",{className:"".concat(m,"-inner-checked")},y),t.createElement("span",{className:"".concat(m,"-inner-unchecked")},S)))});u.displayName="Switch";var g=e.i(121872),m=e.i(242064),p=e.i(937328),h=e.i(517455);e.i(296059);var f=e.i(915654);e.i(262370);var b=e.i(135551),$=e.i(183293),y=e.i(246422),S=e.i(838378);let v=(0,y.genStyleHooks)("Switch",e=>{let t=(0,S.mergeToken)(e,{switchDuration:e.motionDurationMid,switchColor:e.colorPrimary,switchDisabledOpacity:e.opacityLoading,switchLoadingIconSize:e.calc(e.fontSizeIcon).mul(.75).equal(),switchLoadingIconColor:`rgba(0, 0, 0, ${e.opacityLoading})`,switchHandleActiveInset:"-30%"});return[(e=>{let{componentCls:t,trackHeight:n,trackMinWidth:r}=e;return{[t]:Object.assign(Object.assign(Object.assign(Object.assign({},(0,$.resetComponent)(e)),{position:"relative",display:"inline-block",boxSizing:"border-box",minWidth:r,height:n,lineHeight:(0,f.unit)(n),verticalAlign:"middle",background:e.colorTextQuaternary,border:"0",borderRadius:100,cursor:"pointer",transition:`all ${e.motionDurationMid}`,userSelect:"none",[`&:hover:not(${t}-disabled)`]:{background:e.colorTextTertiary}}),(0,$.genFocusStyle)(e)),{[`&${t}-checked`]:{background:e.switchColor,[`&:hover:not(${t}-disabled)`]:{background:e.colorPrimaryHover}},[`&${t}-loading, &${t}-disabled`]:{cursor:"not-allowed",opacity:e.switchDisabledOpacity,"*":{boxShadow:"none",cursor:"not-allowed"}},[`&${t}-rtl`]:{direction:"rtl"}})}})(t),(e=>{let{componentCls:t,trackHeight:n,trackPadding:r,innerMinMargin:i,innerMaxMargin:a,handleSize:l,calc:o}=e,c=`${t}-inner`,s=(0,f.unit)(o(l).add(o(r).mul(2)).equal()),d=(0,f.unit)(o(a).mul(2).equal());return{[t]:{[c]:{display:"block",overflow:"hidden",borderRadius:100,height:"100%",paddingInlineStart:a,paddingInlineEnd:i,transition:`padding-inline-start ${e.switchDuration} ease-in-out, padding-inline-end ${e.switchDuration} ease-in-out`,[`${c}-checked, ${c}-unchecked`]:{display:"block",color:e.colorTextLightSolid,fontSize:e.fontSizeSM,transition:`margin-inline-start ${e.switchDuration} ease-in-out, margin-inline-end ${e.switchDuration} ease-in-out`,pointerEvents:"none",minHeight:n},[`${c}-checked`]:{marginInlineStart:`calc(-100% + ${s} - ${d})`,marginInlineEnd:`calc(100% - ${s} + ${d})`},[`${c}-unchecked`]:{marginTop:o(n).mul(-1).equal(),marginInlineStart:0,marginInlineEnd:0}},[`&${t}-checked ${c}`]:{paddingInlineStart:i,paddingInlineEnd:a,[`${c}-checked`]:{marginInlineStart:0,marginInlineEnd:0},[`${c}-unchecked`]:{marginInlineStart:`calc(100% - ${s} + ${d})`,marginInlineEnd:`calc(-100% + ${s} - ${d})`}},[`&:not(${t}-disabled):active`]:{[`&:not(${t}-checked) ${c}`]:{[`${c}-unchecked`]:{marginInlineStart:o(r).mul(2).equal(),marginInlineEnd:o(r).mul(-1).mul(2).equal()}},[`&${t}-checked ${c}`]:{[`${c}-checked`]:{marginInlineStart:o(r).mul(-1).mul(2).equal(),marginInlineEnd:o(r).mul(2).equal()}}}}}})(t),(e=>{let{componentCls:t,trackPadding:n,handleBg:r,handleShadow:i,handleSize:a,calc:l}=e,o=`${t}-handle`;return{[t]:{[o]:{position:"absolute",top:n,insetInlineStart:n,width:a,height:a,transition:`all ${e.switchDuration} ease-in-out`,"&::before":{position:"absolute",top:0,insetInlineEnd:0,bottom:0,insetInlineStart:0,backgroundColor:r,borderRadius:l(a).div(2).equal(),boxShadow:i,transition:`all ${e.switchDuration} ease-in-out`,content:'""'}},[`&${t}-checked ${o}`]:{insetInlineStart:`calc(100% - ${(0,f.unit)(l(a).add(n).equal())})`},[`&:not(${t}-disabled):active`]:{[`${o}::before`]:{insetInlineEnd:e.switchHandleActiveInset,insetInlineStart:0},[`&${t}-checked ${o}::before`]:{insetInlineEnd:0,insetInlineStart:e.switchHandleActiveInset}}}}})(t),(e=>{let{componentCls:t,handleSize:n,calc:r}=e;return{[t]:{[`${t}-loading-icon${e.iconCls}`]:{position:"relative",top:r(r(n).sub(e.fontSize)).div(2).equal(),color:e.switchLoadingIconColor,verticalAlign:"top"},[`&${t}-checked ${t}-loading-icon`]:{color:e.switchColor}}}})(t),(e=>{let{componentCls:t,trackHeightSM:n,trackPadding:r,trackMinWidthSM:i,innerMinMarginSM:a,innerMaxMarginSM:l,handleSizeSM:o,calc:c}=e,s=`${t}-inner`,d=(0,f.unit)(c(o).add(c(r).mul(2)).equal()),u=(0,f.unit)(c(l).mul(2).equal());return{[t]:{[`&${t}-small`]:{minWidth:i,height:n,lineHeight:(0,f.unit)(n),[`${t}-inner`]:{paddingInlineStart:l,paddingInlineEnd:a,[`${s}-checked, ${s}-unchecked`]:{minHeight:n},[`${s}-checked`]:{marginInlineStart:`calc(-100% + ${d} - ${u})`,marginInlineEnd:`calc(100% - ${d} + ${u})`},[`${s}-unchecked`]:{marginTop:c(n).mul(-1).equal(),marginInlineStart:0,marginInlineEnd:0}},[`${t}-handle`]:{width:o,height:o},[`${t}-loading-icon`]:{top:c(c(o).sub(e.switchLoadingIconSize)).div(2).equal(),fontSize:e.switchLoadingIconSize},[`&${t}-checked`]:{[`${t}-inner`]:{paddingInlineStart:a,paddingInlineEnd:l,[`${s}-checked`]:{marginInlineStart:0,marginInlineEnd:0},[`${s}-unchecked`]:{marginInlineStart:`calc(100% - ${d} + ${u})`,marginInlineEnd:`calc(-100% + ${d} - ${u})`}},[`${t}-handle`]:{insetInlineStart:`calc(100% - ${(0,f.unit)(c(o).add(r).equal())})`}},[`&:not(${t}-disabled):active`]:{[`&:not(${t}-checked) ${s}`]:{[`${s}-unchecked`]:{marginInlineStart:c(e.marginXXS).div(2).equal(),marginInlineEnd:c(e.marginXXS).mul(-1).div(2).equal()}},[`&${t}-checked ${s}`]:{[`${s}-checked`]:{marginInlineStart:c(e.marginXXS).mul(-1).div(2).equal(),marginInlineEnd:c(e.marginXXS).div(2).equal()}}}}}}})(t)]},e=>{let{fontSize:t,lineHeight:n,controlHeight:r,colorWhite:i}=e,a=t*n,l=r/2,o=a-4,c=l-4;return{trackHeight:a,trackHeightSM:l,trackMinWidth:2*o+8,trackMinWidthSM:2*c+4,trackPadding:2,handleBg:i,handleSize:o,handleSizeSM:c,handleShadow:`0 2px 4px 0 ${new b.FastColor("#00230b").setA(.2).toRgbString()}`,innerMinMargin:o/2,innerMaxMargin:o+2+4,innerMinMarginSM:c/2,innerMaxMarginSM:c+2+4}});var k=function(e,t){var n={};for(var r in e)Object.prototype.hasOwnProperty.call(e,r)&&0>t.indexOf(r)&&(n[r]=e[r]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var i=0,r=Object.getOwnPropertySymbols(e);it.indexOf(r[i])&&Object.prototype.propertyIsEnumerable.call(e,r[i])&&(n[r[i]]=e[r[i]]);return n};let C=t.forwardRef((e,i)=>{let{prefixCls:a,size:l,disabled:o,loading:s,className:d,rootClassName:f,style:b,checked:$,value:y,defaultChecked:S,defaultValue:C,onChange:w}=e,I=k(e,["prefixCls","size","disabled","loading","className","rootClassName","style","checked","value","defaultChecked","defaultValue","onChange"]),[x,O]=(0,c.default)(!1,{value:null!=$?$:y,defaultValue:null!=S?S:C}),{getPrefixCls:E,direction:z,switch:j}=t.useContext(m.ConfigContext),N=t.useContext(p.default),P=(null!=o?o:N)||s,T=E("switch",a),M=t.createElement("div",{className:`${T}-handle`},s&&t.createElement(n.default,{className:`${T}-loading-icon`})),[B,H,L]=v(T),R=(0,h.default)(l),q=(0,r.default)(null==j?void 0:j.className,{[`${T}-small`]:"small"===R,[`${T}-loading`]:s,[`${T}-rtl`]:"rtl"===z},d,f,H,L),G=Object.assign(Object.assign({},null==j?void 0:j.style),b);return B(t.createElement(g.default,{component:"Switch",disabled:P},t.createElement(u,Object.assign({},I,{checked:x,onChange:(...e)=>{O(e[0]),null==w||w.apply(void 0,e)},prefixCls:T,className:q,style:G,disabled:P,ref:i,loadingIcon:M}))))});C.__ANT_SWITCH=!0,e.s(["Switch",0,C],790848)},38243,908286,e=>{"use strict";e.i(247167);var t=e.i(271645),n=e.i(343794),r=e.i(876556);function i(e){return["small","middle","large"].includes(e)}function a(e){return!!e&&"number"==typeof e&&!Number.isNaN(e)}e.s(["isPresetSize",()=>i,"isValidGapNumber",()=>a],908286);var l=e.i(242064),o=e.i(249616),c=e.i(372409),s=e.i(246422);let d=(0,s.genStyleHooks)(["Space","Addon"],e=>[(e=>{let{componentCls:t,borderRadius:n,paddingSM:r,colorBorder:i,paddingXS:a,fontSizeLG:l,fontSizeSM:o,borderRadiusLG:s,borderRadiusSM:d,colorBgContainerDisabled:u,lineWidth:g}=e;return{[t]:[{display:"inline-flex",alignItems:"center",gap:0,paddingInline:r,margin:0,background:u,borderWidth:g,borderStyle:"solid",borderColor:i,borderRadius:n,"&-large":{fontSize:l,borderRadius:s},"&-small":{paddingInline:a,borderRadius:d,fontSize:o},"&-compact-last-item":{borderEndStartRadius:0,borderStartStartRadius:0},"&-compact-first-item":{borderEndEndRadius:0,borderStartEndRadius:0},"&-compact-item:not(:first-child):not(:last-child)":{borderRadius:0},"&-compact-item:not(:last-child)":{borderInlineEndWidth:0}},(0,c.genCompactItemStyle)(e,{focus:!1})]}})(e)]);var u=function(e,t){var n={};for(var r in e)Object.prototype.hasOwnProperty.call(e,r)&&0>t.indexOf(r)&&(n[r]=e[r]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var i=0,r=Object.getOwnPropertySymbols(e);it.indexOf(r[i])&&Object.prototype.propertyIsEnumerable.call(e,r[i])&&(n[r[i]]=e[r[i]]);return n};let g=t.default.forwardRef((e,r)=>{let{className:i,children:a,style:c,prefixCls:s}=e,g=u(e,["className","children","style","prefixCls"]),{getPrefixCls:m,direction:p}=t.default.useContext(l.ConfigContext),h=m("space-addon",s),[f,b,$]=d(h),{compactItemClassnames:y,compactSize:S}=(0,o.useCompactItemContext)(h,p),v=(0,n.default)(h,b,y,$,{[`${h}-${S}`]:S},i);return f(t.default.createElement("div",Object.assign({ref:r,className:v,style:c},g),a))}),m=t.default.createContext({latestIndex:0}),p=m.Provider,h=({className:e,index:n,children:r,split:i,style:a})=>{let{latestIndex:l}=t.useContext(m);return null==r?null:t.createElement(t.Fragment,null,t.createElement("div",{className:e,style:a},r),n{let t=(0,f.mergeToken)(e,{spaceGapSmallSize:e.paddingXS,spaceGapMiddleSize:e.padding,spaceGapLargeSize:e.paddingLG});return[(e=>{let{componentCls:t,antCls:n}=e;return{[t]:{display:"inline-flex","&-rtl":{direction:"rtl"},"&-vertical":{flexDirection:"column"},"&-align":{flexDirection:"column","&-center":{alignItems:"center"},"&-start":{alignItems:"flex-start"},"&-end":{alignItems:"flex-end"},"&-baseline":{alignItems:"baseline"}},[`${t}-item:empty`]:{display:"none"},[`${t}-item > ${n}-badge-not-a-wrapper:only-child`]:{display:"block"}}}})(t),(e=>{let{componentCls:t}=e;return{[t]:{"&-gap-row-small":{rowGap:e.spaceGapSmallSize},"&-gap-row-middle":{rowGap:e.spaceGapMiddleSize},"&-gap-row-large":{rowGap:e.spaceGapLargeSize},"&-gap-col-small":{columnGap:e.spaceGapSmallSize},"&-gap-col-middle":{columnGap:e.spaceGapMiddleSize},"&-gap-col-large":{columnGap:e.spaceGapLargeSize}}}})(t)]},()=>({}),{resetStyle:!1});var $=function(e,t){var n={};for(var r in e)Object.prototype.hasOwnProperty.call(e,r)&&0>t.indexOf(r)&&(n[r]=e[r]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var i=0,r=Object.getOwnPropertySymbols(e);it.indexOf(r[i])&&Object.prototype.propertyIsEnumerable.call(e,r[i])&&(n[r[i]]=e[r[i]]);return n};let y=t.forwardRef((e,o)=>{var c;let{getPrefixCls:s,direction:d,size:u,className:g,style:m,classNames:f,styles:y}=(0,l.useComponentConfig)("space"),{size:S=null!=u?u:"small",align:v,className:k,rootClassName:C,children:w,direction:I="horizontal",prefixCls:x,split:O,style:E,wrap:z=!1,classNames:j,styles:N}=e,P=$(e,["size","align","className","rootClassName","children","direction","prefixCls","split","style","wrap","classNames","styles"]),[T,M]=Array.isArray(S)?S:[S,S],B=i(M),H=i(T),L=a(M),R=a(T),q=(0,r.default)(w,{keepEmpty:!0}),G=void 0===v&&"horizontal"===I?"center":v,A=s("space",x),[W,D,X]=b(A),F=(0,n.default)(A,g,D,`${A}-${I}`,{[`${A}-rtl`]:"rtl"===d,[`${A}-align-${G}`]:G,[`${A}-gap-row-${M}`]:B,[`${A}-gap-col-${T}`]:H},k,C,X),K=(0,n.default)(`${A}-item`,null!=(c=null==j?void 0:j.item)?c:f.item),U=Object.assign(Object.assign({},y.item),null==N?void 0:N.item),V=q.map((e,n)=>{let r=(null==e?void 0:e.key)||`${K}-${n}`;return t.createElement(h,{className:K,key:r,index:n,split:O,style:U},e)}),Q=t.useMemo(()=>({latestIndex:q.reduce((e,t,n)=>null!=t?n:e,0)}),[q]);if(0===q.length)return null;let _={};return z&&(_.flexWrap="wrap"),!H&&R&&(_.columnGap=T),!B&&L&&(_.rowGap=M),W(t.createElement("div",Object.assign({ref:o,className:F,style:Object.assign(Object.assign(Object.assign({},_),m),E)},P),t.createElement(p,{value:Q},V)))});y.Compact=o.default,y.Addon=g,e.s(["default",0,y],38243)},770914,e=>{"use strict";var t=e.i(38243);e.s(["Space",()=>t.default])},292639,e=>{"use strict";var t=e.i(764205),n=e.i(266027);let r=(0,e.i(243652).createQueryKeys)("uiSettings");e.s(["useUISettings",0,()=>(0,n.useQuery)({queryKey:r.list({}),queryFn:async()=>await (0,t.getUiSettings)(),staleTime:36e5,gcTime:36e5})])},250980,e=>{"use strict";var t=e.i(271645);let n=t.forwardRef(function(e,n){return t.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",fill:"none",viewBox:"0 0 24 24",strokeWidth:2,stroke:"currentColor","aria-hidden":"true",ref:n},e),t.createElement("path",{strokeLinecap:"round",strokeLinejoin:"round",d:"M12 9v3m0 0v3m0-3h3m-3 0H9m12 0a9 9 0 11-18 0 9 9 0 0118 0z"}))});e.s(["PlusCircleIcon",0,n],250980)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/0549bc9afa7d4888.js b/litellm/proxy/_experimental/out/_next/static/chunks/0549bc9afa7d4888.js deleted file mode 100644 index feba90545f9..00000000000 --- a/litellm/proxy/_experimental/out/_next/static/chunks/0549bc9afa7d4888.js +++ /dev/null @@ -1,41 +0,0 @@ -(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,464571,e=>{"use strict";var t=e.i(920228);e.s(["Button",()=>t.default])},486794,(e,t,n)=>{t.exports=function(){var e=document.getSelection();if(!e.rangeCount)return function(){};for(var t=document.activeElement,n=[],l=0;l{"use strict";var l=e.r(486794),r={"text/plain":"Text","text/html":"Url",default:"Text"};t.exports=function(e,t){var n,o,a,i,c,s,u,d,p=!1;t||(t={}),a=t.debug||!1;try{if(c=l(),s=document.createRange(),u=document.getSelection(),(d=document.createElement("span")).textContent=e,d.ariaHidden="true",d.style.all="unset",d.style.position="fixed",d.style.top=0,d.style.clip="rect(0, 0, 0, 0)",d.style.whiteSpace="pre",d.style.webkitUserSelect="text",d.style.MozUserSelect="text",d.style.msUserSelect="text",d.style.userSelect="text",d.addEventListener("copy",function(n){if(n.stopPropagation(),t.format)if(n.preventDefault(),void 0===n.clipboardData){a&&console.warn("unable to use e.clipboardData"),a&&console.warn("trying IE specific stuff"),window.clipboardData.clearData();var l=r[t.format]||r.default;window.clipboardData.setData(l,e)}else n.clipboardData.clearData(),n.clipboardData.setData(t.format,e);t.onCopy&&(n.preventDefault(),t.onCopy(n.clipboardData))}),document.body.appendChild(d),s.selectNodeContents(d),u.addRange(s),!document.execCommand("copy"))throw Error("copy command was unsuccessful");p=!0}catch(l){a&&console.error("unable to copy using execCommand: ",l),a&&console.warn("trying IE specific stuff");try{window.clipboardData.setData(t.format||"text",e),t.onCopy&&t.onCopy(window.clipboardData),p=!0}catch(l){a&&console.error("unable to copy using clipboardData: ",l),a&&console.error("falling back to prompt"),n="message"in t?t.message:"Copy to clipboard: #{key}, Enter",o=(/mac os x/i.test(navigator.userAgent)?"⌘":"Ctrl")+"+C",i=n.replace(/#{\s*key\s*}/g,o),window.prompt(i,e)}}finally{u&&("function"==typeof u.removeRange?u.removeRange(s):u.removeAllRanges()),d&&document.body.removeChild(d),c()}return p}},898586,401361,335771,e=>{"use strict";e.i(247167);var t=e.i(271645),n=e.i(8211),l=e.i(931067);let r={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M257.7 752c2 0 4-.2 6-.5L431.9 722c2-.4 3.9-1.3 5.3-2.8l423.9-423.9a9.96 9.96 0 000-14.1L694.9 114.9c-1.9-1.9-4.4-2.9-7.1-2.9s-5.2 1-7.1 2.9L256.8 538.8c-1.5 1.5-2.4 3.3-2.8 5.3l-29.5 168.2a33.5 33.5 0 009.4 29.8c6.6 6.4 14.9 9.9 23.8 9.9zm67.4-174.4L687.8 215l73.3 73.3-362.7 362.6-88.9 15.7 15.6-89zM880 836H144c-17.7 0-32 14.3-32 32v36c0 4.4 3.6 8 8 8h784c4.4 0 8-3.6 8-8v-36c0-17.7-14.3-32-32-32z"}}]},name:"edit",theme:"outlined"};var o=e.i(9583),a=t.forwardRef(function(e,n){return t.createElement(o.default,(0,l.default)({},e,{ref:n,icon:r}))});e.s(["default",0,a],401361);var i=e.i(343794),c=e.i(430073),s=e.i(876556),u=e.i(174428),d=e.i(914949),p=e.i(529681),f=e.i(611935),m=e.i(735049),g=e.i(242064),b=e.i(929447),y=e.i(491816);let v={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M864 170h-60c-4.4 0-8 3.6-8 8v518H310v-73c0-6.7-7.8-10.5-13-6.3l-141.9 112a8 8 0 000 12.6l141.9 112c5.3 4.2 13 .4 13-6.3v-75h498c35.3 0 64-28.7 64-64V178c0-4.4-3.6-8-8-8z"}}]},name:"enter",theme:"outlined"};var h=t.forwardRef(function(e,n){return t.createElement(o.default,(0,l.default)({},e,{ref:n,icon:v}))}),x=e.i(404948),O=e.i(763731),E=e.i(635432),S=e.i(183293),w=e.i(246422);e.i(765846);var j=e.i(896091);let C=(0,w.genStyleHooks)("Typography",e=>{let t,{componentCls:n,titleMarginTop:l}=e;return{[n]:Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign(Object.assign({color:e.colorText,wordBreak:"break-word",lineHeight:e.lineHeight,[`&${n}-secondary`]:{color:e.colorTextDescription},[`&${n}-success`]:{color:e.colorSuccessText},[`&${n}-warning`]:{color:e.colorWarningText},[`&${n}-danger`]:{color:e.colorErrorText,"a&:active, a&:focus":{color:e.colorErrorTextActive},"a&:hover":{color:e.colorErrorTextHover}},[`&${n}-disabled`]:{color:e.colorTextDisabled,cursor:"not-allowed",userSelect:"none"},[` - div&, - p - `]:{marginBottom:"1em"}},(t={},[1,2,3,4,5].forEach(n=>{t[` - h${n}&, - div&-h${n}, - div&-h${n} > textarea, - h${n} - `]=((e,t,n,l)=>{let{titleMarginBottom:r,fontWeightStrong:o}=l;return{marginBottom:r,color:n,fontWeight:o,fontSize:e,lineHeight:t}})(e[`fontSizeHeading${n}`],e[`lineHeightHeading${n}`],e.colorTextHeading,e)}),t)),{[` - & + h1${n}, - & + h2${n}, - & + h3${n}, - & + h4${n}, - & + h5${n} - `]:{marginTop:l},[` - div, - ul, - li, - p, - h1, - h2, - h3, - h4, - h5`]:{[` - + h1, - + h2, - + h3, - + h4, - + h5 - `]:{marginTop:l}}}),{code:{margin:"0 0.2em",paddingInline:"0.4em",paddingBlock:"0.2em 0.1em",fontSize:"85%",fontFamily:e.fontFamilyCode,background:"rgba(150, 150, 150, 0.1)",border:"1px solid rgba(100, 100, 100, 0.2)",borderRadius:3},kbd:{margin:"0 0.2em",paddingInline:"0.4em",paddingBlock:"0.15em 0.1em",fontSize:"90%",fontFamily:e.fontFamilyCode,background:"rgba(150, 150, 150, 0.06)",border:"1px solid rgba(100, 100, 100, 0.2)",borderBottomWidth:2,borderRadius:3},mark:{padding:0,backgroundColor:j.gold[2]},"u, ins":{textDecoration:"underline",textDecorationSkipInk:"auto"},"s, del":{textDecoration:"line-through"},strong:{fontWeight:e.fontWeightStrong},"ul, ol":{marginInline:0,marginBlock:"0 1em",padding:0,li:{marginInline:"20px 0",marginBlock:0,paddingInline:"4px 0",paddingBlock:0}},ul:{listStyleType:"circle",ul:{listStyleType:"disc"}},ol:{listStyleType:"decimal"},"pre, blockquote":{margin:"1em 0"},pre:{padding:"0.4em 0.6em",whiteSpace:"pre-wrap",wordWrap:"break-word",background:"rgba(150, 150, 150, 0.1)",border:"1px solid rgba(100, 100, 100, 0.2)",borderRadius:3,fontFamily:e.fontFamilyCode,code:{display:"inline",margin:0,padding:0,fontSize:"inherit",fontFamily:"inherit",background:"transparent",border:0}},blockquote:{paddingInline:"0.6em 0",paddingBlock:0,borderInlineStart:"4px solid rgba(100, 100, 100, 0.2)",opacity:.85}}),(e=>{let{componentCls:t}=e;return{"a&, a":Object.assign(Object.assign({},(0,S.operationUnit)(e)),{userSelect:"text",[`&[disabled], &${t}-disabled`]:{color:e.colorTextDisabled,cursor:"not-allowed","&:active, &:hover":{color:e.colorTextDisabled},"&:active":{pointerEvents:"none"}}})}})(e)),{[` - ${n}-expand, - ${n}-collapse, - ${n}-edit, - ${n}-copy - `]:Object.assign(Object.assign({},(0,S.operationUnit)(e)),{marginInlineStart:e.marginXXS})}),(e=>{let{componentCls:t,paddingSM:n}=e;return{"&-edit-content":{position:"relative","div&":{insetInlineStart:e.calc(e.paddingSM).mul(-1).equal(),insetBlockStart:e.calc(n).div(-2).add(1).equal(),marginBottom:e.calc(n).div(2).sub(2).equal()},[`${t}-edit-content-confirm`]:{position:"absolute",insetInlineEnd:e.calc(e.marginXS).add(2).equal(),insetBlockEnd:e.marginXS,color:e.colorIcon,fontWeight:"normal",fontSize:e.fontSize,fontStyle:"normal",pointerEvents:"none"},textarea:{margin:"0!important",MozTransition:"none",height:"1em"}}}})(e)),{[`${e.componentCls}-copy-success`]:{[` - &, - &:hover, - &:focus`]:{color:e.colorSuccess}},[`${e.componentCls}-copy-icon-only`]:{marginInlineStart:0}}),{[` - a&-ellipsis, - span&-ellipsis - `]:{display:"inline-block",maxWidth:"100%"},"&-ellipsis-single-line":{whiteSpace:"nowrap",overflow:"hidden",textOverflow:"ellipsis","a&, span&":{verticalAlign:"bottom"},"> code":{paddingBlock:0,maxWidth:"calc(100% - 1.2em)",display:"inline-block",overflow:"hidden",textOverflow:"ellipsis",verticalAlign:"bottom",boxSizing:"content-box"}},"&-ellipsis-multiple-line":{display:"-webkit-box",overflow:"hidden",WebkitLineClamp:3,WebkitBoxOrient:"vertical"}}),{"&-rtl":{direction:"rtl"}})}},()=>({titleMarginTop:"1.2em",titleMarginBottom:"0.5em"})),k=e=>{let{prefixCls:n,"aria-label":l,className:r,style:o,direction:a,maxLength:c,autoSize:s=!0,value:u,onSave:d,onCancel:p,onEnd:f,component:m,enterIcon:g=t.createElement(h,null)}=e,b=t.useRef(null),y=t.useRef(!1),v=t.useRef(null),[S,w]=t.useState(u);t.useEffect(()=>{w(u)},[u]),t.useEffect(()=>{var e;if(null==(e=b.current)?void 0:e.resizableTextArea){let{textArea:e}=b.current.resizableTextArea;e.focus();let{length:t}=e.value;e.setSelectionRange(t,t)}},[]);let j=()=>{d(S.trim())},[k,R,$]=C(n),T=(0,i.default)(n,`${n}-edit-content`,{[`${n}-rtl`]:"rtl"===a,[`${n}-${m}`]:!!m},r,R,$);return k(t.createElement("div",{className:T,style:o},t.createElement(E.default,{ref:b,maxLength:c,value:S,onChange:({target:e})=>{w(e.value.replace(/[\n\r]/g,""))},onKeyDown:({keyCode:e})=>{y.current||(v.current=e)},onKeyUp:({keyCode:e,ctrlKey:t,altKey:n,metaKey:l,shiftKey:r})=>{v.current!==e||y.current||t||n||l||r||(e===x.default.ENTER?(j(),null==f||f()):e===x.default.ESC&&p())},onCompositionStart:()=>{y.current=!0},onCompositionEnd:()=>{y.current=!1},onBlur:()=>{j()},"aria-label":l,rows:1,autoSize:s}),null!==g?(0,O.cloneElement)(g,{className:`${n}-edit-content-confirm`}):null))};var R=e.i(844343),$=e.i(175066);function T(e,n){return t.useMemo(()=>{let t=!!e;return[t,Object.assign(Object.assign({},n),t&&"object"==typeof e?e:null)]},[e])}var I=function(e,t){var n={};for(var l in e)Object.prototype.hasOwnProperty.call(e,l)&&0>t.indexOf(l)&&(n[l]=e[l]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var r=0,l=Object.getOwnPropertySymbols(e);rt.indexOf(l[r])&&Object.prototype.propertyIsEnumerable.call(e,l[r])&&(n[l[r]]=e[l[r]]);return n};let D=t.forwardRef((e,n)=>{let{prefixCls:l,component:r="article",className:o,rootClassName:a,setContentRef:c,children:s,direction:u,style:d}=e,p=I(e,["prefixCls","component","className","rootClassName","setContentRef","children","direction","style"]),{getPrefixCls:m,direction:b,className:y,style:v}=(0,g.useComponentConfig)("typography"),h=c?(0,f.composeRef)(n,c):n,x=m("typography",l),[O,E,S]=C(x),w=(0,i.default)(x,y,{[`${x}-rtl`]:"rtl"===(null!=u?u:b)},o,a,E,S),j=Object.assign(Object.assign({},v),d);return O(t.createElement(r,Object.assign({className:w,style:j,ref:h},p),s))});var P=e.i(121229),B=e.i(190144),M=e.i(739295);function H(e){return!1===e?[!1,!1]:Array.isArray(e)?e:[e]}function z(e,t,n){return!0===e||void 0===e?t:e||n&&t}let A=e=>["string","number"].includes(typeof e),W=({prefixCls:e,copied:n,locale:l,iconOnly:r,tooltips:o,icon:a,tabIndex:c,onCopy:s,loading:u})=>{let d=H(o),p=H(a),{copied:f,copy:m}=null!=l?l:{},g=n?f:m,b=z(d[+!!n],g),v="string"==typeof b?b:g;return t.createElement(y.default,{title:b},t.createElement("button",{type:"button",className:(0,i.default)(`${e}-copy`,{[`${e}-copy-success`]:n,[`${e}-copy-icon-only`]:r}),onClick:s,"aria-label":v,tabIndex:c},n?z(p[1],t.createElement(P.default,null),!0):z(p[0],u?t.createElement(M.default,null):t.createElement(B.default,null),!0)))},L=t.forwardRef(({style:e,children:n},l)=>{let r=t.useRef(null);return t.useImperativeHandle(l,()=>({isExceed:()=>{let e=r.current;return e.scrollHeight>e.clientHeight},getHeight:()=>r.current.clientHeight})),t.createElement("span",{"aria-hidden":!0,ref:r,style:Object.assign({position:"fixed",display:"block",left:0,top:0,pointerEvents:"none",backgroundColor:"rgba(255, 0, 0, 0.65)"},e)},n)});function N(e,t){let n=0,l=[];for(let r=0;rt){let e=t-n;return l.push(String(o).slice(0,e)),l}l.push(o),n=a}return e}let U={display:"-webkit-box",overflow:"hidden",WebkitBoxOrient:"vertical"};function F(e){let{enableMeasure:l,width:r,text:o,children:a,rows:i,expanded:c,miscDeps:d,onEllipsis:p}=e,f=t.useMemo(()=>(0,s.default)(o),[o]),m=t.useMemo(()=>f.reduce((e,t)=>e+(A(t)?String(t).length:1),0),[o]),g=t.useMemo(()=>a(f,!1),[o]),[b,y]=t.useState(null),v=t.useRef(null),h=t.useRef(null),x=t.useRef(null),O=t.useRef(null),E=t.useRef(null),[S,w]=t.useState(!1),[j,C]=t.useState(0),[k,R]=t.useState(0),[$,T]=t.useState(null);(0,u.default)(()=>{l&&r&&m?C(1):C(0)},[r,o,i,l,f]),(0,u.default)(()=>{var e,t,n,l;if(1===j)C(2),T(h.current&&getComputedStyle(h.current).whiteSpace);else if(2===j){let r=!!(null==(e=x.current)?void 0:e.isExceed());C(r?3:4),y(r?[0,m]:null),w(r),R(Math.max((null==(t=x.current)?void 0:t.getHeight())||0,(1===i?0:(null==(n=O.current)?void 0:n.getHeight())||0)+((null==(l=E.current)?void 0:l.getHeight())||0))+1),p(r)}},[j]);let I=b?Math.ceil((b[0]+b[1])/2):0;(0,u.default)(()=>{var e;let[t,n]=b||[0,0];if(t!==n){let l=((null==(e=v.current)?void 0:e.getHeight())||0)>k,r=I;n-t==1&&(r=l?t:n),y(l?[t,r]:[r,n])}},[b,I]);let D=t.useMemo(()=>{if(!l)return a(f,!1);if(3!==j||!b||b[0]!==b[1]){let e=a(f,!1);return[4,0].includes(j)?e:t.createElement("span",{style:Object.assign(Object.assign({},U),{WebkitLineClamp:i})},e)}return a(c?f:N(f,b[0]),S)},[c,j,b,f].concat((0,n.default)(d))),P={width:r,margin:0,padding:0,whiteSpace:"nowrap"===$?"normal":"inherit"};return t.createElement(t.Fragment,null,D,2===j&&t.createElement(t.Fragment,null,t.createElement(L,{style:Object.assign(Object.assign(Object.assign({},P),U),{WebkitLineClamp:i}),ref:x},g),t.createElement(L,{style:Object.assign(Object.assign(Object.assign({},P),U),{WebkitLineClamp:i-1}),ref:O},g),t.createElement(L,{style:Object.assign(Object.assign(Object.assign({},P),U),{WebkitLineClamp:1}),ref:E},a([],!0))),3===j&&b&&b[0]!==b[1]&&t.createElement(L,{style:Object.assign(Object.assign({},P),{top:400}),ref:v},a(N(f,I),!0)),1===j&&t.createElement("span",{style:{whiteSpace:"inherit"},ref:h}))}let q=({enableEllipsis:e,isEllipsis:n,children:l,tooltipProps:r})=>(null==r?void 0:r.title)&&e?t.createElement(y.default,Object.assign({open:!!n&&void 0},r),l):l;var X=function(e,t){var n={};for(var l in e)Object.prototype.hasOwnProperty.call(e,l)&&0>t.indexOf(l)&&(n[l]=e[l]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var r=0,l=Object.getOwnPropertySymbols(e);rt.indexOf(l[r])&&Object.prototype.propertyIsEnumerable.call(e,l[r])&&(n[l[r]]=e[l[r]]);return n};let K=["delete","mark","code","underline","strong","keyboard","italic"],V=t.forwardRef((e,l)=>{var r;let o,v,h,{prefixCls:x,className:O,style:E,type:S,disabled:w,children:j,ellipsis:C,editable:I,copyable:P,component:B,title:M}=e,H=X(e,["prefixCls","className","style","type","disabled","children","ellipsis","editable","copyable","component","title"]),{getPrefixCls:z,direction:L}=t.useContext(g.ConfigContext),[N]=(0,b.default)("Text"),U=t.useRef(null),V=t.useRef(null),_=z("typography",x),G=(0,p.default)(H,K),[J,Q]=T(I),[Y,Z]=(0,d.default)(!1,{value:Q.editing}),{triggerType:ee=["icon"]}=Q,et=e=>{var t;e&&(null==(t=Q.onStart)||t.call(Q)),Z(e)},en=(o=(0,t.useRef)(void 0),(0,t.useEffect)(()=>{o.current=Y}),o.current);(0,u.default)(()=>{var e;!Y&&en&&(null==(e=V.current)||e.focus())},[Y]);let el=e=>{null==e||e.preventDefault(),et(!0)},[er,eo]=T(P),{copied:ea,copyLoading:ei,onClick:ec}=(({copyConfig:e,children:n})=>{let[l,r]=t.useState(!1),[o,a]=t.useState(!1),i=t.useRef(null),c=()=>{i.current&&clearTimeout(i.current)},s={};e.format&&(s.format=e.format),t.useEffect(()=>c,[]);let u=(0,$.default)(t=>{var l,o,u,d;return l=void 0,o=void 0,u=void 0,d=function*(){var l;null==t||t.preventDefault(),null==t||t.stopPropagation(),a(!0);try{let o="function"==typeof e.text?yield e.text():e.text;(0,R.default)(o||((e,t=!1)=>t&&null==e?[]:Array.isArray(e)?e:[e])(n,!0).join("")||"",s),a(!1),r(!0),c(),i.current=setTimeout(()=>{r(!1)},3e3),null==(l=e.onCopy)||l.call(e,t)}catch(e){throw a(!1),e}},new(u||(u=Promise))(function(e,t){function n(e){try{a(d.next(e))}catch(e){t(e)}}function r(e){try{a(d.throw(e))}catch(e){t(e)}}function a(t){var l;t.done?e(t.value):((l=t.value)instanceof u?l:new u(function(e){e(l)})).then(n,r)}a((d=d.apply(l,o||[])).next())})});return{copied:l,copyLoading:o,onClick:u}})({copyConfig:eo,children:j}),[es,eu]=t.useState(!1),[ed,ep]=t.useState(!1),[ef,em]=t.useState(!1),[eg,eb]=t.useState(!1),[ey,ev]=t.useState(!0),[eh,ex]=T(C,{expandable:!1,symbol:e=>e?null==N?void 0:N.collapse:null==N?void 0:N.expand}),[eO,eE]=(0,d.default)(ex.defaultExpanded||!1,{value:ex.expanded}),eS=eh&&(!eO||"collapsible"===ex.expandable),{rows:ew=1}=ex,ej=t.useMemo(()=>eS&&(void 0!==ex.suffix||ex.onEllipsis||ex.expandable||J||er),[eS,ex,J,er]);(0,u.default)(()=>{eh&&!ej&&(eu((0,m.isStyleSupport)("webkitLineClamp")),ep((0,m.isStyleSupport)("textOverflow")))},[ej,eh]);let[eC,ek]=t.useState(eS),eR=t.useMemo(()=>!ej&&(1===ew?ed:es),[ej,ed,es]);(0,u.default)(()=>{ek(eR&&eS)},[eR,eS]);let e$=eS&&(eC?eg:ef),eT=eS&&1===ew&&eC,eI=eS&&ew>1&&eC,[eD,eP]=t.useState(0),eB=e=>{var t;em(e),ef!==e&&(null==(t=ex.onEllipsis)||t.call(ex,e))};t.useEffect(()=>{let e=U.current;if(eh&&eC&&e){let t,n,l,r=(t=document.createElement("em"),e.appendChild(t),n=e.getBoundingClientRect(),l=t.getBoundingClientRect(),e.removeChild(t),n.left>l.left||l.right>n.right||n.top>l.top||l.bottom>n.bottom);eg!==r&&eb(r)}},[eh,eC,j,eI,ey,eD]),t.useEffect(()=>{let e=U.current;if("u"{ev(!!e.offsetParent)});return t.observe(e),()=>{t.disconnect()}},[eC,eS]);let eM=(v=ex.tooltip,h=Q.text,(0,t.useMemo)(()=>!0===v?{title:null!=h?h:j}:(0,t.isValidElement)(v)?{title:v}:"object"==typeof v?Object.assign({title:null!=h?h:j},v):{title:v},[v,h,j])),eH=t.useMemo(()=>{if(eh&&!eC)return[Q.text,j,M,eM.title].find(A)},[eh,eC,M,eM.title,e$]);return Y?t.createElement(k,{value:null!=(r=Q.text)?r:"string"==typeof j?j:"",onSave:e=>{var t;null==(t=Q.onChange)||t.call(Q,e),et(!1)},onCancel:()=>{var e;null==(e=Q.onCancel)||e.call(Q),et(!1)},onEnd:Q.onEnd,prefixCls:_,className:O,style:E,direction:L,component:B,maxLength:Q.maxLength,autoSize:Q.autoSize,enterIcon:Q.enterIcon}):t.createElement(c.default,{onResize:({offsetWidth:e})=>{eP(e)},disabled:!eS},r=>t.createElement(q,{tooltipProps:eM,enableEllipsis:eS,isEllipsis:e$},t.createElement(D,Object.assign({className:(0,i.default)({[`${_}-${S}`]:S,[`${_}-disabled`]:w,[`${_}-ellipsis`]:eh,[`${_}-ellipsis-single-line`]:eT,[`${_}-ellipsis-multiple-line`]:eI},O),prefixCls:x,style:Object.assign(Object.assign({},E),{WebkitLineClamp:eI?ew:void 0}),component:B,ref:(0,f.composeRef)(r,U,l),direction:L,onClick:ee.includes("text")?el:void 0,"aria-label":null==eH?void 0:eH.toString(),title:M},G),t.createElement(F,{enableMeasure:eS&&!eC,text:j,rows:ew,width:eD,onEllipsis:eB,expanded:eO,miscDeps:[ea,eO,ei,J,er,N].concat((0,n.default)(K.map(t=>e[t])))},(n,l)=>{let r;return function({mark:e,code:n,underline:l,delete:r,strong:o,keyboard:a,italic:i},c){let s=c;function u(e,n){n&&(s=t.createElement(e,{},s))}return u("strong",o),u("u",l),u("del",r),u("code",n),u("mark",e),u("kbd",a),u("i",i),s}(e,t.createElement(t.Fragment,null,n.length>0&&l&&!eO&&eH?t.createElement("span",{key:"show-content","aria-hidden":!0},n):n,[(r=l)&&!eO&&t.createElement("span",{"aria-hidden":!0,key:"ellipsis"},"..."),ex.suffix,[r&&(()=>{let{expandable:e,symbol:n}=ex;return e?t.createElement("button",{type:"button",key:"expand",className:`${_}-${eO?"collapse":"expand"}`,onClick:e=>{var t,n;eE((t={expanded:!eO}).expanded),null==(n=ex.onExpand)||n.call(ex,e,t)},"aria-label":eO?N.collapse:null==N?void 0:N.expand},"function"==typeof n?n(eO):n):null})(),(()=>{if(!J)return;let{icon:e,tooltip:n,tabIndex:l}=Q,r=(0,s.default)(n)[0]||(null==N?void 0:N.edit),o="string"==typeof r?r:"";return ee.includes("icon")?t.createElement(y.default,{key:"edit",title:!1===n?"":r},t.createElement("button",{type:"button",ref:V,className:`${_}-edit`,onClick:el,"aria-label":o,tabIndex:l},e||t.createElement(a,{role:"button"}))):null})(),er?t.createElement(W,Object.assign({key:"copy"},eo,{prefixCls:_,copied:ea,locale:N,onCopy:ec,loading:ei,iconOnly:null==j})):null]]))}))))});var _=function(e,t){var n={};for(var l in e)Object.prototype.hasOwnProperty.call(e,l)&&0>t.indexOf(l)&&(n[l]=e[l]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var r=0,l=Object.getOwnPropertySymbols(e);rt.indexOf(l[r])&&Object.prototype.propertyIsEnumerable.call(e,l[r])&&(n[l[r]]=e[l[r]]);return n};let G=t.forwardRef((e,n)=>{let{ellipsis:l,rel:r,children:o,navigate:a}=e,i=_(e,["ellipsis","rel","children","navigate"]),c=Object.assign(Object.assign({},i),{rel:void 0===r&&"_blank"===i.target?"noopener noreferrer":r});return t.createElement(V,Object.assign({},c,{ref:n,ellipsis:!!l,component:"a"}),o)});var J=function(e,t){var n={};for(var l in e)Object.prototype.hasOwnProperty.call(e,l)&&0>t.indexOf(l)&&(n[l]=e[l]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var r=0,l=Object.getOwnPropertySymbols(e);rt.indexOf(l[r])&&Object.prototype.propertyIsEnumerable.call(e,l[r])&&(n[l[r]]=e[l[r]]);return n};let Q=t.forwardRef((e,n)=>{let{children:l}=e,r=J(e,["children"]);return t.createElement(V,Object.assign({ref:n},r,{component:"div"}),l)});var Y=function(e,t){var n={};for(var l in e)Object.prototype.hasOwnProperty.call(e,l)&&0>t.indexOf(l)&&(n[l]=e[l]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var r=0,l=Object.getOwnPropertySymbols(e);rt.indexOf(l[r])&&Object.prototype.propertyIsEnumerable.call(e,l[r])&&(n[l[r]]=e[l[r]]);return n};let Z=t.forwardRef((e,n)=>{let{ellipsis:l,children:r}=e,o=Y(e,["ellipsis","children"]),a=t.useMemo(()=>l&&"object"==typeof l?(0,p.default)(l,["expandable","rows"]):l,[l]);return t.createElement(V,Object.assign({ref:n},o,{ellipsis:a,component:"span"}),r)});var ee=function(e,t){var n={};for(var l in e)Object.prototype.hasOwnProperty.call(e,l)&&0>t.indexOf(l)&&(n[l]=e[l]);if(null!=e&&"function"==typeof Object.getOwnPropertySymbols)for(var r=0,l=Object.getOwnPropertySymbols(e);rt.indexOf(l[r])&&Object.prototype.propertyIsEnumerable.call(e,l[r])&&(n[l[r]]=e[l[r]]);return n};let et=[1,2,3,4,5],en=t.forwardRef((e,n)=>{let{level:l=1,children:r}=e,o=ee(e,["level","children"]),a=et.includes(l)?`h${l}`:"h1";return t.createElement(V,Object.assign({ref:n},o,{component:a}),r)});e.s(["default",0,en],335771),D.Text=Z,D.Link=G,D.Title=en,D.Paragraph=Q,e.s(["Typography",0,D],898586)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/36ccc2b555a26ad4.js b/litellm/proxy/_experimental/out/_next/static/chunks/05d4ceb8d45fdc83.js similarity index 96% rename from litellm/proxy/_experimental/out/_next/static/chunks/36ccc2b555a26ad4.js rename to litellm/proxy/_experimental/out/_next/static/chunks/05d4ceb8d45fdc83.js index d601999bfa6..b544627b867 100644 --- a/litellm/proxy/_experimental/out/_next/static/chunks/36ccc2b555a26ad4.js +++ b/litellm/proxy/_experimental/out/_next/static/chunks/05d4ceb8d45fdc83.js @@ -1,4 +1,4 @@ (globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,429427,371330,80758,402155,368578,544508,746725,835696,941444,914189,394487,e=>{"use strict";let t;e.i(247167);var r=e.i(271645);let n="u">typeof document?r.default.useLayoutEffect:()=>{},o=e=>{var t;return null!=(t=null==e?void 0:e.ownerDocument)?t:document},a=e=>e&&"window"in e&&e.window===e?e:o(e).defaultView||window;"u">typeof Element&&Element.prototype;let s=["input:not([disabled]):not([type=hidden])","select:not([disabled])","textarea:not([disabled])","button:not([disabled])","a[href]","area[href]","summary","iframe","object","embed","audio[controls]","video[controls]",'[contenteditable]:not([contenteditable^="false"])',"permission"];s.join(":not([hidden]),"),s.push('[tabindex]:not([tabindex="-1"]):not([disabled])'),s.join(':not([hidden]):not([tabindex="-1"]),');let l=null;function i(e){return e.nativeEvent=e,e.isDefaultPrevented=()=>e.defaultPrevented,e.isPropagationStopped=()=>e.cancelBubble,e.persist=()=>{},e}function u(e){let t=(0,r.useRef)({isFocused:!1,observer:null});return n(()=>{let e=t.current;return()=>{e.observer&&(e.observer.disconnect(),e.observer=null)}},[]),(0,r.useCallback)(r=>{if(r.target instanceof HTMLButtonElement||r.target instanceof HTMLInputElement||r.target instanceof HTMLTextAreaElement||r.target instanceof HTMLSelectElement){t.current.isFocused=!0;let n=r.target;n.addEventListener("focusout",r=>{if(t.current.isFocused=!1,n.disabled){let t=i(r);null==e||e(t)}t.current.observer&&(t.current.observer.disconnect(),t.current.observer=null)},{once:!0}),t.current.observer=new MutationObserver(()=>{if(t.current.isFocused&&n.disabled){var e;null==(e=t.current.observer)||e.disconnect();let r=n===document.activeElement?null:document.activeElement;n.dispatchEvent(new FocusEvent("blur",{relatedTarget:r})),n.dispatchEvent(new FocusEvent("focusout",{bubbles:!0,relatedTarget:r}))}}),t.current.observer.observe(n,{attributes:!0,attributeFilter:["disabled"]})}},[e])}function c(e){var t;if("u"e.test(t.brand))||e.test(window.navigator.userAgent)}function d(e){var t;return"u">typeof window&&null!=window.navigator&&e.test((null==(t=window.navigator.userAgentData)?void 0:t.platform)||window.navigator.platform)}function f(e){let t=null;return()=>(null==t&&(t=e()),t)}let p=f(function(){return d(/^Mac/i)}),m=f(function(){return d(/^iPhone/i)}),v=f(function(){return d(/^iPad/i)||p()&&navigator.maxTouchPoints>1}),b=f(function(){return m()||v()});f(function(){return p()||b()});let g=f(function(){return c(/AppleWebKit/i)&&!h()}),h=f(function(){return c(/Chrome/i)}),y=f(function(){return c(/Android/i)}),E=f(function(){return c(/Firefox/i)});function w(e,t,r=!0){var n,o;let{metaKey:a,ctrlKey:s,altKey:i,shiftKey:u}=t;E()&&(null==(o=window.event)||null==(n=o.type)?void 0:n.startsWith("key"))&&"_blank"===e.target&&(p()?a=!0:s=!0);let c=g()&&p()&&!v()&&1?new KeyboardEvent("keydown",{keyIdentifier:"Enter",metaKey:a,ctrlKey:s,altKey:i,shiftKey:u}):new MouseEvent("click",{metaKey:a,ctrlKey:s,altKey:i,shiftKey:u,detail:1,bubbles:!0,cancelable:!0});if(w.isOpening=r,function(){if(null==l){l=!1;try{document.createElement("div").focus({get preventScroll(){return l=!0,!0}})}catch{}}return l}())e.focus({preventScroll:!0});else{let t=function(e){let t=e.parentNode,r=[],n=document.scrollingElement||document.documentElement;for(;t instanceof HTMLElement&&t!==n;)(t.offsetHeighttypeof window&&window.document&&window.document.createElement,new WeakMap;r.default.useId;let x=null,F=new Set,P=new Map,k=!1,L=!1,N={Tab:!0,Escape:!0};function C(e,t){for(let r of F)r(e,t)}function I(e){k=!0,w.isOpening||e.metaKey||!p()&&e.altKey||e.ctrlKey||"Control"===e.key||"Shift"===e.key||"Meta"===e.key||(x="keyboard",C("keyboard",e))}function S(e){x="pointer","pointerType"in e&&e.pointerType,("mousedown"===e.type||"pointerdown"===e.type)&&(k=!0,C("pointer",e))}function A(e){w.isOpening||(""!==e.pointerType||!e.isTrusted)&&(y()&&e.pointerType?"click"!==e.type||1!==e.buttons:0!==e.detail||e.pointerType)||(k=!0,x="virtual")}function M(e){e.target!==window&&e.target!==document&&e.isTrusted&&(k||L||(x="virtual",C("virtual",e)),k=!1,L=!1)}function R(){k=!1,L=!0}function O(e){if("u"typeof PointerEvent&&(r.addEventListener("pointerdown",S,!0),r.addEventListener("pointermove",S,!0),r.addEventListener("pointerup",S,!0)),t.addEventListener("beforeunload",()=>{D(e)},{once:!0}),P.set(t,{focus:n})}let D=(e,t)=>{let r=a(e),n=o(e);t&&n.removeEventListener("DOMContentLoaded",t),P.has(r)&&(r.HTMLElement.prototype.focus=P.get(r).focus,n.removeEventListener("keydown",I,!0),n.removeEventListener("keyup",I,!0),n.removeEventListener("click",A,!0),r.removeEventListener("focus",M,!0),r.removeEventListener("blur",R,!1),"u">typeof PointerEvent&&(n.removeEventListener("pointerdown",S,!0),n.removeEventListener("pointermove",S,!0),n.removeEventListener("pointerup",S,!0)),P.delete(r))};function H(){return"pointer"!==x}"u">typeof document&&("loading"!==(t=o(void 0)).readyState?O(void 0):t.addEventListener("DOMContentLoaded",()=>{O(void 0)}));let j=new Set(["checkbox","radio","range","color","file","image","button","submit","reset"]);function K(e,t){return!!t&&!!e&&e.contains(t)}function W(){let e=(0,r.useRef)(new Map),t=(0,r.useCallback)((t,r,n,o)=>{let a=(null==o?void 0:o.once)?(...t)=>{e.current.delete(n),n(...t)}:n;e.current.set(n,{type:r,eventTarget:t,fn:a,options:o}),t.addEventListener(r,a,o)},[]),n=(0,r.useCallback)((t,r,n,o)=>{var a;let s=(null==(a=e.current.get(n))?void 0:a.fn)||n;t.removeEventListener(r,s,o),e.current.delete(n)},[]),o=(0,r.useCallback)(()=>{e.current.forEach((e,t)=>{n(e.eventTarget,e.type,t,e.options)})},[n]);return(0,r.useEffect)(()=>o,[o]),{addGlobalListener:t,removeGlobalListener:n,removeAllGlobalListeners:o}}function B(e={}){var t;let{autoFocus:n=!1,isTextInput:s,within:l}=e,c=(0,r.useRef)({isFocused:!1,isFocusVisible:n||H()}),[d,f]=(0,r.useState)(!1),[p,m]=(0,r.useState)(()=>c.current.isFocused&&c.current.isFocusVisible),v=(0,r.useCallback)(()=>m(c.current.isFocused&&c.current.isFocusVisible),[]),b=(0,r.useCallback)(e=>{c.current.isFocused=e,f(e),v()},[v]);t={isTextInput:s},O(),(0,r.useEffect)(()=>{let e=(e,r)=>{var n;let s,l,i,u,d;n=!!(null==t?void 0:t.isTextInput),s=o(null==r?void 0:r.target),l="u">typeof window?a(null==r?void 0:r.target).HTMLInputElement:HTMLInputElement,i="u">typeof window?a(null==r?void 0:r.target).HTMLTextAreaElement:HTMLTextAreaElement,u="u">typeof window?a(null==r?void 0:r.target).HTMLElement:HTMLElement,d="u">typeof window?a(null==r?void 0:r.target).KeyboardEvent:KeyboardEvent,(n=n||s.activeElement instanceof l&&!j.has(s.activeElement.type)||s.activeElement instanceof i||s.activeElement instanceof u&&s.activeElement.isContentEditable)&&"keyboard"===e&&r instanceof d&&!N[r.key]||(e=>{c.current.isFocusVisible=e,v()})(H())};return F.add(e),()=>{F.delete(e)}},[]);let{focusProps:g}=function(e){let{isDisabled:t,onFocus:n,onBlur:a,onFocusChange:s}=e,l=(0,r.useCallback)(e=>{if(e.target===e.currentTarget)return a&&a(e),s&&s(!1),!0},[a,s]),i=u(l),c=(0,r.useCallback)(e=>{var t;let r=o(e.target),a=r?((e=document)=>e.activeElement)(r):((e=document)=>e.activeElement)();e.target===e.currentTarget&&a===(t=e.nativeEvent,t.target)&&(n&&n(e),s&&s(!0),i(e))},[s,n,i]);return{focusProps:{onFocus:!t&&(n||s||a)?c:void 0,onBlur:!t&&(a||s)?l:void 0}}}({isDisabled:l,onFocusChange:b}),{focusWithinProps:h}=function(e){let{isDisabled:t,onBlurWithin:n,onFocusWithin:a,onFocusWithinChange:s}=e,l=(0,r.useRef)({isFocusWithin:!1}),{addGlobalListener:c,removeAllGlobalListeners:d}=W(),f=(0,r.useCallback)(e=>{e.currentTarget.contains(e.target)&&l.current.isFocusWithin&&!e.currentTarget.contains(e.relatedTarget)&&(l.current.isFocusWithin=!1,d(),n&&n(e),s&&s(!1))},[n,s,l,d]),p=u(f),m=(0,r.useCallback)(e=>{var t;if(!e.currentTarget.contains(e.target))return;let r=o(e.target),n=((e=document)=>e.activeElement)(r);if(!l.current.isFocusWithin&&n===(t=e.nativeEvent,t.target)){a&&a(e),s&&s(!0),l.current.isFocusWithin=!0,p(e);let t=e.currentTarget;c(r,"focus",e=>{if(l.current.isFocusWithin&&!K(t,e.target)){let n=new r.defaultView.FocusEvent("blur",{relatedTarget:e.target});Object.defineProperty(n,"target",{value:t}),Object.defineProperty(n,"currentTarget",{value:t}),f(i(n))}},{capture:!0})}},[a,s,p,c,f]);return t?{focusWithinProps:{onFocus:void 0,onBlur:void 0}}:{focusWithinProps:{onFocus:m,onBlur:f}}}({isDisabled:!l,onFocusWithinChange:b});return{isFocused:d,isFocusVisible:p,focusProps:l?h:g}}e.s(["useFocusRing",()=>B],429427);let V=!1,_=0;function G(e){"touch"===e.pointerType&&(V=!0,setTimeout(()=>{V=!1},50))}function U(){if("u">typeof document)return 0===_&&"u">typeof PointerEvent&&document.addEventListener("pointerup",G),_++,()=>{!(--_>0)&&"u">typeof PointerEvent&&document.removeEventListener("pointerup",G)}}function $(e){let{onHoverStart:t,onHoverChange:n,onHoverEnd:a,isDisabled:s}=e,[l,i]=(0,r.useState)(!1),u=(0,r.useRef)({isHovered:!1,ignoreEmulatedMouseEvents:!1,pointerType:"",target:null}).current;(0,r.useEffect)(U,[]);let{addGlobalListener:c,removeAllGlobalListeners:d}=W(),{hoverProps:f,triggerHoverEnd:p}=(0,r.useMemo)(()=>{let e=(e,t)=>{let r=u.target;u.pointerType="",u.target=null,"touch"!==t&&u.isHovered&&r&&(u.isHovered=!1,d(),a&&a({type:"hoverend",target:r,pointerType:t}),n&&n(!1),i(!1))},r={};return"u">typeof PointerEvent&&(r.onPointerEnter=r=>{V&&"mouse"===r.pointerType||((r,a)=>{if(u.pointerType=a,s||"touch"===a||u.isHovered||!r.currentTarget.contains(r.target))return;u.isHovered=!0;let l=r.currentTarget;u.target=l,c(o(r.target),"pointerover",t=>{u.isHovered&&u.target&&!K(u.target,t.target)&&e(t,t.pointerType)},{capture:!0}),t&&t({type:"hoverstart",target:l,pointerType:a}),n&&n(!0),i(!0)})(r,r.pointerType)},r.onPointerLeave=t=>{!s&&t.currentTarget.contains(t.target)&&e(t,t.pointerType)}),{hoverProps:r,triggerHoverEnd:e}},[t,n,a,s,u,c,d]);return(0,r.useEffect)(()=>{s&&p({currentTarget:u.target},u.pointerType)},[s]),{hoverProps:f,isHovered:l}}e.s(["useHover",()=>$],371330);var q=Object.defineProperty,X=(e,t,r)=>{let n;return(n="symbol"!=typeof t?t+"":t)in e?q(e,n,{enumerable:!0,configurable:!0,writable:!0,value:r}):e[n]=r,r};let Y=new class{constructor(){X(this,"current",this.detect()),X(this,"handoffState","pending"),X(this,"currentId",0)}set(e){this.current!==e&&(this.handoffState="pending",this.currentId=0,this.current=e)}reset(){this.set(this.detect())}nextId(){return++this.currentId}get isServer(){return"server"===this.current}get isClient(){return"client"===this.current}detect(){return"u"setTimeout(()=>{throw e}))}function J(){let e=[],t={addEventListener:(e,r,n,o)=>(e.addEventListener(r,n,o),t.add(()=>e.removeEventListener(r,n,o))),requestAnimationFrame(...e){let r=requestAnimationFrame(...e);return t.add(()=>cancelAnimationFrame(r))},nextFrame:(...e)=>t.requestAnimationFrame(()=>t.requestAnimationFrame(...e)),setTimeout(...e){let r=setTimeout(...e);return t.add(()=>clearTimeout(r))},microTask(...e){let r={current:!0};return Z(()=>{r.current&&e[0]()}),t.add(()=>{r.current=!1})},style(e,t,r){let n=e.style.getPropertyValue(t);return Object.assign(e.style,{[t]:r}),this.add(()=>{Object.assign(e.style,{[t]:n})})},group(e){let t=J();return e(t),this.add(()=>t.dispose())},add:t=>(e.includes(t)||e.push(t),()=>{let r=e.indexOf(t);if(r>=0)for(let t of e.splice(r,1))t()}),dispose(){for(let t of e.splice(0))t()}};return t}function Q(){let[e]=(0,r.useState)(J);return(0,r.useEffect)(()=>()=>e.dispose(),[e]),e}e.s(["env",()=>Y],80758),e.s(["getOwnerDocument",()=>z],402155),e.s(["microTask",()=>Z],368578),e.s(["disposables",()=>J],544508),e.s(["useDisposables",()=>Q],746725);let ee=(e,t)=>{Y.isServer?(0,r.useEffect)(e,t):(0,r.useLayoutEffect)(e,t)};function et(e){let t=(0,r.useRef)(e);return ee(()=>{t.current=e},[e]),t}e.s(["useIsoMorphicEffect",()=>ee],835696),e.s(["useLatestValue",()=>et],941444);let er=function(e){let t=et(e);return r.default.useCallback((...e)=>t.current(...e),[t])};function en({disabled:e=!1}={}){let t=(0,r.useRef)(null),[n,o]=(0,r.useState)(!1),a=Q(),s=er(()=>{t.current=null,o(!1),a.dispose()}),l=er(e=>{if(a.dispose(),null===t.current){t.current=e.currentTarget,o(!0);{let r=z(e.currentTarget);a.addEventListener(r,"pointerup",s,!1),a.addEventListener(r,"pointermove",e=>{if(t.current){var r,n;let a,s;o((a=e.width/2,s=e.height/2,r={top:e.clientY-s,right:e.clientX+a,bottom:e.clientY+s,left:e.clientX-a},n=t.current.getBoundingClientRect(),!(!r||!n||r.rightn.right||r.bottomn.bottom)))}},!1),a.addEventListener(r,"pointercancel",s,!1)}}});return{pressed:n,pressProps:e?{}:{onPointerDown:l,onPointerUp:s,onClick:s}}}e.s(["useEvent",()=>er],914189),e.s(["useActivePress",()=>en],394487)},144279,294316,e=>{"use strict";var t=e.i(271645);function r(e,r){return(0,t.useMemo)(()=>{var t;if(e.type)return e.type;let n=null!=(t=e.as)?t:"button";if("string"==typeof n&&"button"===n.toLowerCase()||(null==r?void 0:r.tagName)==="BUTTON"&&!r.hasAttribute("type"))return"button"},[e.type,e.as,r])}e.s(["useResolveButtonType",()=>r],144279);var n=e.i(914189);let o=Symbol();function a(e,t=!0){return Object.assign(e,{[o]:t})}function s(...e){let r=(0,t.useRef)(e);(0,t.useEffect)(()=>{r.current=e},[e]);let a=(0,n.useEvent)(e=>{for(let t of r.current)null!=t&&("function"==typeof t?t(e):t.current=e)});return e.every(e=>null==e||(null==e?void 0:e[o]))?void 0:a}e.s(["optionalRef",()=>a,"useSyncRefs",()=>s],294316)},553521,e=>{"use strict";var t=e.i(271645),r=e.i(835696);function n(){let e=(0,t.useRef)(!1);return(0,r.useIsoMorphicEffect)(()=>(e.current=!0,()=>{e.current=!1}),[]),e}e.s(["useIsMounted",()=>n])},732607,e=>{"use strict";function t(...e){return Array.from(new Set(e.flatMap(e=>"string"==typeof e?e.split(" "):[]))).filter(Boolean).join(" ")}e.s(["classNames",()=>t])},397701,e=>{"use strict";function t(e,r,...n){if(e in r){let t=r[e];return"function"==typeof t?t(...n):t}let o=Error(`Tried to handle "${e}" but there is no handler defined. Only defined handlers are: ${Object.keys(r).map(e=>`"${e}"`).join(", ")}.`);throw Error.captureStackTrace&&Error.captureStackTrace(o,t),o}e.s(["match",()=>t])},700020,e=>{"use strict";let t,r;var n=e.i(271645),o=e.i(732607),a=e.i(397701),s=((t=s||{})[t.None=0]="None",t[t.RenderStrategy=1]="RenderStrategy",t[t.Static=2]="Static",t),l=((r=l||{})[r.Unmount=0]="Unmount",r[r.Hidden=1]="Hidden",r);function i(){let e,t,r=(e=(0,n.useRef)([]),t=(0,n.useCallback)(t=>{for(let r of e.current)null!=r&&("function"==typeof r?r(t):r.current=t)},[]),(...r)=>{if(!r.every(e=>null==e))return e.current=r,t});return(0,n.useCallback)(e=>(function({ourProps:e,theirProps:t,slot:r,defaultTag:n,features:o,visible:s=!0,name:l,mergeRefs:i}){i=null!=i?i:c;let f=d(t,e);if(s)return u(f,r,n,l,i);let p=null!=o?o:0;if(2&p){let{static:e=!1,...t}=f;if(e)return u(t,r,n,l,i)}if(1&p){let{unmount:e=!0,...t}=f;return(0,a.match)(+!e,{0:()=>null,1:()=>u({...t,hidden:!0,style:{display:"none"}},r,n,l,i)})}return u(f,r,n,l,i)})({mergeRefs:r,...e}),[r])}function u(e,t={},r,a,s){let{as:l=r,children:i,refName:c="ref",...f}=v(e,["unmount","static"]),p=void 0!==e.ref?{[c]:e.ref}:{},b="function"==typeof i?i(t):i;"className"in f&&f.className&&"function"==typeof f.className&&(f.className=f.className(t)),f["aria-labelledby"]&&f["aria-labelledby"]===f.id&&(f["aria-labelledby"]=void 0);let g={};if(t){let e=!1,r=[];for(let[n,o]of Object.entries(t))"boolean"==typeof o&&(e=!0),!0===o&&r.push(n.replace(/([A-Z])/g,e=>`-${e.toLowerCase()}`));if(e)for(let e of(g["data-headlessui-state"]=r.join(" "),r))g[`data-${e}`]=""}if(l===n.Fragment&&(Object.keys(m(f)).length>0||Object.keys(m(g)).length>0))if(!(0,n.isValidElement)(b)||Array.isArray(b)&&b.length>1){if(Object.keys(m(f)).length>0)throw Error(['Passing props on "Fragment"!',"",`The current component <${a} /> is rendering a "Fragment".`,"However we need to passthrough the following props:",Object.keys(m(f)).concat(Object.keys(m(g))).map(e=>` - ${e}`).join(` `),"","You can apply a few solutions:",['Add an `as="..."` prop, to ensure that we render an actual element instead of a "Fragment".',"Render a single element as the child so that we can forward the props onto that element."].map(e=>` - ${e}`).join(` `)].join(` -`))}else{var h;let e=b.props,t=null==e?void 0:e.className,r="function"==typeof t?(...e)=>(0,o.classNames)(t(...e),f.className):(0,o.classNames)(t,f.className),a=d(b.props,m(v(f,["ref"])));for(let e in g)e in a&&delete g[e];return(0,n.cloneElement)(b,Object.assign({},a,g,p,{ref:s((h=b,n.default.version.split(".")[0]>="19"?h.props.ref:h.ref),p.ref)},r?{className:r}:{}))}return(0,n.createElement)(l,Object.assign({},v(f,["ref"]),l!==n.Fragment&&p,l!==n.Fragment&&g),b)}function c(...e){return e.every(e=>null==e)?void 0:t=>{for(let r of e)null!=r&&("function"==typeof r?r(t):r.current=t)}}function d(...e){if(0===e.length)return{};if(1===e.length)return e[0];let t={},r={};for(let n of e)for(let e in n)e.startsWith("on")&&"function"==typeof n[e]?(null!=r[e]||(r[e]=[]),r[e].push(n[e])):t[e]=n[e];if(t.disabled||t["aria-disabled"])for(let e in r)/^(on(?:Click|Pointer|Mouse|Key)(?:Down|Up|Press)?)$/.test(e)&&(r[e]=[e=>{var t;return null==(t=null==e?void 0:e.preventDefault)?void 0:t.call(e)}]);for(let e in r)Object.assign(t,{[e](t,...n){for(let o of r[e]){if((t instanceof Event||(null==t?void 0:t.nativeEvent)instanceof Event)&&t.defaultPrevented)return;o(t,...n)}}});return t}function f(...e){if(0===e.length)return{};if(1===e.length)return e[0];let t={},r={};for(let n of e)for(let e in n)e.startsWith("on")&&"function"==typeof n[e]?(null!=r[e]||(r[e]=[]),r[e].push(n[e])):t[e]=n[e];for(let e in r)Object.assign(t,{[e](...t){for(let n of r[e])null==n||n(...t)}});return t}function p(e){var t;return Object.assign((0,n.forwardRef)(e),{displayName:null!=(t=e.displayName)?t:e.name})}function m(e){let t=Object.assign({},e);for(let e in t)void 0===t[e]&&delete t[e];return t}function v(e,t=[]){let r=Object.assign({},e);for(let e of t)e in r&&delete r[e];return r}e.s(["RenderFeatures",()=>s,"RenderStrategy",()=>l,"compact",()=>m,"forwardRefWithAs",()=>p,"mergeProps",()=>f,"useRender",()=>i])},2788,e=>{"use strict";let t;var r=e.i(700020),n=((t=n||{})[t.None=1]="None",t[t.Focusable=2]="Focusable",t[t.Hidden=4]="Hidden",t);let o=(0,r.forwardRefWithAs)(function(e,t){var n;let{features:o=1,...a}=e,s={ref:t,"aria-hidden":(2&o)==2||(null!=(n=a["aria-hidden"])?n:void 0),hidden:(4&o)==4||void 0,style:{position:"fixed",top:1,left:1,width:1,height:0,padding:0,margin:-1,overflow:"hidden",clip:"rect(0, 0, 0, 0)",whiteSpace:"nowrap",borderWidth:"0",...(4&o)==4&&(2&o)!=2&&{display:"none"}}};return(0,r.useRender)()({ourProps:s,theirProps:a,slot:{},defaultTag:"span",name:"Hidden"})});e.s(["Hidden",()=>o,"HiddenFeatures",()=>n])},640497,e=>{"use strict";var t=e.i(271645),r=e.i(553521),n=e.i(2788);function o({onFocus:e}){let[o,a]=(0,t.useState)(!0),s=(0,r.useIsMounted)();return o?t.default.createElement(n.Hidden,{as:"button",type:"button",features:n.HiddenFeatures.Focusable,onFocus:t=>{t.preventDefault();let r,n=50;r=requestAnimationFrame(function t(){if(n--<=0){r&&cancelAnimationFrame(r);return}if(e()){if(cancelAnimationFrame(r),!s.current)return;a(!1);return}r=requestAnimationFrame(t)})}}):null}e.s(["FocusSentinel",()=>o])},652265,e=>{"use strict";let t,r,n,o,a;e.i(544508);var s=e.i(397701),l=e.i(402155);let i=["[contentEditable=true]","[tabindex]","a[href]","area[href]","button:not([disabled])","iframe","input:not([disabled])","select:not([disabled])","textarea:not([disabled])"].map(e=>`${e}:not([tabindex='-1'])`).join(","),u=["[data-autofocus]"].map(e=>`${e}:not([tabindex='-1'])`).join(",");var c=((t=c||{})[t.First=1]="First",t[t.Previous=2]="Previous",t[t.Next=4]="Next",t[t.Last=8]="Last",t[t.WrapAround=16]="WrapAround",t[t.NoScroll=32]="NoScroll",t[t.AutoFocus=64]="AutoFocus",t),d=((r=d||{})[r.Error=0]="Error",r[r.Overflow=1]="Overflow",r[r.Success=2]="Success",r[r.Underflow=3]="Underflow",r),f=((n=f||{})[n.Previous=-1]="Previous",n[n.Next=1]="Next",n);function p(e=document.body){return null==e?[]:Array.from(e.querySelectorAll(i)).sort((e,t)=>Math.sign((e.tabIndex||Number.MAX_SAFE_INTEGER)-(t.tabIndex||Number.MAX_SAFE_INTEGER)))}var m=((o=m||{})[o.Strict=0]="Strict",o[o.Loose=1]="Loose",o);function v(e,t=0){var r;return e!==(null==(r=(0,l.getOwnerDocument)(e))?void 0:r.body)&&(0,s.match)(t,{0:()=>e.matches(i),1(){let t=e;for(;null!==t;){if(t.matches(i))return!0;t=t.parentElement}return!1}})}var b=((a=b||{})[a.Keyboard=0]="Keyboard",a[a.Mouse=1]="Mouse",a);function g(e,t=e=>e){return e.slice().sort((e,r)=>{let n=t(e),o=t(r);if(null===n||null===o)return 0;let a=n.compareDocumentPosition(o);return a&Node.DOCUMENT_POSITION_FOLLOWING?-1:a&Node.DOCUMENT_POSITION_PRECEDING?1:0})}function h(e,t){return y(p(),t,{relativeTo:e})}function y(e,t,{sorted:r=!0,relativeTo:n=null,skipElements:o=[]}={}){var a,s,l;let i=Array.isArray(e)?e.length>0?e[0].ownerDocument:document:e.ownerDocument,c=Array.isArray(e)?r?g(e):e:64&t?function(e=document.body){return null==e?[]:Array.from(e.querySelectorAll(u)).sort((e,t)=>Math.sign((e.tabIndex||Number.MAX_SAFE_INTEGER)-(t.tabIndex||Number.MAX_SAFE_INTEGER)))}(e):p(e);o.length>0&&c.length>1&&(c=c.filter(e=>!o.some(t=>null!=t&&"current"in t?(null==t?void 0:t.current)===e:t===e))),n=null!=n?n:i.activeElement;let d=(()=>{if(5&t)return 1;if(10&t)return -1;throw Error("Missing Focus.First, Focus.Previous, Focus.Next or Focus.Last")})(),f=(()=>{if(1&t)return 0;if(2&t)return Math.max(0,c.indexOf(n))-1;if(4&t)return Math.max(0,c.indexOf(n))+1;if(8&t)return c.length-1;throw Error("Missing Focus.First, Focus.Previous, Focus.Next or Focus.Last")})(),m=32&t?{preventScroll:!0}:{},v=0,b=c.length,h;do{if(v>=b||v+b<=0)return 0;let e=f+v;if(16&t)e=(e+b)%b;else{if(e<0)return 3;if(e>=b)return 1}null==(h=c[e])||h.focus(m),v+=d}while(h!==i.activeElement)return 6&t&&null!=(l=null==(s=null==(a=h)?void 0:a.matches)?void 0:s.call(a,"textarea,input"))&&l&&h.select(),2}"u">typeof window&&"u">typeof document&&(document.addEventListener("keydown",e=>{e.metaKey||e.altKey||e.ctrlKey||(document.documentElement.dataset.headlessuiFocusVisible="")},!0),document.addEventListener("click",e=>{1===e.detail?delete document.documentElement.dataset.headlessuiFocusVisible:0===e.detail&&(document.documentElement.dataset.headlessuiFocusVisible="")},!0)),e.s(["Focus",()=>c,"FocusResult",()=>d,"FocusableMode",()=>m,"focusFrom",()=>h,"focusIn",()=>y,"getFocusableElements",()=>p,"isFocusableElement",()=>v,"sortByDomNode",()=>g])},963703,e=>{"use strict";var t=e.i(271645);let r=t.createContext(null);function n({children:e}){let n=t.useRef({groups:new Map,get(e,t){var r;let n=this.groups.get(e);n||(n=new Map,this.groups.set(e,n));let o=null!=(r=n.get(t))?r:0;return n.set(t,o+1),[Array.from(n.keys()).indexOf(t),function(){let e=n.get(t);e>1?n.set(t,e-1):n.delete(t)}]}});return t.createElement(r.Provider,{value:n},e)}function o(e){let n=t.useContext(r);if(!n)throw Error("You must wrap your component in a ");let o=t.useId(),[a,s]=n.current.get(e,o);return t.useEffect(()=>s,[]),a}e.s(["StableCollection",()=>n,"useStableCollectionIndex",()=>o])},998348,e=>{"use strict";let t;var r=((t=r||{}).Space=" ",t.Enter="Enter",t.Escape="Escape",t.Backspace="Backspace",t.Delete="Delete",t.ArrowLeft="ArrowLeft",t.ArrowUp="ArrowUp",t.ArrowRight="ArrowRight",t.ArrowDown="ArrowDown",t.Home="Home",t.End="End",t.PageUp="PageUp",t.PageDown="PageDown",t.Tab="Tab",t);e.s(["Keys",()=>r])},970554,e=>{"use strict";let t,r,n;var o=e.i(429427),a=e.i(371330),s=e.i(271645),l=e.i(394487),i=e.i(914189),u=e.i(835696),c=e.i(941444),d=e.i(144279),f=e.i(294316),p=e.i(640497),m=e.i(2788),v=e.i(652265),b=e.i(397701),g=e.i(368578),h=e.i(402155),y=e.i(700020),E=e.i(963703),w=e.i(998348),T=((t=T||{})[t.Forwards=0]="Forwards",t[t.Backwards=1]="Backwards",t),x=((r=x||{})[r.Less=-1]="Less",r[r.Equal=0]="Equal",r[r.Greater=1]="Greater",r),F=((n=F||{})[n.SetSelectedIndex=0]="SetSelectedIndex",n[n.RegisterTab=1]="RegisterTab",n[n.UnregisterTab=2]="UnregisterTab",n[n.RegisterPanel=3]="RegisterPanel",n[n.UnregisterPanel=4]="UnregisterPanel",n);let P={0(e,t){var r;let n=(0,v.sortByDomNode)(e.tabs,e=>e.current),o=(0,v.sortByDomNode)(e.panels,e=>e.current),a=n.filter(e=>{var t;return!(null!=(t=e.current)&&t.hasAttribute("disabled"))}),s={...e,tabs:n,panels:o};if(t.index<0||t.index>n.length-1){let r=(0,b.match)(Math.sign(t.index-e.selectedIndex),{[-1]:()=>1,0:()=>(0,b.match)(Math.sign(t.index),{[-1]:()=>0,0:()=>0,1:()=>1}),1:()=>0});if(0===a.length)return s;let o=(0,b.match)(r,{0:()=>n.indexOf(a[0]),1:()=>n.indexOf(a[a.length-1])});return{...s,selectedIndex:-1===o?e.selectedIndex:o}}let l=n.slice(0,t.index),i=[...n.slice(t.index),...l].find(e=>a.includes(e));if(!i)return s;let u=null!=(r=n.indexOf(i))?r:e.selectedIndex;return -1===u&&(u=e.selectedIndex),{...s,selectedIndex:u}},1(e,t){if(e.tabs.includes(t.tab))return e;let r=e.tabs[e.selectedIndex],n=(0,v.sortByDomNode)([...e.tabs,t.tab],e=>e.current),o=e.selectedIndex;return e.info.current.isControlled||-1===(o=n.indexOf(r))&&(o=e.selectedIndex),{...e,tabs:n,selectedIndex:o}},2:(e,t)=>({...e,tabs:e.tabs.filter(e=>e!==t.tab)}),3:(e,t)=>e.panels.includes(t.panel)?e:{...e,panels:(0,v.sortByDomNode)([...e.panels,t.panel],e=>e.current)},4:(e,t)=>({...e,panels:e.panels.filter(e=>e!==t.panel)})},k=(0,s.createContext)(null);function L(e){let t=(0,s.useContext)(k);if(null===t){let t=Error(`<${e} /> is missing a parent component.`);throw Error.captureStackTrace&&Error.captureStackTrace(t,L),t}return t}k.displayName="TabsDataContext";let N=(0,s.createContext)(null);function C(e){let t=(0,s.useContext)(N);if(null===t){let t=Error(`<${e} /> is missing a parent component.`);throw Error.captureStackTrace&&Error.captureStackTrace(t,C),t}return t}function I(e,t){return(0,b.match)(t.type,P,e,t)}N.displayName="TabsActionsContext";let S=y.RenderFeatures.RenderStrategy|y.RenderFeatures.Static,A=Object.assign((0,y.forwardRefWithAs)(function(e,t){var r,n;let c=(0,s.useId)(),{id:p=`headlessui-tabs-tab-${c}`,disabled:m=!1,autoFocus:T=!1,...x}=e,{orientation:F,activation:P,selectedIndex:k,tabs:N,panels:I}=L("Tab"),S=C("Tab"),A=L("Tab"),[M,R]=(0,s.useState)(null),O=(0,s.useRef)(null),D=(0,f.useSyncRefs)(O,t,R);(0,u.useIsoMorphicEffect)(()=>S.registerTab(O),[S,O]);let H=(0,E.useStableCollectionIndex)("tabs"),j=N.indexOf(O);-1===j&&(j=H);let K=j===k,W=(0,i.useEvent)(e=>{var t;let r=e();if(r===v.FocusResult.Success&&"auto"===P){let e=null==(t=(0,h.getOwnerDocument)(O))?void 0:t.activeElement,r=A.tabs.findIndex(t=>t.current===e);-1!==r&&S.change(r)}return r}),B=(0,i.useEvent)(e=>{let t=N.map(e=>e.current).filter(Boolean);if(e.key===w.Keys.Space||e.key===w.Keys.Enter){e.preventDefault(),e.stopPropagation(),S.change(j);return}switch(e.key){case w.Keys.Home:case w.Keys.PageUp:return e.preventDefault(),e.stopPropagation(),W(()=>(0,v.focusIn)(t,v.Focus.First));case w.Keys.End:case w.Keys.PageDown:return e.preventDefault(),e.stopPropagation(),W(()=>(0,v.focusIn)(t,v.Focus.Last))}if(W(()=>(0,b.match)(F,{vertical:()=>e.key===w.Keys.ArrowUp?(0,v.focusIn)(t,v.Focus.Previous|v.Focus.WrapAround):e.key===w.Keys.ArrowDown?(0,v.focusIn)(t,v.Focus.Next|v.Focus.WrapAround):v.FocusResult.Error,horizontal:()=>e.key===w.Keys.ArrowLeft?(0,v.focusIn)(t,v.Focus.Previous|v.Focus.WrapAround):e.key===w.Keys.ArrowRight?(0,v.focusIn)(t,v.Focus.Next|v.Focus.WrapAround):v.FocusResult.Error}))===v.FocusResult.Success)return e.preventDefault()}),V=(0,s.useRef)(!1),_=(0,i.useEvent)(()=>{var e;V.current||(V.current=!0,null==(e=O.current)||e.focus({preventScroll:!0}),S.change(j),(0,g.microTask)(()=>{V.current=!1}))}),G=(0,i.useEvent)(e=>{e.preventDefault()}),{isFocusVisible:U,focusProps:$}=(0,o.useFocusRing)({autoFocus:T}),{isHovered:q,hoverProps:X}=(0,a.useHover)({isDisabled:m}),{pressed:Y,pressProps:z}=(0,l.useActivePress)({disabled:m}),Z=(0,s.useMemo)(()=>({selected:K,hover:q,active:Y,focus:U,autofocus:T,disabled:m}),[K,q,U,Y,T,m]),J=(0,y.mergeProps)({ref:D,onKeyDown:B,onMouseDown:G,onClick:_,id:p,role:"tab",type:(0,d.useResolveButtonType)(e,M),"aria-controls":null==(n=null==(r=I[j])?void 0:r.current)?void 0:n.id,"aria-selected":K,tabIndex:K?0:-1,disabled:m||void 0,autoFocus:T},$,X,z);return(0,y.useRender)()({ourProps:J,theirProps:x,slot:Z,defaultTag:"button",name:"Tabs.Tab"})}),{Group:(0,y.forwardRefWithAs)(function(e,t){let{defaultIndex:r=0,vertical:n=!1,manual:o=!1,onChange:a,selectedIndex:l=null,...d}=e,m=n?"vertical":"horizontal",b=o?"manual":"auto",g=null!==l,h=(0,c.useLatestValue)({isControlled:g}),w=(0,f.useSyncRefs)(t),[T,x]=(0,s.useReducer)(I,{info:h,selectedIndex:null!=l?l:r,tabs:[],panels:[]}),F=(0,s.useMemo)(()=>({selectedIndex:T.selectedIndex}),[T.selectedIndex]),P=(0,c.useLatestValue)(a||(()=>{})),L=(0,c.useLatestValue)(T.tabs),C=(0,s.useMemo)(()=>({orientation:m,activation:b,...T}),[m,b,T]),S=(0,i.useEvent)(e=>(x({type:1,tab:e}),()=>x({type:2,tab:e}))),A=(0,i.useEvent)(e=>(x({type:3,panel:e}),()=>x({type:4,panel:e}))),M=(0,i.useEvent)(e=>{R.current!==e&&P.current(e),g||x({type:0,index:e})}),R=(0,c.useLatestValue)(g?e.selectedIndex:T.selectedIndex),O=(0,s.useMemo)(()=>({registerTab:S,registerPanel:A,change:M}),[]);(0,u.useIsoMorphicEffect)(()=>{x({type:0,index:null!=l?l:r})},[l]),(0,u.useIsoMorphicEffect)(()=>{if(void 0===R.current||T.tabs.length<=0)return;let e=(0,v.sortByDomNode)(T.tabs,e=>e.current);e.some((e,t)=>T.tabs[t]!==e)&&M(e.indexOf(T.tabs[R.current]))});let D=(0,y.useRender)();return s.default.createElement(E.StableCollection,null,s.default.createElement(N.Provider,{value:O},s.default.createElement(k.Provider,{value:C},C.tabs.length<=0&&s.default.createElement(p.FocusSentinel,{onFocus:()=>{var e,t;for(let r of L.current)if((null==(e=r.current)?void 0:e.tabIndex)===0)return null==(t=r.current)||t.focus(),!0;return!1}}),D({ourProps:{ref:w},theirProps:d,slot:F,defaultTag:"div",name:"Tabs"}))))}),List:(0,y.forwardRefWithAs)(function(e,t){let{orientation:r,selectedIndex:n}=L("Tab.List"),o=(0,f.useSyncRefs)(t),a=(0,s.useMemo)(()=>({selectedIndex:n}),[n]);return(0,y.useRender)()({ourProps:{ref:o,role:"tablist","aria-orientation":r},theirProps:e,slot:a,defaultTag:"div",name:"Tabs.List"})}),Panels:(0,y.forwardRefWithAs)(function(e,t){let{selectedIndex:r}=L("Tab.Panels"),n=(0,f.useSyncRefs)(t),o=(0,s.useMemo)(()=>({selectedIndex:r}),[r]);return(0,y.useRender)()({ourProps:{ref:n},theirProps:e,slot:o,defaultTag:"div",name:"Tabs.Panels"})}),Panel:(0,y.forwardRefWithAs)(function(e,t){var r,n,a,l;let i=(0,s.useId)(),{id:c=`headlessui-tabs-panel-${i}`,tabIndex:d=0,...p}=e,{selectedIndex:v,tabs:b,panels:g}=L("Tab.Panel"),h=C("Tab.Panel"),w=(0,s.useRef)(null),T=(0,f.useSyncRefs)(w,t);(0,u.useIsoMorphicEffect)(()=>h.registerPanel(w),[h,w]);let x=(0,E.useStableCollectionIndex)("panels"),F=g.indexOf(w);-1===F&&(F=x);let P=F===v,{isFocusVisible:k,focusProps:N}=(0,o.useFocusRing)(),I=(0,s.useMemo)(()=>({selected:P,focus:k}),[P,k]),A=(0,y.mergeProps)({ref:T,id:c,role:"tabpanel","aria-labelledby":null==(n=null==(r=b[F])?void 0:r.current)?void 0:n.id,tabIndex:P?d:-1},N),M=(0,y.useRender)();return P||null!=(a=p.unmount)&&!a||null!=(l=p.static)&&l?M({ourProps:A,theirProps:p,slot:I,defaultTag:"div",features:S,visible:P,name:"Tabs.Panel"}):s.default.createElement(m.Hidden,{"aria-hidden":"true",...A})})});e.s(["Tab",()=>A])},653824,e=>{"use strict";var t=e.i(290571),r=e.i(970554),n=e.i(444755),o=e.i(673706),a=e.i(271645);let s=(0,o.makeClassName)("TabGroup"),l=a.default.forwardRef((e,o)=>{let{defaultIndex:l,index:i,onIndexChange:u,children:c,className:d}=e,f=(0,t.__rest)(e,["defaultIndex","index","onIndexChange","children","className"]);return a.default.createElement(r.Tab.Group,Object.assign({as:"div",ref:o,defaultIndex:l,selectedIndex:i,onChange:u,className:(0,n.tremorTwMerge)(s("root"),"w-full",d)},f),c)});l.displayName="TabGroup",e.s(["TabGroup",()=>l],653824)},405371,910342,e=>{"use strict";var t=e.i(290571),r=e.i(271645),n=e.i(480731);let o=(0,r.createContext)(n.BaseColors.Blue);e.s(["default",()=>o],910342);var a=e.i(970554),s=e.i(444755);let l=(0,e.i(673706).makeClassName)("TabList"),i=(0,r.createContext)("line"),u={line:(0,s.tremorTwMerge)("flex border-b space-x-4","border-tremor-border","dark:border-dark-tremor-border"),solid:(0,s.tremorTwMerge)("inline-flex p-0.5 rounded-tremor-default space-x-1.5","bg-tremor-background-subtle","dark:bg-dark-tremor-background-subtle")},c=r.default.forwardRef((e,n)=>{let{color:c,variant:d="line",children:f,className:p}=e,m=(0,t.__rest)(e,["color","variant","children","className"]);return r.default.createElement(a.Tab.List,Object.assign({ref:n,className:(0,s.tremorTwMerge)(l("root"),"justify-start overflow-x-clip",u[d],p)},m),r.default.createElement(i.Provider,{value:d},r.default.createElement(o.Provider,{value:c},f)))});c.displayName="TabList",e.s(["TabVariantContext",()=>i,"default",()=>c],405371)},881073,e=>{"use strict";var t=e.i(405371);e.s(["TabList",()=>t.default])},197647,e=>{"use strict";var t=e.i(290571),r=e.i(970554),n=e.i(95779),o=e.i(444755),a=e.i(673706),s=e.i(271645),l=e.i(405371),i=e.i(910342);let u=(0,a.makeClassName)("Tab"),c=s.default.forwardRef((e,c)=>{let{icon:d,className:f,children:p}=e,m=(0,t.__rest)(e,["icon","className","children"]),v=(0,s.useContext)(l.TabVariantContext),b=(0,s.useContext)(i.default);return s.default.createElement(r.Tab,Object.assign({ref:c,className:(0,o.tremorTwMerge)(u("root"),"flex whitespace-nowrap truncate max-w-xs outline-none data-focus-visible:ring text-tremor-default transition duration-100",function(e,t){switch(e){case"line":return(0,o.tremorTwMerge)("data-[selected]:border-b-2 hover:border-b-2 border-transparent transition duration-100 -mb-px px-2 py-2","hover:border-tremor-content hover:text-tremor-content-emphasis text-tremor-content","[&:not([data-selected])]:dark:hover:border-dark-tremor-content-emphasis [&:not([data-selected])]:dark:hover:text-dark-tremor-content-emphasis [&:not([data-selected])]:dark:text-dark-tremor-content",t?(0,a.getColorClassNames)(t,n.colorPalette.border).selectBorderColor:["data-[selected]:border-tremor-brand data-[selected]:text-tremor-brand","data-[selected]:dark:border-dark-tremor-brand data-[selected]:dark:text-dark-tremor-brand"]);case"solid":return(0,o.tremorTwMerge)("border-transparent border rounded-tremor-small px-2.5 py-1","data-[selected]:border-tremor-border data-[selected]:bg-tremor-background data-[selected]:shadow-tremor-input [&:not([data-selected])]:hover:text-tremor-content-emphasis data-[selected]:text-tremor-brand [&:not([data-selected])]:text-tremor-content","dark:data-[selected]:border-dark-tremor-border dark:data-[selected]:bg-dark-tremor-background dark:data-[selected]:shadow-dark-tremor-input dark:[&:not([data-selected])]:hover:text-dark-tremor-content-emphasis dark:data-[selected]:text-dark-tremor-brand dark:[&:not([data-selected])]:text-dark-tremor-content",t?(0,a.getColorClassNames)(t,n.colorPalette.text).selectTextColor:"text-tremor-content dark:text-dark-tremor-content")}}(v,b),f,b&&(0,a.getColorClassNames)(b,n.colorPalette.text).selectTextColor)},m),d?s.default.createElement(d,{className:(0,o.tremorTwMerge)(u("icon"),"flex-none h-5 w-5",p?"mr-2":"")}):null,p?s.default.createElement("span",null,p):null)});c.displayName="Tab",e.s(["Tab",()=>c],197647)},751734,e=>{"use strict";let t=(0,e.i(271645).createContext)(0);e.s(["default",()=>t])},144582,e=>{"use strict";let t=(0,e.i(271645).createContext)({selectedValue:void 0,handleValueChange:void 0});e.s(["default",()=>t])},723731,e=>{"use strict";var t=e.i(290571),r=e.i(970554),n=e.i(751734),o=e.i(144582),a=e.i(444755),s=e.i(673706),l=e.i(271645);let i=(0,s.makeClassName)("TabPanels"),u=l.default.forwardRef((e,s)=>{let{children:u,className:c}=e,d=(0,t.__rest)(e,["children","className"]);return l.default.createElement(r.Tab.Panels,Object.assign({as:"div",ref:s,className:(0,a.tremorTwMerge)(i("root"),"w-full",c)},d),({selectedIndex:e})=>l.default.createElement(o.default.Provider,{value:{selectedValue:e}},l.default.Children.map(u,(e,t)=>l.default.createElement(n.default.Provider,{value:t},e))))});u.displayName="TabPanels",e.s(["TabPanels",()=>u],723731)},404206,e=>{"use strict";var t=e.i(290571),r=e.i(751734),n=e.i(144582),o=e.i(444755),a=e.i(673706),s=e.i(271645);let l=(0,a.makeClassName)("TabPanel"),i=s.default.forwardRef((e,a)=>{let{children:i,className:u}=e,c=(0,t.__rest)(e,["children","className"]),{selectedValue:d}=(0,s.useContext)(n.default),f=d===(0,s.useContext)(r.default);return s.default.createElement("div",Object.assign({ref:a,className:(0,o.tremorTwMerge)(l("root"),"w-full mt-2",f?"":"hidden",u),"aria-selected":f?"true":"false"},c),i)});i.displayName="TabPanel",e.s(["TabPanel",()=>i],404206)}]); \ No newline at end of file +`))}else{var h;let e=b.props,t=null==e?void 0:e.className,r="function"==typeof t?(...e)=>(0,o.classNames)(t(...e),f.className):(0,o.classNames)(t,f.className),a=d(b.props,m(v(f,["ref"])));for(let e in g)e in a&&delete g[e];return(0,n.cloneElement)(b,Object.assign({},a,g,p,{ref:s((h=b,n.default.version.split(".")[0]>="19"?h.props.ref:h.ref),p.ref)},r?{className:r}:{}))}return(0,n.createElement)(l,Object.assign({},v(f,["ref"]),l!==n.Fragment&&p,l!==n.Fragment&&g),b)}function c(...e){return e.every(e=>null==e)?void 0:t=>{for(let r of e)null!=r&&("function"==typeof r?r(t):r.current=t)}}function d(...e){if(0===e.length)return{};if(1===e.length)return e[0];let t={},r={};for(let n of e)for(let e in n)e.startsWith("on")&&"function"==typeof n[e]?(null!=r[e]||(r[e]=[]),r[e].push(n[e])):t[e]=n[e];if(t.disabled||t["aria-disabled"])for(let e in r)/^(on(?:Click|Pointer|Mouse|Key)(?:Down|Up|Press)?)$/.test(e)&&(r[e]=[e=>{var t;return null==(t=null==e?void 0:e.preventDefault)?void 0:t.call(e)}]);for(let e in r)Object.assign(t,{[e](t,...n){for(let o of r[e]){if((t instanceof Event||(null==t?void 0:t.nativeEvent)instanceof Event)&&t.defaultPrevented)return;o(t,...n)}}});return t}function f(...e){if(0===e.length)return{};if(1===e.length)return e[0];let t={},r={};for(let n of e)for(let e in n)e.startsWith("on")&&"function"==typeof n[e]?(null!=r[e]||(r[e]=[]),r[e].push(n[e])):t[e]=n[e];for(let e in r)Object.assign(t,{[e](...t){for(let n of r[e])null==n||n(...t)}});return t}function p(e){var t;return Object.assign((0,n.forwardRef)(e),{displayName:null!=(t=e.displayName)?t:e.name})}function m(e){let t=Object.assign({},e);for(let e in t)void 0===t[e]&&delete t[e];return t}function v(e,t=[]){let r=Object.assign({},e);for(let e of t)e in r&&delete r[e];return r}e.s(["RenderFeatures",()=>s,"RenderStrategy",()=>l,"compact",()=>m,"forwardRefWithAs",()=>p,"mergeProps",()=>f,"useRender",()=>i])},2788,e=>{"use strict";let t;var r=e.i(700020),n=((t=n||{})[t.None=1]="None",t[t.Focusable=2]="Focusable",t[t.Hidden=4]="Hidden",t);let o=(0,r.forwardRefWithAs)(function(e,t){var n;let{features:o=1,...a}=e,s={ref:t,"aria-hidden":(2&o)==2||(null!=(n=a["aria-hidden"])?n:void 0),hidden:(4&o)==4||void 0,style:{position:"fixed",top:1,left:1,width:1,height:0,padding:0,margin:-1,overflow:"hidden",clip:"rect(0, 0, 0, 0)",whiteSpace:"nowrap",borderWidth:"0",...(4&o)==4&&(2&o)!=2&&{display:"none"}}};return(0,r.useRender)()({ourProps:s,theirProps:a,slot:{},defaultTag:"span",name:"Hidden"})});e.s(["Hidden",()=>o,"HiddenFeatures",()=>n])},640497,e=>{"use strict";var t=e.i(271645),r=e.i(553521),n=e.i(2788);function o({onFocus:e}){let[o,a]=(0,t.useState)(!0),s=(0,r.useIsMounted)();return o?t.default.createElement(n.Hidden,{as:"button",type:"button",features:n.HiddenFeatures.Focusable,onFocus:t=>{t.preventDefault();let r,n=50;r=requestAnimationFrame(function t(){if(n--<=0){r&&cancelAnimationFrame(r);return}if(e()){if(cancelAnimationFrame(r),!s.current)return;a(!1);return}r=requestAnimationFrame(t)})}}):null}e.s(["FocusSentinel",()=>o])},652265,e=>{"use strict";let t,r,n,o,a;e.i(544508);var s=e.i(397701),l=e.i(402155);let i=["[contentEditable=true]","[tabindex]","a[href]","area[href]","button:not([disabled])","iframe","input:not([disabled])","select:not([disabled])","textarea:not([disabled])"].map(e=>`${e}:not([tabindex='-1'])`).join(","),u=["[data-autofocus]"].map(e=>`${e}:not([tabindex='-1'])`).join(",");var c=((t=c||{})[t.First=1]="First",t[t.Previous=2]="Previous",t[t.Next=4]="Next",t[t.Last=8]="Last",t[t.WrapAround=16]="WrapAround",t[t.NoScroll=32]="NoScroll",t[t.AutoFocus=64]="AutoFocus",t),d=((r=d||{})[r.Error=0]="Error",r[r.Overflow=1]="Overflow",r[r.Success=2]="Success",r[r.Underflow=3]="Underflow",r),f=((n=f||{})[n.Previous=-1]="Previous",n[n.Next=1]="Next",n);function p(e=document.body){return null==e?[]:Array.from(e.querySelectorAll(i)).sort((e,t)=>Math.sign((e.tabIndex||Number.MAX_SAFE_INTEGER)-(t.tabIndex||Number.MAX_SAFE_INTEGER)))}var m=((o=m||{})[o.Strict=0]="Strict",o[o.Loose=1]="Loose",o);function v(e,t=0){var r;return e!==(null==(r=(0,l.getOwnerDocument)(e))?void 0:r.body)&&(0,s.match)(t,{0:()=>e.matches(i),1(){let t=e;for(;null!==t;){if(t.matches(i))return!0;t=t.parentElement}return!1}})}var b=((a=b||{})[a.Keyboard=0]="Keyboard",a[a.Mouse=1]="Mouse",a);function g(e,t=e=>e){return e.slice().sort((e,r)=>{let n=t(e),o=t(r);if(null===n||null===o)return 0;let a=n.compareDocumentPosition(o);return a&Node.DOCUMENT_POSITION_FOLLOWING?-1:a&Node.DOCUMENT_POSITION_PRECEDING?1:0})}function h(e,t){return y(p(),t,{relativeTo:e})}function y(e,t,{sorted:r=!0,relativeTo:n=null,skipElements:o=[]}={}){var a,s,l;let i=Array.isArray(e)?e.length>0?e[0].ownerDocument:document:e.ownerDocument,c=Array.isArray(e)?r?g(e):e:64&t?function(e=document.body){return null==e?[]:Array.from(e.querySelectorAll(u)).sort((e,t)=>Math.sign((e.tabIndex||Number.MAX_SAFE_INTEGER)-(t.tabIndex||Number.MAX_SAFE_INTEGER)))}(e):p(e);o.length>0&&c.length>1&&(c=c.filter(e=>!o.some(t=>null!=t&&"current"in t?(null==t?void 0:t.current)===e:t===e))),n=null!=n?n:i.activeElement;let d=(()=>{if(5&t)return 1;if(10&t)return -1;throw Error("Missing Focus.First, Focus.Previous, Focus.Next or Focus.Last")})(),f=(()=>{if(1&t)return 0;if(2&t)return Math.max(0,c.indexOf(n))-1;if(4&t)return Math.max(0,c.indexOf(n))+1;if(8&t)return c.length-1;throw Error("Missing Focus.First, Focus.Previous, Focus.Next or Focus.Last")})(),m=32&t?{preventScroll:!0}:{},v=0,b=c.length,h;do{if(v>=b||v+b<=0)return 0;let e=f+v;if(16&t)e=(e+b)%b;else{if(e<0)return 3;if(e>=b)return 1}null==(h=c[e])||h.focus(m),v+=d}while(h!==i.activeElement)return 6&t&&null!=(l=null==(s=null==(a=h)?void 0:a.matches)?void 0:s.call(a,"textarea,input"))&&l&&h.select(),2}"u">typeof window&&"u">typeof document&&(document.addEventListener("keydown",e=>{e.metaKey||e.altKey||e.ctrlKey||(document.documentElement.dataset.headlessuiFocusVisible="")},!0),document.addEventListener("click",e=>{1===e.detail?delete document.documentElement.dataset.headlessuiFocusVisible:0===e.detail&&(document.documentElement.dataset.headlessuiFocusVisible="")},!0)),e.s(["Focus",()=>c,"FocusResult",()=>d,"FocusableMode",()=>m,"focusFrom",()=>h,"focusIn",()=>y,"getFocusableElements",()=>p,"isFocusableElement",()=>v,"sortByDomNode",()=>g])},963703,e=>{"use strict";var t=e.i(271645);let r=t.createContext(null);function n({children:e}){let n=t.useRef({groups:new Map,get(e,t){var r;let n=this.groups.get(e);n||(n=new Map,this.groups.set(e,n));let o=null!=(r=n.get(t))?r:0;return n.set(t,o+1),[Array.from(n.keys()).indexOf(t),function(){let e=n.get(t);e>1?n.set(t,e-1):n.delete(t)}]}});return t.createElement(r.Provider,{value:n},e)}function o(e){let n=t.useContext(r);if(!n)throw Error("You must wrap your component in a ");let o=t.useId(),[a,s]=n.current.get(e,o);return t.useEffect(()=>s,[]),a}e.s(["StableCollection",()=>n,"useStableCollectionIndex",()=>o])},998348,e=>{"use strict";let t;var r=((t=r||{}).Space=" ",t.Enter="Enter",t.Escape="Escape",t.Backspace="Backspace",t.Delete="Delete",t.ArrowLeft="ArrowLeft",t.ArrowUp="ArrowUp",t.ArrowRight="ArrowRight",t.ArrowDown="ArrowDown",t.Home="Home",t.End="End",t.PageUp="PageUp",t.PageDown="PageDown",t.Tab="Tab",t);e.s(["Keys",()=>r])},970554,e=>{"use strict";let t,r,n;var o=e.i(429427),a=e.i(371330),s=e.i(271645),l=e.i(394487),i=e.i(914189),u=e.i(835696),c=e.i(941444),d=e.i(144279),f=e.i(294316),p=e.i(640497),m=e.i(2788),v=e.i(652265),b=e.i(397701),g=e.i(368578),h=e.i(402155),y=e.i(700020),E=e.i(963703),w=e.i(998348),T=((t=T||{})[t.Forwards=0]="Forwards",t[t.Backwards=1]="Backwards",t),x=((r=x||{})[r.Less=-1]="Less",r[r.Equal=0]="Equal",r[r.Greater=1]="Greater",r),F=((n=F||{})[n.SetSelectedIndex=0]="SetSelectedIndex",n[n.RegisterTab=1]="RegisterTab",n[n.UnregisterTab=2]="UnregisterTab",n[n.RegisterPanel=3]="RegisterPanel",n[n.UnregisterPanel=4]="UnregisterPanel",n);let P={0(e,t){var r;let n=(0,v.sortByDomNode)(e.tabs,e=>e.current),o=(0,v.sortByDomNode)(e.panels,e=>e.current),a=n.filter(e=>{var t;return!(null!=(t=e.current)&&t.hasAttribute("disabled"))}),s={...e,tabs:n,panels:o};if(t.index<0||t.index>n.length-1){let r=(0,b.match)(Math.sign(t.index-e.selectedIndex),{[-1]:()=>1,0:()=>(0,b.match)(Math.sign(t.index),{[-1]:()=>0,0:()=>0,1:()=>1}),1:()=>0});if(0===a.length)return s;let o=(0,b.match)(r,{0:()=>n.indexOf(a[0]),1:()=>n.indexOf(a[a.length-1])});return{...s,selectedIndex:-1===o?e.selectedIndex:o}}let l=n.slice(0,t.index),i=[...n.slice(t.index),...l].find(e=>a.includes(e));if(!i)return s;let u=null!=(r=n.indexOf(i))?r:e.selectedIndex;return -1===u&&(u=e.selectedIndex),{...s,selectedIndex:u}},1(e,t){if(e.tabs.includes(t.tab))return e;let r=e.tabs[e.selectedIndex],n=(0,v.sortByDomNode)([...e.tabs,t.tab],e=>e.current),o=e.selectedIndex;return e.info.current.isControlled||-1===(o=n.indexOf(r))&&(o=e.selectedIndex),{...e,tabs:n,selectedIndex:o}},2:(e,t)=>({...e,tabs:e.tabs.filter(e=>e!==t.tab)}),3:(e,t)=>e.panels.includes(t.panel)?e:{...e,panels:(0,v.sortByDomNode)([...e.panels,t.panel],e=>e.current)},4:(e,t)=>({...e,panels:e.panels.filter(e=>e!==t.panel)})},k=(0,s.createContext)(null);function L(e){let t=(0,s.useContext)(k);if(null===t){let t=Error(`<${e} /> is missing a parent component.`);throw Error.captureStackTrace&&Error.captureStackTrace(t,L),t}return t}k.displayName="TabsDataContext";let N=(0,s.createContext)(null);function C(e){let t=(0,s.useContext)(N);if(null===t){let t=Error(`<${e} /> is missing a parent component.`);throw Error.captureStackTrace&&Error.captureStackTrace(t,C),t}return t}function I(e,t){return(0,b.match)(t.type,P,e,t)}N.displayName="TabsActionsContext";let S=y.RenderFeatures.RenderStrategy|y.RenderFeatures.Static,A=Object.assign((0,y.forwardRefWithAs)(function(e,t){var r,n;let c=(0,s.useId)(),{id:p=`headlessui-tabs-tab-${c}`,disabled:m=!1,autoFocus:T=!1,...x}=e,{orientation:F,activation:P,selectedIndex:k,tabs:N,panels:I}=L("Tab"),S=C("Tab"),A=L("Tab"),[M,R]=(0,s.useState)(null),O=(0,s.useRef)(null),D=(0,f.useSyncRefs)(O,t,R);(0,u.useIsoMorphicEffect)(()=>S.registerTab(O),[S,O]);let H=(0,E.useStableCollectionIndex)("tabs"),j=N.indexOf(O);-1===j&&(j=H);let K=j===k,W=(0,i.useEvent)(e=>{var t;let r=e();if(r===v.FocusResult.Success&&"auto"===P){let e=null==(t=(0,h.getOwnerDocument)(O))?void 0:t.activeElement,r=A.tabs.findIndex(t=>t.current===e);-1!==r&&S.change(r)}return r}),B=(0,i.useEvent)(e=>{let t=N.map(e=>e.current).filter(Boolean);if(e.key===w.Keys.Space||e.key===w.Keys.Enter){e.preventDefault(),e.stopPropagation(),S.change(j);return}switch(e.key){case w.Keys.Home:case w.Keys.PageUp:return e.preventDefault(),e.stopPropagation(),W(()=>(0,v.focusIn)(t,v.Focus.First));case w.Keys.End:case w.Keys.PageDown:return e.preventDefault(),e.stopPropagation(),W(()=>(0,v.focusIn)(t,v.Focus.Last))}if(W(()=>(0,b.match)(F,{vertical:()=>e.key===w.Keys.ArrowUp?(0,v.focusIn)(t,v.Focus.Previous|v.Focus.WrapAround):e.key===w.Keys.ArrowDown?(0,v.focusIn)(t,v.Focus.Next|v.Focus.WrapAround):v.FocusResult.Error,horizontal:()=>e.key===w.Keys.ArrowLeft?(0,v.focusIn)(t,v.Focus.Previous|v.Focus.WrapAround):e.key===w.Keys.ArrowRight?(0,v.focusIn)(t,v.Focus.Next|v.Focus.WrapAround):v.FocusResult.Error}))===v.FocusResult.Success)return e.preventDefault()}),V=(0,s.useRef)(!1),_=(0,i.useEvent)(()=>{var e;V.current||(V.current=!0,null==(e=O.current)||e.focus({preventScroll:!0}),S.change(j),(0,g.microTask)(()=>{V.current=!1}))}),G=(0,i.useEvent)(e=>{e.preventDefault()}),{isFocusVisible:U,focusProps:$}=(0,o.useFocusRing)({autoFocus:T}),{isHovered:q,hoverProps:X}=(0,a.useHover)({isDisabled:m}),{pressed:Y,pressProps:z}=(0,l.useActivePress)({disabled:m}),Z=(0,s.useMemo)(()=>({selected:K,hover:q,active:Y,focus:U,autofocus:T,disabled:m}),[K,q,U,Y,T,m]),J=(0,y.mergeProps)({ref:D,onKeyDown:B,onMouseDown:G,onClick:_,id:p,role:"tab",type:(0,d.useResolveButtonType)(e,M),"aria-controls":null==(n=null==(r=I[j])?void 0:r.current)?void 0:n.id,"aria-selected":K,tabIndex:K?0:-1,disabled:m||void 0,autoFocus:T},$,X,z);return(0,y.useRender)()({ourProps:J,theirProps:x,slot:Z,defaultTag:"button",name:"Tabs.Tab"})}),{Group:(0,y.forwardRefWithAs)(function(e,t){let{defaultIndex:r=0,vertical:n=!1,manual:o=!1,onChange:a,selectedIndex:l=null,...d}=e,m=n?"vertical":"horizontal",b=o?"manual":"auto",g=null!==l,h=(0,c.useLatestValue)({isControlled:g}),w=(0,f.useSyncRefs)(t),[T,x]=(0,s.useReducer)(I,{info:h,selectedIndex:null!=l?l:r,tabs:[],panels:[]}),F=(0,s.useMemo)(()=>({selectedIndex:T.selectedIndex}),[T.selectedIndex]),P=(0,c.useLatestValue)(a||(()=>{})),L=(0,c.useLatestValue)(T.tabs),C=(0,s.useMemo)(()=>({orientation:m,activation:b,...T}),[m,b,T]),S=(0,i.useEvent)(e=>(x({type:1,tab:e}),()=>x({type:2,tab:e}))),A=(0,i.useEvent)(e=>(x({type:3,panel:e}),()=>x({type:4,panel:e}))),M=(0,i.useEvent)(e=>{R.current!==e&&P.current(e),g||x({type:0,index:e})}),R=(0,c.useLatestValue)(g?e.selectedIndex:T.selectedIndex),O=(0,s.useMemo)(()=>({registerTab:S,registerPanel:A,change:M}),[]);(0,u.useIsoMorphicEffect)(()=>{x({type:0,index:null!=l?l:r})},[l]),(0,u.useIsoMorphicEffect)(()=>{if(void 0===R.current||T.tabs.length<=0)return;let e=(0,v.sortByDomNode)(T.tabs,e=>e.current);e.some((e,t)=>T.tabs[t]!==e)&&M(e.indexOf(T.tabs[R.current]))});let D=(0,y.useRender)();return s.default.createElement(E.StableCollection,null,s.default.createElement(N.Provider,{value:O},s.default.createElement(k.Provider,{value:C},C.tabs.length<=0&&s.default.createElement(p.FocusSentinel,{onFocus:()=>{var e,t;for(let r of L.current)if((null==(e=r.current)?void 0:e.tabIndex)===0)return null==(t=r.current)||t.focus(),!0;return!1}}),D({ourProps:{ref:w},theirProps:d,slot:F,defaultTag:"div",name:"Tabs"}))))}),List:(0,y.forwardRefWithAs)(function(e,t){let{orientation:r,selectedIndex:n}=L("Tab.List"),o=(0,f.useSyncRefs)(t),a=(0,s.useMemo)(()=>({selectedIndex:n}),[n]);return(0,y.useRender)()({ourProps:{ref:o,role:"tablist","aria-orientation":r},theirProps:e,slot:a,defaultTag:"div",name:"Tabs.List"})}),Panels:(0,y.forwardRefWithAs)(function(e,t){let{selectedIndex:r}=L("Tab.Panels"),n=(0,f.useSyncRefs)(t),o=(0,s.useMemo)(()=>({selectedIndex:r}),[r]);return(0,y.useRender)()({ourProps:{ref:n},theirProps:e,slot:o,defaultTag:"div",name:"Tabs.Panels"})}),Panel:(0,y.forwardRefWithAs)(function(e,t){var r,n,a,l;let i=(0,s.useId)(),{id:c=`headlessui-tabs-panel-${i}`,tabIndex:d=0,...p}=e,{selectedIndex:v,tabs:b,panels:g}=L("Tab.Panel"),h=C("Tab.Panel"),w=(0,s.useRef)(null),T=(0,f.useSyncRefs)(w,t);(0,u.useIsoMorphicEffect)(()=>h.registerPanel(w),[h,w]);let x=(0,E.useStableCollectionIndex)("panels"),F=g.indexOf(w);-1===F&&(F=x);let P=F===v,{isFocusVisible:k,focusProps:N}=(0,o.useFocusRing)(),I=(0,s.useMemo)(()=>({selected:P,focus:k}),[P,k]),A=(0,y.mergeProps)({ref:T,id:c,role:"tabpanel","aria-labelledby":null==(n=null==(r=b[F])?void 0:r.current)?void 0:n.id,tabIndex:P?d:-1},N),M=(0,y.useRender)();return P||null!=(a=p.unmount)&&!a||null!=(l=p.static)&&l?M({ourProps:A,theirProps:p,slot:I,defaultTag:"div",features:S,visible:P,name:"Tabs.Panel"}):s.default.createElement(m.Hidden,{"aria-hidden":"true",...A})})});e.s(["Tab",()=>A])},653824,e=>{"use strict";var t=e.i(290571),r=e.i(970554),n=e.i(444755),o=e.i(673706),a=e.i(271645);let s=(0,o.makeClassName)("TabGroup"),l=a.default.forwardRef((e,o)=>{let{defaultIndex:l,index:i,onIndexChange:u,children:c,className:d}=e,f=(0,t.__rest)(e,["defaultIndex","index","onIndexChange","children","className"]);return a.default.createElement(r.Tab.Group,Object.assign({as:"div",ref:o,defaultIndex:l,selectedIndex:i,onChange:u,className:(0,n.tremorTwMerge)(s("root"),"w-full",d)},f),c)});l.displayName="TabGroup",e.s(["TabGroup",()=>l],653824)},405371,910342,e=>{"use strict";var t=e.i(290571),r=e.i(271645),n=e.i(480731);let o=(0,r.createContext)(n.BaseColors.Blue);e.s(["default",()=>o],910342);var a=e.i(970554),s=e.i(444755);let l=(0,e.i(673706).makeClassName)("TabList"),i=(0,r.createContext)("line"),u={line:(0,s.tremorTwMerge)("flex border-b space-x-4","border-tremor-border","dark:border-dark-tremor-border"),solid:(0,s.tremorTwMerge)("inline-flex p-0.5 rounded-tremor-default space-x-1.5","bg-tremor-background-subtle","dark:bg-dark-tremor-background-subtle")},c=r.default.forwardRef((e,n)=>{let{color:c,variant:d="line",children:f,className:p}=e,m=(0,t.__rest)(e,["color","variant","children","className"]);return r.default.createElement(a.Tab.List,Object.assign({ref:n,className:(0,s.tremorTwMerge)(l("root"),"justify-start overflow-x-clip",u[d],p)},m),r.default.createElement(i.Provider,{value:d},r.default.createElement(o.Provider,{value:c},f)))});c.displayName="TabList",e.s(["TabVariantContext",()=>i,"default",()=>c],405371)},881073,e=>{"use strict";var t=e.i(405371);e.s(["TabList",()=>t.default])},197647,e=>{"use strict";var t=e.i(290571),r=e.i(970554),n=e.i(95779),o=e.i(444755),a=e.i(673706),s=e.i(271645),l=e.i(405371),i=e.i(910342);let u=(0,a.makeClassName)("Tab"),c=s.default.forwardRef((e,c)=>{let{icon:d,className:f,children:p}=e,m=(0,t.__rest)(e,["icon","className","children"]),v=(0,s.useContext)(l.TabVariantContext),b=(0,s.useContext)(i.default);return s.default.createElement(r.Tab,Object.assign({ref:c,className:(0,o.tremorTwMerge)(u("root"),"flex whitespace-nowrap truncate max-w-xs outline-none data-focus-visible:ring text-tremor-default transition duration-100",function(e,t){switch(e){case"line":return(0,o.tremorTwMerge)("data-[selected]:border-b-2 hover:border-b-2 border-transparent transition duration-100 -mb-px px-2 py-2","hover:border-tremor-content hover:text-tremor-content-emphasis text-tremor-content","[&:not([data-selected])]:dark:hover:border-dark-tremor-content-emphasis [&:not([data-selected])]:dark:hover:text-dark-tremor-content-emphasis [&:not([data-selected])]:dark:text-dark-tremor-content",t?(0,a.getColorClassNames)(t,n.colorPalette.border).selectBorderColor:["data-[selected]:border-tremor-brand data-[selected]:text-tremor-brand","data-[selected]:dark:border-dark-tremor-brand data-[selected]:dark:text-dark-tremor-brand"]);case"solid":return(0,o.tremorTwMerge)("border-transparent border rounded-tremor-small px-2.5 py-1","data-[selected]:border-tremor-border data-[selected]:bg-tremor-background data-[selected]:shadow-tremor-input [&:not([data-selected])]:hover:text-tremor-content-emphasis data-[selected]:text-tremor-brand [&:not([data-selected])]:text-tremor-content","dark:data-[selected]:border-dark-tremor-border dark:data-[selected]:bg-dark-tremor-background dark:data-[selected]:shadow-dark-tremor-input dark:[&:not([data-selected])]:hover:text-dark-tremor-content-emphasis dark:data-[selected]:text-dark-tremor-brand dark:[&:not([data-selected])]:text-dark-tremor-content",t?(0,a.getColorClassNames)(t,n.colorPalette.text).selectTextColor:"text-tremor-content dark:text-dark-tremor-content")}}(v,b),f,b&&(0,a.getColorClassNames)(b,n.colorPalette.text).selectTextColor)},m),d?s.default.createElement(d,{className:(0,o.tremorTwMerge)(u("icon"),"flex-none h-5 w-5",p?"mr-2":"")}):null,p?s.default.createElement("span",null,p):null)});c.displayName="Tab",e.s(["Tab",()=>c],197647)},751734,144582,e=>{"use strict";var t=e.i(271645);let r=(0,t.createContext)(0);e.s(["default",()=>r],751734);let n=(0,t.createContext)({selectedValue:void 0,handleValueChange:void 0});e.s(["default",()=>n],144582)},723731,e=>{"use strict";var t=e.i(290571),r=e.i(970554),n=e.i(751734),o=e.i(144582),a=e.i(444755),s=e.i(673706),l=e.i(271645);let i=(0,s.makeClassName)("TabPanels"),u=l.default.forwardRef((e,s)=>{let{children:u,className:c}=e,d=(0,t.__rest)(e,["children","className"]);return l.default.createElement(r.Tab.Panels,Object.assign({as:"div",ref:s,className:(0,a.tremorTwMerge)(i("root"),"w-full",c)},d),({selectedIndex:e})=>l.default.createElement(o.default.Provider,{value:{selectedValue:e}},l.default.Children.map(u,(e,t)=>l.default.createElement(n.default.Provider,{value:t},e))))});u.displayName="TabPanels",e.s(["TabPanels",()=>u],723731)},404206,e=>{"use strict";var t=e.i(290571),r=e.i(751734),n=e.i(144582),o=e.i(444755),a=e.i(673706),s=e.i(271645);let l=(0,a.makeClassName)("TabPanel"),i=s.default.forwardRef((e,a)=>{let{children:i,className:u}=e,c=(0,t.__rest)(e,["children","className"]),{selectedValue:d}=(0,s.useContext)(n.default),f=d===(0,s.useContext)(r.default);return s.default.createElement("div",Object.assign({ref:a,className:(0,o.tremorTwMerge)(l("root"),"w-full mt-2",f?"":"hidden",u),"aria-selected":f?"true":"false"},c),i)});i.displayName="TabPanel",e.s(["TabPanel",()=>i],404206)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/05e9ff30be0ddaae.js b/litellm/proxy/_experimental/out/_next/static/chunks/05e9ff30be0ddaae.js new file mode 100644 index 00000000000..f926944354f --- /dev/null +++ b/litellm/proxy/_experimental/out/_next/static/chunks/05e9ff30be0ddaae.js @@ -0,0 +1,4 @@ +(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,751734,144582,e=>{"use strict";var t=e.i(271645);let r=(0,t.createContext)(0);e.s(["default",()=>r],751734);let n=(0,t.createContext)({selectedValue:void 0,handleValueChange:void 0});e.s(["default",()=>n],144582)},404206,e=>{"use strict";var t=e.i(290571),r=e.i(751734),n=e.i(144582),o=e.i(444755),a=e.i(673706),s=e.i(271645);let l=(0,a.makeClassName)("TabPanel"),i=s.default.forwardRef((e,a)=>{let{children:i,className:u}=e,c=(0,t.__rest)(e,["children","className"]),{selectedValue:d}=(0,s.useContext)(n.default),f=d===(0,s.useContext)(r.default);return s.default.createElement("div",Object.assign({ref:a,className:(0,o.tremorTwMerge)(l("root"),"w-full mt-2",f?"":"hidden",u),"aria-selected":f?"true":"false"},c),i)});i.displayName="TabPanel",e.s(["TabPanel",()=>i],404206)},429427,371330,80758,402155,368578,544508,746725,835696,941444,914189,394487,e=>{"use strict";let t;e.i(247167);var r=e.i(271645);let n="u">typeof document?r.default.useLayoutEffect:()=>{},o=e=>{var t;return null!=(t=null==e?void 0:e.ownerDocument)?t:document},a=e=>e&&"window"in e&&e.window===e?e:o(e).defaultView||window;"u">typeof Element&&Element.prototype;let s=["input:not([disabled]):not([type=hidden])","select:not([disabled])","textarea:not([disabled])","button:not([disabled])","a[href]","area[href]","summary","iframe","object","embed","audio[controls]","video[controls]",'[contenteditable]:not([contenteditable^="false"])',"permission"];s.join(":not([hidden]),"),s.push('[tabindex]:not([tabindex="-1"]):not([disabled])'),s.join(':not([hidden]):not([tabindex="-1"]),');let l=null;function i(e){return e.nativeEvent=e,e.isDefaultPrevented=()=>e.defaultPrevented,e.isPropagationStopped=()=>e.cancelBubble,e.persist=()=>{},e}function u(e){let t=(0,r.useRef)({isFocused:!1,observer:null});return n(()=>{let e=t.current;return()=>{e.observer&&(e.observer.disconnect(),e.observer=null)}},[]),(0,r.useCallback)(r=>{if(r.target instanceof HTMLButtonElement||r.target instanceof HTMLInputElement||r.target instanceof HTMLTextAreaElement||r.target instanceof HTMLSelectElement){t.current.isFocused=!0;let n=r.target;n.addEventListener("focusout",r=>{if(t.current.isFocused=!1,n.disabled){let t=i(r);null==e||e(t)}t.current.observer&&(t.current.observer.disconnect(),t.current.observer=null)},{once:!0}),t.current.observer=new MutationObserver(()=>{if(t.current.isFocused&&n.disabled){var e;null==(e=t.current.observer)||e.disconnect();let r=n===document.activeElement?null:document.activeElement;n.dispatchEvent(new FocusEvent("blur",{relatedTarget:r})),n.dispatchEvent(new FocusEvent("focusout",{bubbles:!0,relatedTarget:r}))}}),t.current.observer.observe(n,{attributes:!0,attributeFilter:["disabled"]})}},[e])}function c(e){var t;if("u"e.test(t.brand))||e.test(window.navigator.userAgent)}function d(e){var t;return"u">typeof window&&null!=window.navigator&&e.test((null==(t=window.navigator.userAgentData)?void 0:t.platform)||window.navigator.platform)}function f(e){let t=null;return()=>(null==t&&(t=e()),t)}let p=f(function(){return d(/^Mac/i)}),m=f(function(){return d(/^iPhone/i)}),v=f(function(){return d(/^iPad/i)||p()&&navigator.maxTouchPoints>1}),b=f(function(){return m()||v()});f(function(){return p()||b()});let g=f(function(){return c(/AppleWebKit/i)&&!h()}),h=f(function(){return c(/Chrome/i)}),y=f(function(){return c(/Android/i)}),E=f(function(){return c(/Firefox/i)});function w(e,t,r=!0){var n,o;let{metaKey:a,ctrlKey:s,altKey:i,shiftKey:u}=t;E()&&(null==(o=window.event)||null==(n=o.type)?void 0:n.startsWith("key"))&&"_blank"===e.target&&(p()?a=!0:s=!0);let c=g()&&p()&&!v()&&1?new KeyboardEvent("keydown",{keyIdentifier:"Enter",metaKey:a,ctrlKey:s,altKey:i,shiftKey:u}):new MouseEvent("click",{metaKey:a,ctrlKey:s,altKey:i,shiftKey:u,detail:1,bubbles:!0,cancelable:!0});if(w.isOpening=r,function(){if(null==l){l=!1;try{document.createElement("div").focus({get preventScroll(){return l=!0,!0}})}catch{}}return l}())e.focus({preventScroll:!0});else{let t=function(e){let t=e.parentNode,r=[],n=document.scrollingElement||document.documentElement;for(;t instanceof HTMLElement&&t!==n;)(t.offsetHeighttypeof window&&window.document&&window.document.createElement,new WeakMap;r.default.useId;let x=null,F=new Set,P=new Map,k=!1,L=!1,N={Tab:!0,Escape:!0};function C(e,t){for(let r of F)r(e,t)}function I(e){k=!0,w.isOpening||e.metaKey||!p()&&e.altKey||e.ctrlKey||"Control"===e.key||"Shift"===e.key||"Meta"===e.key||(x="keyboard",C("keyboard",e))}function S(e){x="pointer","pointerType"in e&&e.pointerType,("mousedown"===e.type||"pointerdown"===e.type)&&(k=!0,C("pointer",e))}function A(e){w.isOpening||(""!==e.pointerType||!e.isTrusted)&&(y()&&e.pointerType?"click"!==e.type||1!==e.buttons:0!==e.detail||e.pointerType)||(k=!0,x="virtual")}function M(e){e.target!==window&&e.target!==document&&e.isTrusted&&(k||L||(x="virtual",C("virtual",e)),k=!1,L=!1)}function R(){k=!1,L=!0}function O(e){if("u"typeof PointerEvent&&(r.addEventListener("pointerdown",S,!0),r.addEventListener("pointermove",S,!0),r.addEventListener("pointerup",S,!0)),t.addEventListener("beforeunload",()=>{D(e)},{once:!0}),P.set(t,{focus:n})}let D=(e,t)=>{let r=a(e),n=o(e);t&&n.removeEventListener("DOMContentLoaded",t),P.has(r)&&(r.HTMLElement.prototype.focus=P.get(r).focus,n.removeEventListener("keydown",I,!0),n.removeEventListener("keyup",I,!0),n.removeEventListener("click",A,!0),r.removeEventListener("focus",M,!0),r.removeEventListener("blur",R,!1),"u">typeof PointerEvent&&(n.removeEventListener("pointerdown",S,!0),n.removeEventListener("pointermove",S,!0),n.removeEventListener("pointerup",S,!0)),P.delete(r))};function H(){return"pointer"!==x}"u">typeof document&&("loading"!==(t=o(void 0)).readyState?O(void 0):t.addEventListener("DOMContentLoaded",()=>{O(void 0)}));let j=new Set(["checkbox","radio","range","color","file","image","button","submit","reset"]);function K(e,t){return!!t&&!!e&&e.contains(t)}function W(){let e=(0,r.useRef)(new Map),t=(0,r.useCallback)((t,r,n,o)=>{let a=(null==o?void 0:o.once)?(...t)=>{e.current.delete(n),n(...t)}:n;e.current.set(n,{type:r,eventTarget:t,fn:a,options:o}),t.addEventListener(r,a,o)},[]),n=(0,r.useCallback)((t,r,n,o)=>{var a;let s=(null==(a=e.current.get(n))?void 0:a.fn)||n;t.removeEventListener(r,s,o),e.current.delete(n)},[]),o=(0,r.useCallback)(()=>{e.current.forEach((e,t)=>{n(e.eventTarget,e.type,t,e.options)})},[n]);return(0,r.useEffect)(()=>o,[o]),{addGlobalListener:t,removeGlobalListener:n,removeAllGlobalListeners:o}}function B(e={}){var t;let{autoFocus:n=!1,isTextInput:s,within:l}=e,c=(0,r.useRef)({isFocused:!1,isFocusVisible:n||H()}),[d,f]=(0,r.useState)(!1),[p,m]=(0,r.useState)(()=>c.current.isFocused&&c.current.isFocusVisible),v=(0,r.useCallback)(()=>m(c.current.isFocused&&c.current.isFocusVisible),[]),b=(0,r.useCallback)(e=>{c.current.isFocused=e,f(e),v()},[v]);t={isTextInput:s},O(),(0,r.useEffect)(()=>{let e=(e,r)=>{var n;let s,l,i,u,d;n=!!(null==t?void 0:t.isTextInput),s=o(null==r?void 0:r.target),l="u">typeof window?a(null==r?void 0:r.target).HTMLInputElement:HTMLInputElement,i="u">typeof window?a(null==r?void 0:r.target).HTMLTextAreaElement:HTMLTextAreaElement,u="u">typeof window?a(null==r?void 0:r.target).HTMLElement:HTMLElement,d="u">typeof window?a(null==r?void 0:r.target).KeyboardEvent:KeyboardEvent,(n=n||s.activeElement instanceof l&&!j.has(s.activeElement.type)||s.activeElement instanceof i||s.activeElement instanceof u&&s.activeElement.isContentEditable)&&"keyboard"===e&&r instanceof d&&!N[r.key]||(e=>{c.current.isFocusVisible=e,v()})(H())};return F.add(e),()=>{F.delete(e)}},[]);let{focusProps:g}=function(e){let{isDisabled:t,onFocus:n,onBlur:a,onFocusChange:s}=e,l=(0,r.useCallback)(e=>{if(e.target===e.currentTarget)return a&&a(e),s&&s(!1),!0},[a,s]),i=u(l),c=(0,r.useCallback)(e=>{var t;let r=o(e.target),a=r?((e=document)=>e.activeElement)(r):((e=document)=>e.activeElement)();e.target===e.currentTarget&&a===(t=e.nativeEvent,t.target)&&(n&&n(e),s&&s(!0),i(e))},[s,n,i]);return{focusProps:{onFocus:!t&&(n||s||a)?c:void 0,onBlur:!t&&(a||s)?l:void 0}}}({isDisabled:l,onFocusChange:b}),{focusWithinProps:h}=function(e){let{isDisabled:t,onBlurWithin:n,onFocusWithin:a,onFocusWithinChange:s}=e,l=(0,r.useRef)({isFocusWithin:!1}),{addGlobalListener:c,removeAllGlobalListeners:d}=W(),f=(0,r.useCallback)(e=>{e.currentTarget.contains(e.target)&&l.current.isFocusWithin&&!e.currentTarget.contains(e.relatedTarget)&&(l.current.isFocusWithin=!1,d(),n&&n(e),s&&s(!1))},[n,s,l,d]),p=u(f),m=(0,r.useCallback)(e=>{var t;if(!e.currentTarget.contains(e.target))return;let r=o(e.target),n=((e=document)=>e.activeElement)(r);if(!l.current.isFocusWithin&&n===(t=e.nativeEvent,t.target)){a&&a(e),s&&s(!0),l.current.isFocusWithin=!0,p(e);let t=e.currentTarget;c(r,"focus",e=>{if(l.current.isFocusWithin&&!K(t,e.target)){let n=new r.defaultView.FocusEvent("blur",{relatedTarget:e.target});Object.defineProperty(n,"target",{value:t}),Object.defineProperty(n,"currentTarget",{value:t}),f(i(n))}},{capture:!0})}},[a,s,p,c,f]);return t?{focusWithinProps:{onFocus:void 0,onBlur:void 0}}:{focusWithinProps:{onFocus:m,onBlur:f}}}({isDisabled:!l,onFocusWithinChange:b});return{isFocused:d,isFocusVisible:p,focusProps:l?h:g}}e.s(["useFocusRing",()=>B],429427);let V=!1,_=0;function G(e){"touch"===e.pointerType&&(V=!0,setTimeout(()=>{V=!1},50))}function U(){if("u">typeof document)return 0===_&&"u">typeof PointerEvent&&document.addEventListener("pointerup",G),_++,()=>{!(--_>0)&&"u">typeof PointerEvent&&document.removeEventListener("pointerup",G)}}function $(e){let{onHoverStart:t,onHoverChange:n,onHoverEnd:a,isDisabled:s}=e,[l,i]=(0,r.useState)(!1),u=(0,r.useRef)({isHovered:!1,ignoreEmulatedMouseEvents:!1,pointerType:"",target:null}).current;(0,r.useEffect)(U,[]);let{addGlobalListener:c,removeAllGlobalListeners:d}=W(),{hoverProps:f,triggerHoverEnd:p}=(0,r.useMemo)(()=>{let e=(e,t)=>{let r=u.target;u.pointerType="",u.target=null,"touch"!==t&&u.isHovered&&r&&(u.isHovered=!1,d(),a&&a({type:"hoverend",target:r,pointerType:t}),n&&n(!1),i(!1))},r={};return"u">typeof PointerEvent&&(r.onPointerEnter=r=>{V&&"mouse"===r.pointerType||((r,a)=>{if(u.pointerType=a,s||"touch"===a||u.isHovered||!r.currentTarget.contains(r.target))return;u.isHovered=!0;let l=r.currentTarget;u.target=l,c(o(r.target),"pointerover",t=>{u.isHovered&&u.target&&!K(u.target,t.target)&&e(t,t.pointerType)},{capture:!0}),t&&t({type:"hoverstart",target:l,pointerType:a}),n&&n(!0),i(!0)})(r,r.pointerType)},r.onPointerLeave=t=>{!s&&t.currentTarget.contains(t.target)&&e(t,t.pointerType)}),{hoverProps:r,triggerHoverEnd:e}},[t,n,a,s,u,c,d]);return(0,r.useEffect)(()=>{s&&p({currentTarget:u.target},u.pointerType)},[s]),{hoverProps:f,isHovered:l}}e.s(["useHover",()=>$],371330);var q=Object.defineProperty,X=(e,t,r)=>{let n;return(n="symbol"!=typeof t?t+"":t)in e?q(e,n,{enumerable:!0,configurable:!0,writable:!0,value:r}):e[n]=r,r};let Y=new class{constructor(){X(this,"current",this.detect()),X(this,"handoffState","pending"),X(this,"currentId",0)}set(e){this.current!==e&&(this.handoffState="pending",this.currentId=0,this.current=e)}reset(){this.set(this.detect())}nextId(){return++this.currentId}get isServer(){return"server"===this.current}get isClient(){return"client"===this.current}detect(){return"u"setTimeout(()=>{throw e}))}function J(){let e=[],t={addEventListener:(e,r,n,o)=>(e.addEventListener(r,n,o),t.add(()=>e.removeEventListener(r,n,o))),requestAnimationFrame(...e){let r=requestAnimationFrame(...e);return t.add(()=>cancelAnimationFrame(r))},nextFrame:(...e)=>t.requestAnimationFrame(()=>t.requestAnimationFrame(...e)),setTimeout(...e){let r=setTimeout(...e);return t.add(()=>clearTimeout(r))},microTask(...e){let r={current:!0};return Z(()=>{r.current&&e[0]()}),t.add(()=>{r.current=!1})},style(e,t,r){let n=e.style.getPropertyValue(t);return Object.assign(e.style,{[t]:r}),this.add(()=>{Object.assign(e.style,{[t]:n})})},group(e){let t=J();return e(t),this.add(()=>t.dispose())},add:t=>(e.includes(t)||e.push(t),()=>{let r=e.indexOf(t);if(r>=0)for(let t of e.splice(r,1))t()}),dispose(){for(let t of e.splice(0))t()}};return t}function Q(){let[e]=(0,r.useState)(J);return(0,r.useEffect)(()=>()=>e.dispose(),[e]),e}e.s(["env",()=>Y],80758),e.s(["getOwnerDocument",()=>z],402155),e.s(["microTask",()=>Z],368578),e.s(["disposables",()=>J],544508),e.s(["useDisposables",()=>Q],746725);let ee=(e,t)=>{Y.isServer?(0,r.useEffect)(e,t):(0,r.useLayoutEffect)(e,t)};function et(e){let t=(0,r.useRef)(e);return ee(()=>{t.current=e},[e]),t}e.s(["useIsoMorphicEffect",()=>ee],835696),e.s(["useLatestValue",()=>et],941444);let er=function(e){let t=et(e);return r.default.useCallback((...e)=>t.current(...e),[t])};function en({disabled:e=!1}={}){let t=(0,r.useRef)(null),[n,o]=(0,r.useState)(!1),a=Q(),s=er(()=>{t.current=null,o(!1),a.dispose()}),l=er(e=>{if(a.dispose(),null===t.current){t.current=e.currentTarget,o(!0);{let r=z(e.currentTarget);a.addEventListener(r,"pointerup",s,!1),a.addEventListener(r,"pointermove",e=>{if(t.current){var r,n;let a,s;o((a=e.width/2,s=e.height/2,r={top:e.clientY-s,right:e.clientX+a,bottom:e.clientY+s,left:e.clientX-a},n=t.current.getBoundingClientRect(),!(!r||!n||r.rightn.right||r.bottomn.bottom)))}},!1),a.addEventListener(r,"pointercancel",s,!1)}}});return{pressed:n,pressProps:e?{}:{onPointerDown:l,onPointerUp:s,onClick:s}}}e.s(["useEvent",()=>er],914189),e.s(["useActivePress",()=>en],394487)},397701,e=>{"use strict";function t(e,r,...n){if(e in r){let t=r[e];return"function"==typeof t?t(...n):t}let o=Error(`Tried to handle "${e}" but there is no handler defined. Only defined handlers are: ${Object.keys(r).map(e=>`"${e}"`).join(", ")}.`);throw Error.captureStackTrace&&Error.captureStackTrace(o,t),o}e.s(["match",()=>t])},652265,e=>{"use strict";let t,r,n,o,a;e.i(544508);var s=e.i(397701),l=e.i(402155);let i=["[contentEditable=true]","[tabindex]","a[href]","area[href]","button:not([disabled])","iframe","input:not([disabled])","select:not([disabled])","textarea:not([disabled])"].map(e=>`${e}:not([tabindex='-1'])`).join(","),u=["[data-autofocus]"].map(e=>`${e}:not([tabindex='-1'])`).join(",");var c=((t=c||{})[t.First=1]="First",t[t.Previous=2]="Previous",t[t.Next=4]="Next",t[t.Last=8]="Last",t[t.WrapAround=16]="WrapAround",t[t.NoScroll=32]="NoScroll",t[t.AutoFocus=64]="AutoFocus",t),d=((r=d||{})[r.Error=0]="Error",r[r.Overflow=1]="Overflow",r[r.Success=2]="Success",r[r.Underflow=3]="Underflow",r),f=((n=f||{})[n.Previous=-1]="Previous",n[n.Next=1]="Next",n);function p(e=document.body){return null==e?[]:Array.from(e.querySelectorAll(i)).sort((e,t)=>Math.sign((e.tabIndex||Number.MAX_SAFE_INTEGER)-(t.tabIndex||Number.MAX_SAFE_INTEGER)))}var m=((o=m||{})[o.Strict=0]="Strict",o[o.Loose=1]="Loose",o);function v(e,t=0){var r;return e!==(null==(r=(0,l.getOwnerDocument)(e))?void 0:r.body)&&(0,s.match)(t,{0:()=>e.matches(i),1(){let t=e;for(;null!==t;){if(t.matches(i))return!0;t=t.parentElement}return!1}})}var b=((a=b||{})[a.Keyboard=0]="Keyboard",a[a.Mouse=1]="Mouse",a);function g(e,t=e=>e){return e.slice().sort((e,r)=>{let n=t(e),o=t(r);if(null===n||null===o)return 0;let a=n.compareDocumentPosition(o);return a&Node.DOCUMENT_POSITION_FOLLOWING?-1:a&Node.DOCUMENT_POSITION_PRECEDING?1:0})}function h(e,t){return y(p(),t,{relativeTo:e})}function y(e,t,{sorted:r=!0,relativeTo:n=null,skipElements:o=[]}={}){var a,s,l;let i=Array.isArray(e)?e.length>0?e[0].ownerDocument:document:e.ownerDocument,c=Array.isArray(e)?r?g(e):e:64&t?function(e=document.body){return null==e?[]:Array.from(e.querySelectorAll(u)).sort((e,t)=>Math.sign((e.tabIndex||Number.MAX_SAFE_INTEGER)-(t.tabIndex||Number.MAX_SAFE_INTEGER)))}(e):p(e);o.length>0&&c.length>1&&(c=c.filter(e=>!o.some(t=>null!=t&&"current"in t?(null==t?void 0:t.current)===e:t===e))),n=null!=n?n:i.activeElement;let d=(()=>{if(5&t)return 1;if(10&t)return -1;throw Error("Missing Focus.First, Focus.Previous, Focus.Next or Focus.Last")})(),f=(()=>{if(1&t)return 0;if(2&t)return Math.max(0,c.indexOf(n))-1;if(4&t)return Math.max(0,c.indexOf(n))+1;if(8&t)return c.length-1;throw Error("Missing Focus.First, Focus.Previous, Focus.Next or Focus.Last")})(),m=32&t?{preventScroll:!0}:{},v=0,b=c.length,h;do{if(v>=b||v+b<=0)return 0;let e=f+v;if(16&t)e=(e+b)%b;else{if(e<0)return 3;if(e>=b)return 1}null==(h=c[e])||h.focus(m),v+=d}while(h!==i.activeElement)return 6&t&&null!=(l=null==(s=null==(a=h)?void 0:a.matches)?void 0:s.call(a,"textarea,input"))&&l&&h.select(),2}"u">typeof window&&"u">typeof document&&(document.addEventListener("keydown",e=>{e.metaKey||e.altKey||e.ctrlKey||(document.documentElement.dataset.headlessuiFocusVisible="")},!0),document.addEventListener("click",e=>{1===e.detail?delete document.documentElement.dataset.headlessuiFocusVisible:0===e.detail&&(document.documentElement.dataset.headlessuiFocusVisible="")},!0)),e.s(["Focus",()=>c,"FocusResult",()=>d,"FocusableMode",()=>m,"focusFrom",()=>h,"focusIn",()=>y,"getFocusableElements",()=>p,"isFocusableElement",()=>v,"sortByDomNode",()=>g])},144279,294316,e=>{"use strict";var t=e.i(271645);function r(e,r){return(0,t.useMemo)(()=>{var t;if(e.type)return e.type;let n=null!=(t=e.as)?t:"button";if("string"==typeof n&&"button"===n.toLowerCase()||(null==r?void 0:r.tagName)==="BUTTON"&&!r.hasAttribute("type"))return"button"},[e.type,e.as,r])}e.s(["useResolveButtonType",()=>r],144279);var n=e.i(914189);let o=Symbol();function a(e,t=!0){return Object.assign(e,{[o]:t})}function s(...e){let r=(0,t.useRef)(e);(0,t.useEffect)(()=>{r.current=e},[e]);let a=(0,n.useEvent)(e=>{for(let t of r.current)null!=t&&("function"==typeof t?t(e):t.current=e)});return e.every(e=>null==e||(null==e?void 0:e[o]))?void 0:a}e.s(["optionalRef",()=>a,"useSyncRefs",()=>s],294316)},732607,e=>{"use strict";function t(...e){return Array.from(new Set(e.flatMap(e=>"string"==typeof e?e.split(" "):[]))).filter(Boolean).join(" ")}e.s(["classNames",()=>t])},700020,e=>{"use strict";let t,r;var n=e.i(271645),o=e.i(732607),a=e.i(397701),s=((t=s||{})[t.None=0]="None",t[t.RenderStrategy=1]="RenderStrategy",t[t.Static=2]="Static",t),l=((r=l||{})[r.Unmount=0]="Unmount",r[r.Hidden=1]="Hidden",r);function i(){let e,t,r=(e=(0,n.useRef)([]),t=(0,n.useCallback)(t=>{for(let r of e.current)null!=r&&("function"==typeof r?r(t):r.current=t)},[]),(...r)=>{if(!r.every(e=>null==e))return e.current=r,t});return(0,n.useCallback)(e=>(function({ourProps:e,theirProps:t,slot:r,defaultTag:n,features:o,visible:s=!0,name:l,mergeRefs:i}){i=null!=i?i:c;let f=d(t,e);if(s)return u(f,r,n,l,i);let p=null!=o?o:0;if(2&p){let{static:e=!1,...t}=f;if(e)return u(t,r,n,l,i)}if(1&p){let{unmount:e=!0,...t}=f;return(0,a.match)(+!e,{0:()=>null,1:()=>u({...t,hidden:!0,style:{display:"none"}},r,n,l,i)})}return u(f,r,n,l,i)})({mergeRefs:r,...e}),[r])}function u(e,t={},r,a,s){let{as:l=r,children:i,refName:c="ref",...f}=v(e,["unmount","static"]),p=void 0!==e.ref?{[c]:e.ref}:{},b="function"==typeof i?i(t):i;"className"in f&&f.className&&"function"==typeof f.className&&(f.className=f.className(t)),f["aria-labelledby"]&&f["aria-labelledby"]===f.id&&(f["aria-labelledby"]=void 0);let g={};if(t){let e=!1,r=[];for(let[n,o]of Object.entries(t))"boolean"==typeof o&&(e=!0),!0===o&&r.push(n.replace(/([A-Z])/g,e=>`-${e.toLowerCase()}`));if(e)for(let e of(g["data-headlessui-state"]=r.join(" "),r))g[`data-${e}`]=""}if(l===n.Fragment&&(Object.keys(m(f)).length>0||Object.keys(m(g)).length>0))if(!(0,n.isValidElement)(b)||Array.isArray(b)&&b.length>1){if(Object.keys(m(f)).length>0)throw Error(['Passing props on "Fragment"!',"",`The current component <${a} /> is rendering a "Fragment".`,"However we need to passthrough the following props:",Object.keys(m(f)).concat(Object.keys(m(g))).map(e=>` - ${e}`).join(` +`),"","You can apply a few solutions:",['Add an `as="..."` prop, to ensure that we render an actual element instead of a "Fragment".',"Render a single element as the child so that we can forward the props onto that element."].map(e=>` - ${e}`).join(` +`)].join(` +`))}else{var h;let e=b.props,t=null==e?void 0:e.className,r="function"==typeof t?(...e)=>(0,o.classNames)(t(...e),f.className):(0,o.classNames)(t,f.className),a=d(b.props,m(v(f,["ref"])));for(let e in g)e in a&&delete g[e];return(0,n.cloneElement)(b,Object.assign({},a,g,p,{ref:s((h=b,n.default.version.split(".")[0]>="19"?h.props.ref:h.ref),p.ref)},r?{className:r}:{}))}return(0,n.createElement)(l,Object.assign({},v(f,["ref"]),l!==n.Fragment&&p,l!==n.Fragment&&g),b)}function c(...e){return e.every(e=>null==e)?void 0:t=>{for(let r of e)null!=r&&("function"==typeof r?r(t):r.current=t)}}function d(...e){if(0===e.length)return{};if(1===e.length)return e[0];let t={},r={};for(let n of e)for(let e in n)e.startsWith("on")&&"function"==typeof n[e]?(null!=r[e]||(r[e]=[]),r[e].push(n[e])):t[e]=n[e];if(t.disabled||t["aria-disabled"])for(let e in r)/^(on(?:Click|Pointer|Mouse|Key)(?:Down|Up|Press)?)$/.test(e)&&(r[e]=[e=>{var t;return null==(t=null==e?void 0:e.preventDefault)?void 0:t.call(e)}]);for(let e in r)Object.assign(t,{[e](t,...n){for(let o of r[e]){if((t instanceof Event||(null==t?void 0:t.nativeEvent)instanceof Event)&&t.defaultPrevented)return;o(t,...n)}}});return t}function f(...e){if(0===e.length)return{};if(1===e.length)return e[0];let t={},r={};for(let n of e)for(let e in n)e.startsWith("on")&&"function"==typeof n[e]?(null!=r[e]||(r[e]=[]),r[e].push(n[e])):t[e]=n[e];for(let e in r)Object.assign(t,{[e](...t){for(let n of r[e])null==n||n(...t)}});return t}function p(e){var t;return Object.assign((0,n.forwardRef)(e),{displayName:null!=(t=e.displayName)?t:e.name})}function m(e){let t=Object.assign({},e);for(let e in t)void 0===t[e]&&delete t[e];return t}function v(e,t=[]){let r=Object.assign({},e);for(let e of t)e in r&&delete r[e];return r}e.s(["RenderFeatures",()=>s,"RenderStrategy",()=>l,"compact",()=>m,"forwardRefWithAs",()=>p,"mergeProps",()=>f,"useRender",()=>i])},2788,e=>{"use strict";let t;var r=e.i(700020),n=((t=n||{})[t.None=1]="None",t[t.Focusable=2]="Focusable",t[t.Hidden=4]="Hidden",t);let o=(0,r.forwardRefWithAs)(function(e,t){var n;let{features:o=1,...a}=e,s={ref:t,"aria-hidden":(2&o)==2||(null!=(n=a["aria-hidden"])?n:void 0),hidden:(4&o)==4||void 0,style:{position:"fixed",top:1,left:1,width:1,height:0,padding:0,margin:-1,overflow:"hidden",clip:"rect(0, 0, 0, 0)",whiteSpace:"nowrap",borderWidth:"0",...(4&o)==4&&(2&o)!=2&&{display:"none"}}};return(0,r.useRender)()({ourProps:s,theirProps:a,slot:{},defaultTag:"span",name:"Hidden"})});e.s(["Hidden",()=>o,"HiddenFeatures",()=>n])},998348,e=>{"use strict";let t;var r=((t=r||{}).Space=" ",t.Enter="Enter",t.Escape="Escape",t.Backspace="Backspace",t.Delete="Delete",t.ArrowLeft="ArrowLeft",t.ArrowUp="ArrowUp",t.ArrowRight="ArrowRight",t.ArrowDown="ArrowDown",t.Home="Home",t.End="End",t.PageUp="PageUp",t.PageDown="PageDown",t.Tab="Tab",t);e.s(["Keys",()=>r])},553521,e=>{"use strict";var t=e.i(271645),r=e.i(835696);function n(){let e=(0,t.useRef)(!1);return(0,r.useIsoMorphicEffect)(()=>(e.current=!0,()=>{e.current=!1}),[]),e}e.s(["useIsMounted",()=>n])},640497,e=>{"use strict";var t=e.i(271645),r=e.i(553521),n=e.i(2788);function o({onFocus:e}){let[o,a]=(0,t.useState)(!0),s=(0,r.useIsMounted)();return o?t.default.createElement(n.Hidden,{as:"button",type:"button",features:n.HiddenFeatures.Focusable,onFocus:t=>{t.preventDefault();let r,n=50;r=requestAnimationFrame(function t(){if(n--<=0){r&&cancelAnimationFrame(r);return}if(e()){if(cancelAnimationFrame(r),!s.current)return;a(!1);return}r=requestAnimationFrame(t)})}}):null}e.s(["FocusSentinel",()=>o])},963703,e=>{"use strict";var t=e.i(271645);let r=t.createContext(null);function n({children:e}){let n=t.useRef({groups:new Map,get(e,t){var r;let n=this.groups.get(e);n||(n=new Map,this.groups.set(e,n));let o=null!=(r=n.get(t))?r:0;return n.set(t,o+1),[Array.from(n.keys()).indexOf(t),function(){let e=n.get(t);e>1?n.set(t,e-1):n.delete(t)}]}});return t.createElement(r.Provider,{value:n},e)}function o(e){let n=t.useContext(r);if(!n)throw Error("You must wrap your component in a ");let o=t.useId(),[a,s]=n.current.get(e,o);return t.useEffect(()=>s,[]),a}e.s(["StableCollection",()=>n,"useStableCollectionIndex",()=>o])},970554,e=>{"use strict";let t,r,n;var o=e.i(429427),a=e.i(371330),s=e.i(271645),l=e.i(394487),i=e.i(914189),u=e.i(835696),c=e.i(941444),d=e.i(144279),f=e.i(294316),p=e.i(640497),m=e.i(2788),v=e.i(652265),b=e.i(397701),g=e.i(368578),h=e.i(402155),y=e.i(700020),E=e.i(963703),w=e.i(998348),T=((t=T||{})[t.Forwards=0]="Forwards",t[t.Backwards=1]="Backwards",t),x=((r=x||{})[r.Less=-1]="Less",r[r.Equal=0]="Equal",r[r.Greater=1]="Greater",r),F=((n=F||{})[n.SetSelectedIndex=0]="SetSelectedIndex",n[n.RegisterTab=1]="RegisterTab",n[n.UnregisterTab=2]="UnregisterTab",n[n.RegisterPanel=3]="RegisterPanel",n[n.UnregisterPanel=4]="UnregisterPanel",n);let P={0(e,t){var r;let n=(0,v.sortByDomNode)(e.tabs,e=>e.current),o=(0,v.sortByDomNode)(e.panels,e=>e.current),a=n.filter(e=>{var t;return!(null!=(t=e.current)&&t.hasAttribute("disabled"))}),s={...e,tabs:n,panels:o};if(t.index<0||t.index>n.length-1){let r=(0,b.match)(Math.sign(t.index-e.selectedIndex),{[-1]:()=>1,0:()=>(0,b.match)(Math.sign(t.index),{[-1]:()=>0,0:()=>0,1:()=>1}),1:()=>0});if(0===a.length)return s;let o=(0,b.match)(r,{0:()=>n.indexOf(a[0]),1:()=>n.indexOf(a[a.length-1])});return{...s,selectedIndex:-1===o?e.selectedIndex:o}}let l=n.slice(0,t.index),i=[...n.slice(t.index),...l].find(e=>a.includes(e));if(!i)return s;let u=null!=(r=n.indexOf(i))?r:e.selectedIndex;return -1===u&&(u=e.selectedIndex),{...s,selectedIndex:u}},1(e,t){if(e.tabs.includes(t.tab))return e;let r=e.tabs[e.selectedIndex],n=(0,v.sortByDomNode)([...e.tabs,t.tab],e=>e.current),o=e.selectedIndex;return e.info.current.isControlled||-1===(o=n.indexOf(r))&&(o=e.selectedIndex),{...e,tabs:n,selectedIndex:o}},2:(e,t)=>({...e,tabs:e.tabs.filter(e=>e!==t.tab)}),3:(e,t)=>e.panels.includes(t.panel)?e:{...e,panels:(0,v.sortByDomNode)([...e.panels,t.panel],e=>e.current)},4:(e,t)=>({...e,panels:e.panels.filter(e=>e!==t.panel)})},k=(0,s.createContext)(null);function L(e){let t=(0,s.useContext)(k);if(null===t){let t=Error(`<${e} /> is missing a parent component.`);throw Error.captureStackTrace&&Error.captureStackTrace(t,L),t}return t}k.displayName="TabsDataContext";let N=(0,s.createContext)(null);function C(e){let t=(0,s.useContext)(N);if(null===t){let t=Error(`<${e} /> is missing a parent component.`);throw Error.captureStackTrace&&Error.captureStackTrace(t,C),t}return t}function I(e,t){return(0,b.match)(t.type,P,e,t)}N.displayName="TabsActionsContext";let S=y.RenderFeatures.RenderStrategy|y.RenderFeatures.Static,A=Object.assign((0,y.forwardRefWithAs)(function(e,t){var r,n;let c=(0,s.useId)(),{id:p=`headlessui-tabs-tab-${c}`,disabled:m=!1,autoFocus:T=!1,...x}=e,{orientation:F,activation:P,selectedIndex:k,tabs:N,panels:I}=L("Tab"),S=C("Tab"),A=L("Tab"),[M,R]=(0,s.useState)(null),O=(0,s.useRef)(null),D=(0,f.useSyncRefs)(O,t,R);(0,u.useIsoMorphicEffect)(()=>S.registerTab(O),[S,O]);let H=(0,E.useStableCollectionIndex)("tabs"),j=N.indexOf(O);-1===j&&(j=H);let K=j===k,W=(0,i.useEvent)(e=>{var t;let r=e();if(r===v.FocusResult.Success&&"auto"===P){let e=null==(t=(0,h.getOwnerDocument)(O))?void 0:t.activeElement,r=A.tabs.findIndex(t=>t.current===e);-1!==r&&S.change(r)}return r}),B=(0,i.useEvent)(e=>{let t=N.map(e=>e.current).filter(Boolean);if(e.key===w.Keys.Space||e.key===w.Keys.Enter){e.preventDefault(),e.stopPropagation(),S.change(j);return}switch(e.key){case w.Keys.Home:case w.Keys.PageUp:return e.preventDefault(),e.stopPropagation(),W(()=>(0,v.focusIn)(t,v.Focus.First));case w.Keys.End:case w.Keys.PageDown:return e.preventDefault(),e.stopPropagation(),W(()=>(0,v.focusIn)(t,v.Focus.Last))}if(W(()=>(0,b.match)(F,{vertical:()=>e.key===w.Keys.ArrowUp?(0,v.focusIn)(t,v.Focus.Previous|v.Focus.WrapAround):e.key===w.Keys.ArrowDown?(0,v.focusIn)(t,v.Focus.Next|v.Focus.WrapAround):v.FocusResult.Error,horizontal:()=>e.key===w.Keys.ArrowLeft?(0,v.focusIn)(t,v.Focus.Previous|v.Focus.WrapAround):e.key===w.Keys.ArrowRight?(0,v.focusIn)(t,v.Focus.Next|v.Focus.WrapAround):v.FocusResult.Error}))===v.FocusResult.Success)return e.preventDefault()}),V=(0,s.useRef)(!1),_=(0,i.useEvent)(()=>{var e;V.current||(V.current=!0,null==(e=O.current)||e.focus({preventScroll:!0}),S.change(j),(0,g.microTask)(()=>{V.current=!1}))}),G=(0,i.useEvent)(e=>{e.preventDefault()}),{isFocusVisible:U,focusProps:$}=(0,o.useFocusRing)({autoFocus:T}),{isHovered:q,hoverProps:X}=(0,a.useHover)({isDisabled:m}),{pressed:Y,pressProps:z}=(0,l.useActivePress)({disabled:m}),Z=(0,s.useMemo)(()=>({selected:K,hover:q,active:Y,focus:U,autofocus:T,disabled:m}),[K,q,U,Y,T,m]),J=(0,y.mergeProps)({ref:D,onKeyDown:B,onMouseDown:G,onClick:_,id:p,role:"tab",type:(0,d.useResolveButtonType)(e,M),"aria-controls":null==(n=null==(r=I[j])?void 0:r.current)?void 0:n.id,"aria-selected":K,tabIndex:K?0:-1,disabled:m||void 0,autoFocus:T},$,X,z);return(0,y.useRender)()({ourProps:J,theirProps:x,slot:Z,defaultTag:"button",name:"Tabs.Tab"})}),{Group:(0,y.forwardRefWithAs)(function(e,t){let{defaultIndex:r=0,vertical:n=!1,manual:o=!1,onChange:a,selectedIndex:l=null,...d}=e,m=n?"vertical":"horizontal",b=o?"manual":"auto",g=null!==l,h=(0,c.useLatestValue)({isControlled:g}),w=(0,f.useSyncRefs)(t),[T,x]=(0,s.useReducer)(I,{info:h,selectedIndex:null!=l?l:r,tabs:[],panels:[]}),F=(0,s.useMemo)(()=>({selectedIndex:T.selectedIndex}),[T.selectedIndex]),P=(0,c.useLatestValue)(a||(()=>{})),L=(0,c.useLatestValue)(T.tabs),C=(0,s.useMemo)(()=>({orientation:m,activation:b,...T}),[m,b,T]),S=(0,i.useEvent)(e=>(x({type:1,tab:e}),()=>x({type:2,tab:e}))),A=(0,i.useEvent)(e=>(x({type:3,panel:e}),()=>x({type:4,panel:e}))),M=(0,i.useEvent)(e=>{R.current!==e&&P.current(e),g||x({type:0,index:e})}),R=(0,c.useLatestValue)(g?e.selectedIndex:T.selectedIndex),O=(0,s.useMemo)(()=>({registerTab:S,registerPanel:A,change:M}),[]);(0,u.useIsoMorphicEffect)(()=>{x({type:0,index:null!=l?l:r})},[l]),(0,u.useIsoMorphicEffect)(()=>{if(void 0===R.current||T.tabs.length<=0)return;let e=(0,v.sortByDomNode)(T.tabs,e=>e.current);e.some((e,t)=>T.tabs[t]!==e)&&M(e.indexOf(T.tabs[R.current]))});let D=(0,y.useRender)();return s.default.createElement(E.StableCollection,null,s.default.createElement(N.Provider,{value:O},s.default.createElement(k.Provider,{value:C},C.tabs.length<=0&&s.default.createElement(p.FocusSentinel,{onFocus:()=>{var e,t;for(let r of L.current)if((null==(e=r.current)?void 0:e.tabIndex)===0)return null==(t=r.current)||t.focus(),!0;return!1}}),D({ourProps:{ref:w},theirProps:d,slot:F,defaultTag:"div",name:"Tabs"}))))}),List:(0,y.forwardRefWithAs)(function(e,t){let{orientation:r,selectedIndex:n}=L("Tab.List"),o=(0,f.useSyncRefs)(t),a=(0,s.useMemo)(()=>({selectedIndex:n}),[n]);return(0,y.useRender)()({ourProps:{ref:o,role:"tablist","aria-orientation":r},theirProps:e,slot:a,defaultTag:"div",name:"Tabs.List"})}),Panels:(0,y.forwardRefWithAs)(function(e,t){let{selectedIndex:r}=L("Tab.Panels"),n=(0,f.useSyncRefs)(t),o=(0,s.useMemo)(()=>({selectedIndex:r}),[r]);return(0,y.useRender)()({ourProps:{ref:n},theirProps:e,slot:o,defaultTag:"div",name:"Tabs.Panels"})}),Panel:(0,y.forwardRefWithAs)(function(e,t){var r,n,a,l;let i=(0,s.useId)(),{id:c=`headlessui-tabs-panel-${i}`,tabIndex:d=0,...p}=e,{selectedIndex:v,tabs:b,panels:g}=L("Tab.Panel"),h=C("Tab.Panel"),w=(0,s.useRef)(null),T=(0,f.useSyncRefs)(w,t);(0,u.useIsoMorphicEffect)(()=>h.registerPanel(w),[h,w]);let x=(0,E.useStableCollectionIndex)("panels"),F=g.indexOf(w);-1===F&&(F=x);let P=F===v,{isFocusVisible:k,focusProps:N}=(0,o.useFocusRing)(),I=(0,s.useMemo)(()=>({selected:P,focus:k}),[P,k]),A=(0,y.mergeProps)({ref:T,id:c,role:"tabpanel","aria-labelledby":null==(n=null==(r=b[F])?void 0:r.current)?void 0:n.id,tabIndex:P?d:-1},N),M=(0,y.useRender)();return P||null!=(a=p.unmount)&&!a||null!=(l=p.static)&&l?M({ourProps:A,theirProps:p,slot:I,defaultTag:"div",features:S,visible:P,name:"Tabs.Panel"}):s.default.createElement(m.Hidden,{"aria-hidden":"true",...A})})});e.s(["Tab",()=>A])},405371,910342,e=>{"use strict";var t=e.i(290571),r=e.i(271645),n=e.i(480731);let o=(0,r.createContext)(n.BaseColors.Blue);e.s(["default",()=>o],910342);var a=e.i(970554),s=e.i(444755);let l=(0,e.i(673706).makeClassName)("TabList"),i=(0,r.createContext)("line"),u={line:(0,s.tremorTwMerge)("flex border-b space-x-4","border-tremor-border","dark:border-dark-tremor-border"),solid:(0,s.tremorTwMerge)("inline-flex p-0.5 rounded-tremor-default space-x-1.5","bg-tremor-background-subtle","dark:bg-dark-tremor-background-subtle")},c=r.default.forwardRef((e,n)=>{let{color:c,variant:d="line",children:f,className:p}=e,m=(0,t.__rest)(e,["color","variant","children","className"]);return r.default.createElement(a.Tab.List,Object.assign({ref:n,className:(0,s.tremorTwMerge)(l("root"),"justify-start overflow-x-clip",u[d],p)},m),r.default.createElement(i.Provider,{value:d},r.default.createElement(o.Provider,{value:c},f)))});c.displayName="TabList",e.s(["TabVariantContext",()=>i,"default",()=>c],405371)},197647,e=>{"use strict";var t=e.i(290571),r=e.i(970554),n=e.i(95779),o=e.i(444755),a=e.i(673706),s=e.i(271645),l=e.i(405371),i=e.i(910342);let u=(0,a.makeClassName)("Tab"),c=s.default.forwardRef((e,c)=>{let{icon:d,className:f,children:p}=e,m=(0,t.__rest)(e,["icon","className","children"]),v=(0,s.useContext)(l.TabVariantContext),b=(0,s.useContext)(i.default);return s.default.createElement(r.Tab,Object.assign({ref:c,className:(0,o.tremorTwMerge)(u("root"),"flex whitespace-nowrap truncate max-w-xs outline-none data-focus-visible:ring text-tremor-default transition duration-100",function(e,t){switch(e){case"line":return(0,o.tremorTwMerge)("data-[selected]:border-b-2 hover:border-b-2 border-transparent transition duration-100 -mb-px px-2 py-2","hover:border-tremor-content hover:text-tremor-content-emphasis text-tremor-content","[&:not([data-selected])]:dark:hover:border-dark-tremor-content-emphasis [&:not([data-selected])]:dark:hover:text-dark-tremor-content-emphasis [&:not([data-selected])]:dark:text-dark-tremor-content",t?(0,a.getColorClassNames)(t,n.colorPalette.border).selectBorderColor:["data-[selected]:border-tremor-brand data-[selected]:text-tremor-brand","data-[selected]:dark:border-dark-tremor-brand data-[selected]:dark:text-dark-tremor-brand"]);case"solid":return(0,o.tremorTwMerge)("border-transparent border rounded-tremor-small px-2.5 py-1","data-[selected]:border-tremor-border data-[selected]:bg-tremor-background data-[selected]:shadow-tremor-input [&:not([data-selected])]:hover:text-tremor-content-emphasis data-[selected]:text-tremor-brand [&:not([data-selected])]:text-tremor-content","dark:data-[selected]:border-dark-tremor-border dark:data-[selected]:bg-dark-tremor-background dark:data-[selected]:shadow-dark-tremor-input dark:[&:not([data-selected])]:hover:text-dark-tremor-content-emphasis dark:data-[selected]:text-dark-tremor-brand dark:[&:not([data-selected])]:text-dark-tremor-content",t?(0,a.getColorClassNames)(t,n.colorPalette.text).selectTextColor:"text-tremor-content dark:text-dark-tremor-content")}}(v,b),f,b&&(0,a.getColorClassNames)(b,n.colorPalette.text).selectTextColor)},m),d?s.default.createElement(d,{className:(0,o.tremorTwMerge)(u("icon"),"flex-none h-5 w-5",p?"mr-2":"")}):null,p?s.default.createElement("span",null,p):null)});c.displayName="Tab",e.s(["Tab",()=>c],197647)},653824,e=>{"use strict";var t=e.i(290571),r=e.i(970554),n=e.i(444755),o=e.i(673706),a=e.i(271645);let s=(0,o.makeClassName)("TabGroup"),l=a.default.forwardRef((e,o)=>{let{defaultIndex:l,index:i,onIndexChange:u,children:c,className:d}=e,f=(0,t.__rest)(e,["defaultIndex","index","onIndexChange","children","className"]);return a.default.createElement(r.Tab.Group,Object.assign({as:"div",ref:o,defaultIndex:l,selectedIndex:i,onChange:u,className:(0,n.tremorTwMerge)(s("root"),"w-full",d)},f),c)});l.displayName="TabGroup",e.s(["TabGroup",()=>l],653824)},881073,e=>{"use strict";var t=e.i(405371);e.s(["TabList",()=>t.default])},723731,e=>{"use strict";var t=e.i(290571),r=e.i(970554),n=e.i(751734),o=e.i(144582),a=e.i(444755),s=e.i(673706),l=e.i(271645);let i=(0,s.makeClassName)("TabPanels"),u=l.default.forwardRef((e,s)=>{let{children:u,className:c}=e,d=(0,t.__rest)(e,["children","className"]);return l.default.createElement(r.Tab.Panels,Object.assign({as:"div",ref:s,className:(0,a.tremorTwMerge)(i("root"),"w-full",c)},d),({selectedIndex:e})=>l.default.createElement(o.default.Provider,{value:{selectedValue:e}},l.default.Children.map(u,(e,t)=>l.default.createElement(n.default.Provider,{value:t},e))))});u.displayName="TabPanels",e.s(["TabPanels",()=>u],723731)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/05fcbaa2a2d4ce24.js b/litellm/proxy/_experimental/out/_next/static/chunks/05fcbaa2a2d4ce24.js deleted file mode 100644 index 3bd408347f0..00000000000 --- a/litellm/proxy/_experimental/out/_next/static/chunks/05fcbaa2a2d4ce24.js +++ /dev/null @@ -1 +0,0 @@ -(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,213205,e=>{"use strict";e.i(247167);var s=e.i(931067),t=e.i(271645);let l={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M678.3 642.4c24.2-13 51.9-20.4 81.4-20.4h.1c3 0 4.4-3.6 2.2-5.6a371.67 371.67 0 00-103.7-65.8c-.4-.2-.8-.3-1.2-.5C719.2 505 759.6 431.7 759.6 349c0-137-110.8-248-247.5-248S264.7 212 264.7 349c0 82.7 40.4 156 102.6 201.1-.4.2-.8.3-1.2.5-44.7 18.9-84.8 46-119.3 80.6a373.42 373.42 0 00-80.4 119.5A373.6 373.6 0 00137 888.8a8 8 0 008 8.2h59.9c4.3 0 7.9-3.5 8-7.8 2-77.2 32.9-149.5 87.6-204.3C357 628.2 432.2 597 512.2 597c56.7 0 111.1 15.7 158 45.1a8.1 8.1 0 008.1.3zM512.2 521c-45.8 0-88.9-17.9-121.4-50.4A171.2 171.2 0 01340.5 349c0-45.9 17.9-89.1 50.3-121.6S466.3 177 512.2 177s88.9 17.9 121.4 50.4A171.2 171.2 0 01683.9 349c0 45.9-17.9 89.1-50.3 121.6C601.1 503.1 558 521 512.2 521zM880 759h-84v-84c0-4.4-3.6-8-8-8h-56c-4.4 0-8 3.6-8 8v84h-84c-4.4 0-8 3.6-8 8v56c0 4.4 3.6 8 8 8h84v84c0 4.4 3.6 8 8 8h56c4.4 0 8-3.6 8-8v-84h84c4.4 0 8-3.6 8-8v-56c0-4.4-3.6-8-8-8z"}}]},name:"user-add",theme:"outlined"};var a=e.i(9583),r=t.forwardRef(function(e,r){return t.createElement(a.default,(0,s.default)({},e,{ref:r,icon:l}))});e.s(["UserAddOutlined",0,r],213205)},355619,e=>{"use strict";var s=e.i(764205);let t=async(e,t,l)=>{try{if(null===e||null===t)return;if(null!==l){let a=(await (0,s.modelAvailableCall)(l,e,t,!0,null,!0)).data.map(e=>e.id),r=[],i=[];return a.forEach(e=>{e.endsWith("/*")?r.push(e):i.push(e)}),[...r,...i]}}catch(e){console.error("Error fetching user models:",e)}};e.s(["fetchAvailableModelsForTeamOrKey",0,t,"getModelDisplayName",0,e=>{if("all-proxy-models"===e)return"All Proxy Models";if(e.endsWith("/*")){let s=e.replace("/*","");return`All ${s} models`}return e},"unfurlWildcardModelsInList",0,(e,s)=>{let t=[],l=[];return console.log("teamModels",e),console.log("allModels",s),e.forEach(e=>{if(e.endsWith("/*")){let a=e.replace("/*",""),r=s.filter(e=>e.startsWith(a+"/"));l.push(...r),t.push(e)}else l.push(e)}),[...t,...l].filter((e,s,t)=>t.indexOf(e)===s)}])},860585,e=>{"use strict";var s=e.i(843476),t=e.i(199133);let{Option:l}=t.Select;e.s(["default",0,({value:e,onChange:a,className:r="",style:i={}})=>(0,s.jsxs)(t.Select,{style:{width:"100%",...i},value:e||void 0,onChange:a,className:r,placeholder:"n/a",allowClear:!0,children:[(0,s.jsx)(l,{value:"1h",children:"hourly"}),(0,s.jsx)(l,{value:"24h",children:"daily"}),(0,s.jsx)(l,{value:"7d",children:"weekly"}),(0,s.jsx)(l,{value:"30d",children:"monthly"})]}),"getBudgetDurationLabel",0,e=>e?({"1h":"hourly","24h":"daily","7d":"weekly","30d":"monthly"})[e]||e:"Not set"])},285027,e=>{"use strict";e.i(247167);var s=e.i(931067),t=e.i(271645);let l={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M464 720a48 48 0 1096 0 48 48 0 10-96 0zm16-304v184c0 4.4 3.6 8 8 8h48c4.4 0 8-3.6 8-8V416c0-4.4-3.6-8-8-8h-48c-4.4 0-8 3.6-8 8zm475.7 440l-416-720c-6.2-10.7-16.9-16-27.7-16s-21.6 5.3-27.7 16l-416 720C56 877.4 71.4 904 96 904h832c24.6 0 40-26.6 27.7-48zm-783.5-27.9L512 239.9l339.8 588.2H172.2z"}}]},name:"warning",theme:"outlined"};var a=e.i(9583),r=t.forwardRef(function(e,r){return t.createElement(a.default,(0,s.default)({},e,{ref:r,icon:l}))});e.s(["WarningOutlined",0,r],285027)},447082,e=>{"use strict";var s=e.i(843476),t=e.i(271645),l=e.i(599724),a=e.i(464571),r=e.i(212931),i=e.i(291542),n=e.i(515831),d=e.i(898586),o=e.i(519756),c=e.i(737434),m=e.i(285027),u=e.i(993914),x=e.i(955135);e.i(247167);var h=e.i(931067);let p={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M854.6 288.6L639.4 73.4c-6-6-14.1-9.4-22.6-9.4H192c-17.7 0-32 14.3-32 32v832c0 17.7 14.3 32 32 32h640c17.7 0 32-14.3 32-32V311.3c0-8.5-3.4-16.7-9.4-22.7zM790.2 326H602V137.8L790.2 326zm1.8 562H232V136h302v216a42 42 0 0042 42h216v494zM472 744a40 40 0 1080 0 40 40 0 10-80 0zm16-104h48c4.4 0 8-3.6 8-8V448c0-4.4-3.6-8-8-8h-48c-4.4 0-8 3.6-8 8v184c0 4.4 3.6 8 8 8z"}}]},name:"file-exclamation",theme:"outlined"};var f=e.i(9583),g=t.forwardRef(function(e,s){return t.createElement(f.default,(0,h.default)({},e,{ref:s,icon:p}))}),j=e.i(764205),v=e.i(59935),y=e.i(220508),b=e.i(964306);let N=t.forwardRef(function(e,s){return t.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",fill:"none",viewBox:"0 0 24 24",strokeWidth:2,stroke:"currentColor","aria-hidden":"true",ref:s},e),t.createElement("path",{strokeLinecap:"round",strokeLinejoin:"round",d:"M12 9v2m0 4h.01m-6.938 4h13.856c1.54 0 2.502-1.667 1.732-3L13.732 4c-.77-1.333-2.694-1.333-3.464 0L3.34 16c-.77 1.333.192 3 1.732 3z"}))});var w=e.i(237016),_=e.i(727749);e.s(["default",0,({accessToken:e,teams:h,possibleUIRoles:p,onUsersCreated:f})=>{let[C,S]=(0,t.useState)(!1),[k,I]=(0,t.useState)([]),[T,U]=(0,t.useState)(!1),[V,B]=(0,t.useState)(null),[O,M]=(0,t.useState)(null),[F,L]=(0,t.useState)(null),[z,P]=(0,t.useState)(null),[E,A]=(0,t.useState)(null),[R,D]=(0,t.useState)("http://localhost:4000");(0,t.useEffect)(()=>{(async()=>{try{let s=await (0,j.getProxyUISettings)(e);A(s)}catch(e){console.error("Error fetching UI settings:",e)}})(),D(new URL("/",window.location.href).toString())},[e]);let $=async()=>{U(!0);let s=k.map(e=>({...e,status:"pending"}));I(s);let t=!1;for(let l=0;le.trim()).filter(Boolean),0===s.teams.length&&delete s.teams),a.models&&"string"==typeof a.models&&""!==a.models.trim()&&(s.models=a.models.split(",").map(e=>e.trim()).filter(Boolean),0===s.models.length&&delete s.models),a.max_budget&&""!==a.max_budget.toString().trim()){let e=parseFloat(a.max_budget.toString());!isNaN(e)&&e>0&&(s.max_budget=e)}a.budget_duration&&""!==a.budget_duration.trim()&&(s.budget_duration=a.budget_duration.trim()),a.metadata&&"string"==typeof a.metadata&&""!==a.metadata.trim()&&(s.metadata=a.metadata.trim()),console.log("Sending user data:",s);let r=await (0,j.userCreateCall)(e,null,s);if(console.log("Full response:",r),r&&(r.key||r.user_id)){t=!0,console.log("Success case triggered");let s=r.data?.user_id||r.user_id;try{if(E?.SSO_ENABLED){let e=new URL("/ui",R).toString();I(s=>s.map((s,t)=>t===l?{...s,status:"success",key:r.key||r.user_id,invitation_link:e}:s))}else{let t=await (0,j.invitationCreateCall)(e,s),a=new URL(`/ui?invitation_id=${t.id}`,R).toString();I(e=>e.map((e,s)=>s===l?{...e,status:"success",key:r.key||r.user_id,invitation_link:a}:e))}}catch(e){console.error("Error creating invitation:",e),I(e=>e.map((e,s)=>s===l?{...e,status:"success",key:r.key||r.user_id,error:"User created but failed to generate invitation link"}:e))}}else{console.log("Error case triggered");let e=r?.error||"Failed to create user";console.log("Error message:",e),I(s=>s.map((s,t)=>t===l?{...s,status:"failed",error:e}:s))}}catch(s){console.error("Caught error:",s);let e=s?.response?.data?.error||s?.message||String(s);I(s=>s.map((s,t)=>t===l?{...s,status:"failed",error:e}:s))}}U(!1),t&&f&&f()},W=[{title:"Row",dataIndex:"rowNumber",key:"rowNumber",width:80},{title:"Email",dataIndex:"user_email",key:"user_email"},{title:"Role",dataIndex:"user_role",key:"user_role"},{title:"Teams",dataIndex:"teams",key:"teams"},{title:"Budget",dataIndex:"max_budget",key:"max_budget"},{title:"Status",key:"status",render:(e,t)=>t.isValid?t.status&&"pending"!==t.status?"success"===t.status?(0,s.jsxs)("div",{children:[(0,s.jsxs)("div",{className:"flex items-center",children:[(0,s.jsx)(y.CheckCircleIcon,{className:"h-5 w-5 text-green-500 mr-2"}),(0,s.jsx)("span",{className:"text-green-500",children:"Success"})]}),t.invitation_link&&(0,s.jsx)("div",{className:"mt-1",children:(0,s.jsxs)("div",{className:"flex items-center",children:[(0,s.jsx)("span",{className:"text-xs text-gray-500 truncate max-w-[150px]",children:t.invitation_link}),(0,s.jsx)(w.CopyToClipboard,{text:t.invitation_link,onCopy:()=>_.default.success("Invitation link copied!"),children:(0,s.jsx)("button",{className:"ml-1 text-blue-500 text-xs hover:text-blue-700",children:"Copy"})})]})})]}):(0,s.jsxs)("div",{children:[(0,s.jsxs)("div",{className:"flex items-center",children:[(0,s.jsx)(b.XCircleIcon,{className:"h-5 w-5 text-red-500 mr-2"}),(0,s.jsx)("span",{className:"text-red-500",children:"Failed"})]}),t.error&&(0,s.jsx)("span",{className:"text-sm text-red-500 ml-7",children:JSON.stringify(t.error)})]}):(0,s.jsx)("span",{className:"text-gray-500",children:"Pending"}):(0,s.jsxs)("div",{children:[(0,s.jsxs)("div",{className:"flex items-center",children:[(0,s.jsx)(b.XCircleIcon,{className:"h-5 w-5 text-red-500 mr-2"}),(0,s.jsx)("span",{className:"text-red-500",children:"Invalid"})]}),t.error&&(0,s.jsx)("span",{className:"text-sm text-red-500 ml-7",children:t.error})]})}];return(0,s.jsxs)(s.Fragment,{children:[(0,s.jsx)(a.Button,{type:"primary",className:"mb-0",onClick:()=>S(!0),children:"+ Bulk Invite Users"}),(0,s.jsx)(r.Modal,{title:"Bulk Invite Users",open:C,width:800,onCancel:()=>S(!1),bodyStyle:{maxHeight:"70vh",overflow:"auto"},footer:null,children:(0,s.jsx)("div",{className:"flex flex-col",children:0===k.length?(0,s.jsxs)("div",{className:"mb-6",children:[(0,s.jsxs)("div",{className:"flex items-center mb-4",children:[(0,s.jsx)("div",{className:"w-8 h-8 rounded-full bg-blue-500 text-white flex items-center justify-center mr-3",children:"1"}),(0,s.jsx)("h3",{className:"text-lg font-medium",children:"Download and fill the template"})]}),(0,s.jsxs)("div",{className:"ml-11 mb-6",children:[(0,s.jsx)("p",{className:"mb-4",children:"Add multiple users at once by following these steps:"}),(0,s.jsxs)("ol",{className:"list-decimal list-inside space-y-2 ml-2 mb-4",children:[(0,s.jsx)("li",{children:"Download our CSV template"}),(0,s.jsx)("li",{children:"Add your users' information to the spreadsheet"}),(0,s.jsx)("li",{children:"Save the file and upload it here"}),(0,s.jsx)("li",{children:"After creation, download the results file containing the Virtual Keys for each user"})]}),(0,s.jsxs)("div",{className:"bg-gray-50 p-4 rounded-md border border-gray-200 mb-4",children:[(0,s.jsx)("h4",{className:"font-medium mb-2",children:"Template Column Names"}),(0,s.jsxs)("div",{className:"grid grid-cols-1 md:grid-cols-2 gap-3",children:[(0,s.jsxs)("div",{className:"flex items-start",children:[(0,s.jsx)("div",{className:"w-3 h-3 rounded-full bg-red-500 mt-1.5 mr-2 flex-shrink-0"}),(0,s.jsxs)("div",{children:[(0,s.jsx)("p",{className:"font-medium",children:"user_email"}),(0,s.jsx)("p",{className:"text-sm text-gray-600",children:"User's email address (required)"})]})]}),(0,s.jsxs)("div",{className:"flex items-start",children:[(0,s.jsx)("div",{className:"w-3 h-3 rounded-full bg-red-500 mt-1.5 mr-2 flex-shrink-0"}),(0,s.jsxs)("div",{children:[(0,s.jsx)("p",{className:"font-medium",children:"user_role"}),(0,s.jsx)("p",{className:"text-sm text-gray-600",children:'User\'s role (one of: "proxy_admin", "proxy_admin_viewer", "internal_user", "internal_user_viewer")'})]})]}),(0,s.jsxs)("div",{className:"flex items-start",children:[(0,s.jsx)("div",{className:"w-3 h-3 rounded-full bg-gray-300 mt-1.5 mr-2 flex-shrink-0"}),(0,s.jsxs)("div",{children:[(0,s.jsx)("p",{className:"font-medium",children:"teams"}),(0,s.jsx)("p",{className:"text-sm text-gray-600",children:'Comma-separated team IDs (e.g., "team-1,team-2")'})]})]}),(0,s.jsxs)("div",{className:"flex items-start",children:[(0,s.jsx)("div",{className:"w-3 h-3 rounded-full bg-gray-300 mt-1.5 mr-2 flex-shrink-0"}),(0,s.jsxs)("div",{children:[(0,s.jsx)("p",{className:"font-medium",children:"max_budget"}),(0,s.jsx)("p",{className:"text-sm text-gray-600",children:'Maximum budget as a number (e.g., "100")'})]})]}),(0,s.jsxs)("div",{className:"flex items-start",children:[(0,s.jsx)("div",{className:"w-3 h-3 rounded-full bg-gray-300 mt-1.5 mr-2 flex-shrink-0"}),(0,s.jsxs)("div",{children:[(0,s.jsx)("p",{className:"font-medium",children:"budget_duration"}),(0,s.jsx)("p",{className:"text-sm text-gray-600",children:'Budget reset period (e.g., "30d", "1mo")'})]})]}),(0,s.jsxs)("div",{className:"flex items-start",children:[(0,s.jsx)("div",{className:"w-3 h-3 rounded-full bg-gray-300 mt-1.5 mr-2 flex-shrink-0"}),(0,s.jsxs)("div",{children:[(0,s.jsx)("p",{className:"font-medium",children:"models"}),(0,s.jsx)("p",{className:"text-sm text-gray-600",children:'Comma-separated allowed models (e.g., "gpt-3.5-turbo,gpt-4")'})]})]})]})]}),(0,s.jsx)(a.Button,{type:"primary",size:"large",className:"w-full md:w-auto",icon:(0,s.jsx)(c.DownloadOutlined,{}),children:"Download CSV Template"})]}),(0,s.jsxs)("div",{className:"flex items-center mb-4",children:[(0,s.jsx)("div",{className:"w-8 h-8 rounded-full bg-blue-500 text-white flex items-center justify-center mr-3",children:"2"}),(0,s.jsx)("h3",{className:"text-lg font-medium",children:"Upload your completed CSV"})]}),(0,s.jsxs)("div",{className:"ml-11",children:[z?(0,s.jsxs)("div",{className:`mb-4 p-4 rounded-md border ${F?"bg-red-50 border-red-200":"bg-blue-50 border-blue-200"}`,children:[(0,s.jsxs)("div",{className:"flex items-center justify-between",children:[(0,s.jsxs)("div",{className:"flex items-center",children:[F?(0,s.jsx)(g,{className:"text-red-500 text-xl mr-3"}):(0,s.jsx)(u.FileTextOutlined,{className:"text-blue-500 text-xl mr-3"}),(0,s.jsxs)("div",{children:[(0,s.jsx)(d.Typography.Text,{strong:!0,className:F?"text-red-800":"text-blue-800",children:z.name}),(0,s.jsxs)(d.Typography.Text,{className:`block text-xs ${F?"text-red-600":"text-blue-600"}`,children:[(z.size/1024).toFixed(1)," KB • ",new Date().toLocaleDateString()]})]})]}),(0,s.jsx)(a.Button,{size:"small",onClick:()=>{P(null),I([]),B(null),M(null),L(null)},className:"flex items-center",icon:(0,s.jsx)(x.DeleteOutlined,{}),children:"Remove"})]}),F?(0,s.jsxs)("div",{className:"mt-3 text-red-600 text-sm flex items-start",children:[(0,s.jsx)(m.WarningOutlined,{className:"mr-2 mt-0.5"}),(0,s.jsx)("span",{children:F})]}):!O&&(0,s.jsxs)("div",{className:"mt-3 flex items-center",children:[(0,s.jsx)("div",{className:"w-full bg-gray-200 rounded-full h-1.5",children:(0,s.jsx)("div",{className:"bg-blue-500 h-1.5 rounded-full w-full animate-pulse"})}),(0,s.jsx)("span",{className:"ml-2 text-xs text-blue-600",children:"Processing..."})]})]}):(0,s.jsx)(n.Upload,{beforeUpload:e=>((B(null),M(null),L(null),P(e),"text/csv"===e.type||e.name.endsWith(".csv"))?e.size>5242880?L(`File is too large (${(e.size/1048576).toFixed(1)} MB). Please upload a CSV file smaller than 5MB.`):v.default.parse(e,{complete:e=>{if(!e.data||0===e.data.length){M("The CSV file appears to be empty. Please upload a file with data."),I([]);return}if(1===e.data.length){M("The CSV file only contains headers but no user data. Please add user data to your CSV."),I([]);return}let s=e.data[0];if(0===s.length||1===s.length&&""===s[0]){M("The CSV file doesn't contain any column headers. Please make sure your CSV has headers."),I([]);return}let t=["user_email","user_role"].filter(e=>!s.includes(e));if(t.length>0){M(`Your CSV is missing these required columns: ${t.join(", ")}. Please add these columns to your CSV file.`),I([]);return}try{let t=e.data.slice(1).map((e,t)=>{if(0===e.length||1===e.length&&""===e[0])return null;if(e.length=parseFloat(l.max_budget.toString())&&a.push("Max budget must be greater than 0")),l.budget_duration&&!l.budget_duration.match(/^\d+[dhmwy]$|^\d+mo$/)&&a.push(`Invalid budget duration format "${l.budget_duration}". Use format like "30d", "1mo", "2w", "6h"`),l.teams&&"string"==typeof l.teams&&h&&h.length>0){let e=h.map(e=>e.team_id),s=l.teams.split(",").map(e=>e.trim()).filter(s=>!e.includes(s));s.length>0&&a.push(`Unknown team(s): ${s.join(", ")}`)}return a.length>0&&(l.isValid=!1,l.error=a.join(", ")),l}).filter(Boolean),l=t.filter(e=>e.isValid);I(t),0===t.length?M("No valid data rows found in the CSV file. Please check your file format."):0===l.length?B("No valid users found in the CSV. Please check the errors below and fix your CSV file."):l.length{B(`Failed to parse CSV file: ${e.message}`),I([])},header:!1}):(L(`Invalid file type: ${e.name}. Please upload a CSV file (.csv extension).`),_.default.fromBackend("Invalid file type. Please upload a CSV file.")),!1),accept:".csv",maxCount:1,showUploadList:!1,children:(0,s.jsxs)("div",{className:"border-2 border-dashed border-gray-300 rounded-lg p-8 text-center hover:border-blue-500 transition-colors cursor-pointer",children:[(0,s.jsx)(o.UploadOutlined,{className:"text-3xl text-gray-400 mb-2"}),(0,s.jsx)("p",{className:"mb-1",children:"Drag and drop your CSV file here"}),(0,s.jsx)("p",{className:"text-sm text-gray-500 mb-3",children:"or"}),(0,s.jsx)(a.Button,{size:"small",children:"Browse files"}),(0,s.jsx)("p",{className:"text-xs text-gray-500 mt-4",children:"Only CSV files (.csv) are supported"})]})}),O&&(0,s.jsx)("div",{className:"mb-4 p-4 bg-yellow-50 border border-yellow-200 rounded-md",children:(0,s.jsxs)("div",{className:"flex items-start",children:[(0,s.jsx)(N,{className:"h-5 w-5 text-yellow-500 mr-2 mt-0.5"}),(0,s.jsxs)("div",{children:[(0,s.jsx)(d.Typography.Text,{strong:!0,className:"text-yellow-800",children:"CSV Structure Error"}),(0,s.jsx)(d.Typography.Paragraph,{className:"text-yellow-700 mt-1 mb-0",children:O}),(0,s.jsx)(d.Typography.Paragraph,{className:"text-yellow-700 mt-2 mb-0",children:"Please download our template and ensure your CSV follows the required format."})]})]})})]})]}):(0,s.jsxs)("div",{className:"mb-6",children:[(0,s.jsxs)("div",{className:"flex items-center mb-4",children:[(0,s.jsx)("div",{className:"w-8 h-8 rounded-full bg-blue-500 text-white flex items-center justify-center mr-3",children:"3"}),(0,s.jsx)("h3",{className:"text-lg font-medium",children:k.some(e=>"success"===e.status||"failed"===e.status)?"User Creation Results":"Review and create users"})]}),V&&(0,s.jsx)("div",{className:"ml-11 mb-4 p-4 bg-red-50 border border-red-200 rounded-md",children:(0,s.jsxs)("div",{className:"flex items-start",children:[(0,s.jsx)(m.WarningOutlined,{className:"text-red-500 mr-2 mt-1"}),(0,s.jsxs)("div",{children:[(0,s.jsx)(l.Text,{className:"text-red-600 font-medium",children:V}),k.some(e=>!e.isValid)&&(0,s.jsxs)("ul",{className:"mt-2 list-disc list-inside text-red-600 text-sm",children:[(0,s.jsx)("li",{children:"Check the table below for specific errors in each row"}),(0,s.jsx)("li",{children:"Common issues include invalid email formats, missing required fields, or incorrect role values"}),(0,s.jsx)("li",{children:"Fix these issues in your CSV file and upload again"})]})]})]})}),(0,s.jsxs)("div",{className:"ml-11",children:[(0,s.jsxs)("div",{className:"flex justify-between items-center mb-3",children:[(0,s.jsx)("div",{className:"flex items-center",children:k.some(e=>"success"===e.status||"failed"===e.status)?(0,s.jsxs)("div",{className:"flex items-center",children:[(0,s.jsx)(l.Text,{className:"text-lg font-medium mr-3",children:"Creation Summary"}),(0,s.jsxs)(l.Text,{className:"text-sm bg-green-100 text-green-800 px-2 py-1 rounded mr-2",children:[k.filter(e=>"success"===e.status).length," Successful"]}),k.some(e=>"failed"===e.status)&&(0,s.jsxs)(l.Text,{className:"text-sm bg-red-100 text-red-800 px-2 py-1 rounded",children:[k.filter(e=>"failed"===e.status).length," Failed"]})]}):(0,s.jsxs)("div",{className:"flex items-center",children:[(0,s.jsx)(l.Text,{className:"text-lg font-medium mr-3",children:"User Preview"}),(0,s.jsxs)(l.Text,{className:"text-sm bg-blue-100 text-blue-800 px-2 py-1 rounded",children:[k.filter(e=>e.isValid).length," of ",k.length," users valid"]})]})}),!k.some(e=>"success"===e.status||"failed"===e.status)&&(0,s.jsxs)("div",{className:"flex space-x-3",children:[(0,s.jsx)(a.Button,{onClick:()=>{I([]),B(null)},children:"Back"}),(0,s.jsx)(a.Button,{type:"primary",onClick:$,disabled:0===k.filter(e=>e.isValid).length||T,children:T?"Creating...":`Create ${k.filter(e=>e.isValid).length} Users`})]})]}),k.some(e=>"success"===e.status)&&(0,s.jsx)("div",{className:"mb-4 p-4 bg-blue-50 border border-blue-200 rounded-md",children:(0,s.jsxs)("div",{className:"flex items-start",children:[(0,s.jsx)("div",{className:"mr-3 mt-1",children:(0,s.jsx)(y.CheckCircleIcon,{className:"h-5 w-5 text-blue-500"})}),(0,s.jsxs)("div",{children:[(0,s.jsx)(l.Text,{className:"font-medium text-blue-800",children:"User creation complete"}),(0,s.jsxs)(l.Text,{className:"block text-sm text-blue-700 mt-1",children:[(0,s.jsx)("span",{className:"font-medium",children:"Next step:"})," Download the credentials file containing Virtual Keys and invitation links. Users will need these Virtual Keys to make LLM requests through LiteLLM."]})]})]})}),(0,s.jsx)(i.Table,{dataSource:k,columns:W,size:"small",pagination:{pageSize:5},scroll:{y:300},rowClassName:e=>e.isValid?"":"bg-red-50"}),!k.some(e=>"success"===e.status||"failed"===e.status)&&(0,s.jsxs)("div",{className:"flex justify-end mt-4",children:[(0,s.jsx)(a.Button,{onClick:()=>{I([]),B(null)},className:"mr-3",children:"Back"}),(0,s.jsx)(a.Button,{type:"primary",onClick:$,disabled:0===k.filter(e=>e.isValid).length||T,children:T?"Creating...":`Create ${k.filter(e=>e.isValid).length} Users`})]}),k.some(e=>"success"===e.status||"failed"===e.status)&&(0,s.jsxs)("div",{className:"flex justify-end mt-4",children:[(0,s.jsx)(a.Button,{onClick:()=>{I([]),B(null)},className:"mr-3",children:"Start New Bulk Import"}),(0,s.jsx)(a.Button,{type:"primary",onClick:()=>{let e=k.map(e=>({user_email:e.user_email,user_role:e.user_role,status:e.status,key:e.key||"",invitation_link:e.invitation_link||"",error:e.error||""})),s=new Blob([v.default.unparse(e)],{type:"text/csv"}),t=window.URL.createObjectURL(s),l=document.createElement("a");l.href=t,l.download="bulk_users_results.csv",document.body.appendChild(l),l.click(),document.body.removeChild(l),window.URL.revokeObjectURL(t)},icon:(0,s.jsx)(c.DownloadOutlined,{}),children:"Download User Credentials"})]})]})]})})})]})}],447082)},371455,172372,e=>{"use strict";var s=e.i(843476),t=e.i(827252),l=e.i(213205),a=e.i(912598),r=e.i(109799),i=e.i(677667),n=e.i(130643),d=e.i(898667),o=e.i(35983),c=e.i(779241),m=e.i(560445),u=e.i(464571),x=e.i(536916),h=e.i(808613),p=e.i(311451),f=e.i(212931),g=e.i(199133),j=e.i(770914),v=e.i(592968),y=e.i(898586),b=e.i(271645),N=e.i(447082),w=e.i(663435),_=e.i(355619),C=e.i(727749),S=e.i(764205),k=e.i(237016),I=e.i(599724);function T({isInvitationLinkModalVisible:e,setIsInvitationLinkModalVisible:t,baseUrl:l,invitationLinkData:a,modalType:r="invitation"}){let{Title:i,Paragraph:n}=y.Typography,d=()=>{if(!l)return"";let e=new URL(l).pathname,s=e&&"/"!==e?`${e}/ui`:"ui";if(a?.has_user_setup_sso)return new URL(s,l).toString();let t=`${s}?invitation_id=${a?.id}`;return"resetPassword"===r&&(t+="&action=reset_password"),new URL(t,l).toString()};return(0,s.jsxs)(f.Modal,{title:"invitation"===r?"Invitation Link":"Reset Password Link",open:e,width:800,footer:null,onOk:()=>{t(!1)},onCancel:()=>{t(!1)},children:[(0,s.jsx)(n,{children:"invitation"===r?"Copy and send the generated link to onboard this user to the proxy.":"Copy and send the generated link to the user to reset their password."}),(0,s.jsxs)("div",{className:"flex justify-between pt-5 pb-2",children:[(0,s.jsx)(I.Text,{className:"text-base",children:"User ID"}),(0,s.jsx)(I.Text,{children:a?.user_id})]}),(0,s.jsxs)("div",{className:"flex justify-between pt-5 pb-2",children:[(0,s.jsx)(I.Text,{children:"invitation"===r?"Invitation Link":"Reset Password Link"}),(0,s.jsx)(I.Text,{children:(0,s.jsx)(I.Text,{children:d()})})]}),(0,s.jsx)("div",{className:"flex justify-end mt-5",children:(0,s.jsx)(k.CopyToClipboard,{text:d(),onCopy:()=>C.default.success("Copied!"),children:(0,s.jsx)(u.Button,{type:"primary",children:"invitation"===r?"Copy invitation link":"Copy password reset link"})})})]})}e.s(["default",()=>T],172372);let{Option:U}=g.Select,{Text:V,Link:B,Title:O}=y.Typography;e.s(["CreateUserButton",0,({userID:e,accessToken:y,teams:k,possibleUIRoles:I,onUserCreated:O,isEmbedded:M=!1})=>{let F=(0,a.useQueryClient)(),[L,z]=(0,b.useState)(null),[P]=h.Form.useForm(),[E,A]=(0,b.useState)(!1),[R,D]=(0,b.useState)(!1),[$,W]=(0,b.useState)([]),[K,q]=(0,b.useState)(!1),[H,G]=(0,b.useState)(null),[J,Q]=(0,b.useState)(null),{data:X=[]}=(0,r.useOrganizations)();(0,b.useMemo)(()=>{let e=X.flatMap(e=>e.teams||[]);return e.length>0?e:k||[]},[X,k]),(0,b.useEffect)(()=>{let s=async()=>{try{let s=await (0,S.modelAvailableCall)(y,e,"any"),t=[];for(let e=0;e{try{C.default.info("Making API Call"),M||A(!0),s.models&&0!==s.models.length||"proxy_admin"===s.user_role||(s.models=["no-default-models"]),s.organization_ids&&(s.organizations=s.organization_ids,delete s.organization_ids);let t=await (0,S.userCreateCall)(y,null,s);await F.invalidateQueries({queryKey:["userList"]}),D(!0);let l=t.data?.user_id||t.user_id;if(O&&M){O(l),P.resetFields();return}if(L?.SSO_ENABLED){let s={id:"u">typeof crypto&&crypto.randomUUID?crypto.randomUUID():"xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx".replace(/[xy]/g,function(e){let s=16*Math.random()|0;return("x"==e?s:3&s|8).toString(16)}),user_id:l,is_accepted:!1,accepted_at:null,expires_at:new Date(Date.now()+6048e5),created_at:new Date,created_by:e,updated_at:new Date,updated_by:e,has_user_setup_sso:!0};G(s),q(!0)}else(0,S.invitationCreateCall)(y,l).then(e=>{e.has_user_setup_sso=!1,G(e),q(!0)});C.default.success("API user Created"),P.resetFields(),localStorage.removeItem("userData"+e)}catch(s){let e=s.response?.data?.detail||s?.message||"Error creating the user";C.default.fromBackend(e),console.error("Error creating the user:",s)}};return M?(0,s.jsxs)(h.Form,{form:P,onFinish:Y,labelCol:{span:8},wrapperCol:{span:16},labelAlign:"left",initialValues:{user_role:"internal_user_viewer",send_invite_email:!0},children:[(0,s.jsx)(m.Alert,{message:"Email invitations",description:(0,s.jsxs)(s.Fragment,{children:["New users receive an email invite only when an email integration (SMTP, Resend, or SendGrid) is configured."," ",(0,s.jsx)(B,{href:"https://docs.litellm.ai/docs/proxy/email",target:"_blank",children:"Learn how to set up email notifications"})]}),type:"info",showIcon:!0,className:"mb-4"}),(0,s.jsx)(h.Form.Item,{label:"User Email",name:"user_email",children:(0,s.jsx)(c.TextInput,{placeholder:""})}),(0,s.jsx)(h.Form.Item,{label:"User Role",name:"user_role",children:(0,s.jsx)(g.Select,{children:I&&Object.entries(I).map(([e,{ui_label:t,description:l}])=>(0,s.jsx)(o.SelectItem,{value:e,title:t,children:(0,s.jsxs)("div",{className:"flex",children:[t," ",(0,s.jsx)(V,{className:"ml-2",style:{color:"gray",fontSize:"12px"},children:l})]})},e))})}),(0,s.jsx)(h.Form.Item,{label:"Team",name:"team_id",children:(0,s.jsx)(w.default,{})}),(0,s.jsx)(h.Form.Item,{label:"Metadata",name:"metadata",children:(0,s.jsx)(p.Input.TextArea,{rows:4,placeholder:"Enter metadata as JSON"})}),(0,s.jsx)(h.Form.Item,{label:"Send invitation email",name:"send_invite_email",valuePropName:"checked",children:(0,s.jsx)(x.Checkbox,{})}),(0,s.jsx)("div",{style:{textAlign:"right",marginTop:"10px"},children:(0,s.jsx)(u.Button,{htmlType:"submit",children:"Create User"})})]}):(0,s.jsxs)("div",{className:"flex gap-2",children:[(0,s.jsx)(u.Button,{type:"primary",className:"mb-0",onClick:()=>A(!0),children:"+ Invite User"}),(0,s.jsx)(N.default,{accessToken:y,teams:k,possibleUIRoles:I}),(0,s.jsxs)(f.Modal,{title:"Invite User",open:E,width:800,footer:null,onOk:()=>{A(!1),P.resetFields()},onCancel:()=>{A(!1),D(!1),P.resetFields()},children:[(0,s.jsxs)(j.Space,{direction:"vertical",size:"middle",children:[(0,s.jsx)(V,{className:"mb-1",children:"Create a User who can own keys"}),(0,s.jsx)(m.Alert,{message:"Email invitations",description:(0,s.jsxs)(s.Fragment,{children:["New users receive an email invite only when an email integration (SMTP, Resend, or SendGrid) is configured."," ",(0,s.jsx)(B,{href:"https://docs.litellm.ai/docs/proxy/email",target:"_blank",children:"Learn how to set up email notifications"})]}),type:"info",showIcon:!0,className:"mb-4"})]}),(0,s.jsxs)(h.Form,{form:P,onFinish:Y,labelCol:{span:8},wrapperCol:{span:16},labelAlign:"left",initialValues:{user_role:"internal_user_viewer",send_invite_email:!0},children:[(0,s.jsx)(h.Form.Item,{label:"User Email",name:"user_email",children:(0,s.jsx)(p.Input,{})}),(0,s.jsx)(h.Form.Item,{label:(0,s.jsxs)("span",{children:["Global Proxy Role"," ",(0,s.jsx)(v.Tooltip,{title:"This role is independent of any team/org specific roles. Configure Team / Organization Admins in the Settings",children:(0,s.jsx)(t.InfoCircleOutlined,{})})]}),name:"user_role",children:(0,s.jsx)(g.Select,{children:I&&Object.entries(I).map(([e,{ui_label:t,description:l}])=>(0,s.jsxs)(o.SelectItem,{value:e,title:t,children:[(0,s.jsx)(V,{children:t}),(0,s.jsxs)(V,{type:"secondary",children:[" - ",l]})]},e))})}),(0,s.jsx)(h.Form.Item,{label:"Team",className:"gap-2",name:"team_id",help:"If selected, user will be added as a 'user' role to the team.",children:(0,s.jsx)(w.default,{})}),(0,s.jsx)(h.Form.Item,{label:"Organization",name:"organization_ids",help:"The user will be added to the selected organization(s).",children:(0,s.jsx)(g.Select,{mode:"multiple",placeholder:"Select Organization",style:{width:"100%"},children:X.map(e=>(0,s.jsxs)(U,{value:e.organization_id,children:[e.organization_alias," (",e.organization_id,")"]},e.organization_id))})}),(0,s.jsx)(h.Form.Item,{label:"Metadata",name:"metadata",children:(0,s.jsx)(p.Input.TextArea,{rows:4,placeholder:"Enter metadata as JSON"})}),(0,s.jsx)(h.Form.Item,{label:"Send invitation email",name:"send_invite_email",valuePropName:"checked",children:(0,s.jsx)(x.Checkbox,{})}),(0,s.jsxs)(i.Accordion,{children:[(0,s.jsx)(d.AccordionHeader,{children:(0,s.jsx)(V,{strong:!0,children:"Personal Key Creation"})}),(0,s.jsx)(n.AccordionBody,{children:(0,s.jsx)(h.Form.Item,{className:"gap-2",label:(0,s.jsxs)("span",{children:["Models"," ",(0,s.jsx)(v.Tooltip,{title:"Models user has access to, outside of team scope.",children:(0,s.jsx)(t.InfoCircleOutlined,{style:{marginLeft:"4px"}})})]}),name:"models",help:"Models user has access to, outside of team scope.",children:(0,s.jsxs)(g.Select,{mode:"multiple",placeholder:"Select models",style:{width:"100%"},children:[(0,s.jsx)(g.Select.Option,{value:"all-proxy-models",children:"All Proxy Models"},"all-proxy-models"),(0,s.jsx)(g.Select.Option,{value:"no-default-models",children:"No Default Models"},"no-default-models"),$.map(e=>(0,s.jsx)(g.Select.Option,{value:e,children:(0,_.getModelDisplayName)(e)},e))]})})})]}),(0,s.jsx)("div",{style:{textAlign:"right",marginTop:"10px"},children:(0,s.jsx)(u.Button,{type:"primary",icon:(0,s.jsx)(l.UserAddOutlined,{}),htmlType:"submit",children:"Invite User"})})]})]}),R&&(0,s.jsx)(T,{isInvitationLinkModalVisible:K,setIsInvitationLinkModalVisible:q,baseUrl:J||"",invitationLinkData:H})]})}],371455)}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/0606c92ecd600e0c.js b/litellm/proxy/_experimental/out/_next/static/chunks/0606c92ecd600e0c.js deleted file mode 100644 index 4d144c04ad4..00000000000 --- a/litellm/proxy/_experimental/out/_next/static/chunks/0606c92ecd600e0c.js +++ /dev/null @@ -1,420 +0,0 @@ -(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,492030,e=>{"use strict";var t=e.i(121229);e.s(["CheckOutlined",()=>t.default])},166406,e=>{"use strict";var t=e.i(190144);e.s(["CopyOutlined",()=>t.default])},447566,e=>{"use strict";e.i(247167);var t=e.i(931067),i=e.i(271645);let a={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M872 474H286.9l350.2-304c5.6-4.9 2.2-14-5.2-14h-88.5c-3.9 0-7.6 1.4-10.5 3.9L155 487.8a31.96 31.96 0 000 48.3L535.1 866c1.5 1.3 3.3 2 5.2 2h91.5c7.4 0 10.8-9.2 5.2-14L286.9 550H872c4.4 0 8-3.6 8-8v-60c0-4.4-3.6-8-8-8z"}}]},name:"arrow-left",theme:"outlined"};var o=e.i(9583),r=i.forwardRef(function(e,r){return i.createElement(o.default,(0,t.default)({},e,{ref:r,icon:a}))});e.s(["ArrowLeftOutlined",0,r],447566)},596239,e=>{"use strict";e.i(247167);var t=e.i(931067),i=e.i(271645);let a={icon:{tag:"svg",attrs:{viewBox:"64 64 896 896",focusable:"false"},children:[{tag:"path",attrs:{d:"M574 665.4a8.03 8.03 0 00-11.3 0L446.5 781.6c-53.8 53.8-144.6 59.5-204 0-59.5-59.5-53.8-150.2 0-204l116.2-116.2c3.1-3.1 3.1-8.2 0-11.3l-39.8-39.8a8.03 8.03 0 00-11.3 0L191.4 526.5c-84.6 84.6-84.6 221.5 0 306s221.5 84.6 306 0l116.2-116.2c3.1-3.1 3.1-8.2 0-11.3L574 665.4zm258.6-474c-84.6-84.6-221.5-84.6-306 0L410.3 307.6a8.03 8.03 0 000 11.3l39.7 39.7c3.1 3.1 8.2 3.1 11.3 0l116.2-116.2c53.8-53.8 144.6-59.5 204 0 59.5 59.5 53.8 150.2 0 204L665.3 562.6a8.03 8.03 0 000 11.3l39.8 39.8c3.1 3.1 8.2 3.1 11.3 0l116.2-116.2c84.5-84.6 84.5-221.5 0-306.1zM610.1 372.3a8.03 8.03 0 00-11.3 0L372.3 598.7a8.03 8.03 0 000 11.3l39.6 39.6c3.1 3.1 8.2 3.1 11.3 0l226.4-226.4c3.1-3.1 3.1-8.2 0-11.3l-39.5-39.6z"}}]},name:"link",theme:"outlined"};var o=e.i(9583),r=i.forwardRef(function(e,r){return i.createElement(o.default,(0,t.default)({},e,{ref:r,icon:a}))});e.s(["LinkOutlined",0,r],596239)},190272,785913,e=>{"use strict";var t,i,a=((t={}).AUDIO_SPEECH="audio_speech",t.AUDIO_TRANSCRIPTION="audio_transcription",t.IMAGE_GENERATION="image_generation",t.VIDEO_GENERATION="video_generation",t.CHAT="chat",t.RESPONSES="responses",t.IMAGE_EDITS="image_edits",t.ANTHROPIC_MESSAGES="anthropic_messages",t.EMBEDDING="embedding",t),o=((i={}).IMAGE="image",i.VIDEO="video",i.CHAT="chat",i.RESPONSES="responses",i.IMAGE_EDITS="image_edits",i.ANTHROPIC_MESSAGES="anthropic_messages",i.EMBEDDINGS="embeddings",i.SPEECH="speech",i.TRANSCRIPTION="transcription",i.A2A_AGENTS="a2a_agents",i.MCP="mcp",i.REALTIME="realtime",i);let r={image_generation:"image",video_generation:"video",chat:"chat",responses:"responses",image_edits:"image_edits",anthropic_messages:"anthropic_messages",audio_speech:"speech",audio_transcription:"transcription",embedding:"embeddings"};e.s(["EndpointType",()=>o,"getEndpointType",0,e=>{if(console.log("getEndpointType:",e),Object.values(a).includes(e)){let t=r[e];return console.log("endpointType:",t),t}return"chat"}],785913),e.s(["generateCodeSnippet",0,e=>{let t,{apiKeySource:i,accessToken:a,apiKey:r,inputMessage:n,chatHistory:s,selectedTags:l,selectedVectorStores:p,selectedGuardrails:d,selectedPolicies:c,selectedMCPServers:g,mcpServers:m,mcpServerToolRestrictions:u,selectedVoice:f,endpointType:_,selectedModel:h,selectedSdk:A,proxySettings:x}=e,b="session"===i?a:r,y=window.location.origin,I=x?.LITELLM_UI_API_DOC_BASE_URL;I&&I.trim()?y=I:x?.PROXY_BASE_URL&&(y=x.PROXY_BASE_URL);let v=n||"Your prompt here",C=v.replace(/\\/g,"\\\\").replace(/"/g,'\\"').replace(/\n/g,"\\n"),E=s.filter(e=>!e.isImage).map(({role:e,content:t})=>({role:e,content:t})),O={};l.length>0&&(O.tags=l),p.length>0&&(O.vector_stores=p),d.length>0&&(O.guardrails=d),c.length>0&&(O.policies=c);let T=h||"your-model-name",S="azure"===A?`import openai - -client = openai.AzureOpenAI( - api_key="${b||"YOUR_LITELLM_API_KEY"}", - azure_endpoint="${y}", - api_version="2024-02-01" -)`:`import openai - -client = openai.OpenAI( - api_key="${b||"YOUR_LITELLM_API_KEY"}", - base_url="${y}" -)`;switch(_){case o.CHAT:{let e=Object.keys(O).length>0,i="";if(e){let e=JSON.stringify({metadata:O},null,2).split("\n").map(e=>" ".repeat(4)+e).join("\n").trim();i=`, - extra_body=${e}`}let a=E.length>0?E:[{role:"user",content:v}];t=` -import base64 - -# Helper function to encode images to base64 -def encode_image(image_path): - with open(image_path, "rb") as image_file: - return base64.b64encode(image_file.read()).decode('utf-8') - -# Example with text only -response = client.chat.completions.create( - model="${T}", - messages=${JSON.stringify(a,null,4)}${i} -) - -print(response) - -# Example with image or PDF (uncomment and provide file path to use) -# base64_file = encode_image("path/to/your/file.jpg") # or .pdf -# response_with_file = client.chat.completions.create( -# model="${T}", -# messages=[ -# { -# "role": "user", -# "content": [ -# { -# "type": "text", -# "text": "${C}" -# }, -# { -# "type": "image_url", -# "image_url": { -# "url": f"data:image/jpeg;base64,{base64_file}" # or data:application/pdf;base64,{base64_file} -# } -# } -# ] -# } -# ]${i} -# ) -# print(response_with_file) -`;break}case o.RESPONSES:{let e=Object.keys(O).length>0,i="";if(e){let e=JSON.stringify({metadata:O},null,2).split("\n").map(e=>" ".repeat(4)+e).join("\n").trim();i=`, - extra_body=${e}`}let a=E.length>0?E:[{role:"user",content:v}];t=` -import base64 - -# Helper function to encode images to base64 -def encode_image(image_path): - with open(image_path, "rb") as image_file: - return base64.b64encode(image_file.read()).decode('utf-8') - -# Example with text only -response = client.responses.create( - model="${T}", - input=${JSON.stringify(a,null,4)}${i} -) - -print(response.output_text) - -# Example with image or PDF (uncomment and provide file path to use) -# base64_file = encode_image("path/to/your/file.jpg") # or .pdf -# response_with_file = client.responses.create( -# model="${T}", -# input=[ -# { -# "role": "user", -# "content": [ -# {"type": "input_text", "text": "${C}"}, -# { -# "type": "input_image", -# "image_url": f"data:image/jpeg;base64,{base64_file}", # or data:application/pdf;base64,{base64_file} -# }, -# ], -# } -# ]${i} -# ) -# print(response_with_file.output_text) -`;break}case o.IMAGE:t="azure"===A?` -# NOTE: The Azure SDK does not have a direct equivalent to the multi-modal 'responses.create' method shown for OpenAI. -# This snippet uses 'client.images.generate' and will create a new image based on your prompt. -# It does not use the uploaded image, as 'client.images.generate' does not support image inputs in this context. -import os -import requests -import json -import time -from PIL import Image - -result = client.images.generate( - model="${T}", - prompt="${n}", - n=1 -) - -json_response = json.loads(result.model_dump_json()) - -# Set the directory for the stored image -image_dir = os.path.join(os.curdir, 'images') - -# If the directory doesn't exist, create it -if not os.path.isdir(image_dir): - os.mkdir(image_dir) - -# Initialize the image path -image_filename = f"generated_image_{int(time.time())}.png" -image_path = os.path.join(image_dir, image_filename) - -try: - # Retrieve the generated image - if json_response.get("data") && len(json_response["data"]) > 0 && json_response["data"][0].get("url"): - image_url = json_response["data"][0]["url"] - generated_image = requests.get(image_url).content - with open(image_path, "wb") as image_file: - image_file.write(generated_image) - - print(f"Image saved to {image_path}") - # Display the image - image = Image.open(image_path) - image.show() - else: - print("Could not find image URL in response.") - print("Full response:", json_response) -except Exception as e: - print(f"An error occurred: {e}") - print("Full response:", json_response) -`:` -import base64 -import os -import time -import json -from PIL import Image -import requests - -# Helper function to encode images to base64 -def encode_image(image_path): - with open(image_path, "rb") as image_file: - return base64.b64encode(image_file.read()).decode('utf-8') - -# Helper function to create a file (simplified for this example) -def create_file(image_path): - # In a real implementation, this would upload the file to OpenAI - # For this example, we'll just return a placeholder ID - return f"file_{os.path.basename(image_path).replace('.', '_')}" - -# The prompt entered by the user -prompt = "${C}" - -# Encode images to base64 -base64_image1 = encode_image("body-lotion.png") -base64_image2 = encode_image("soap.png") - -# Create file IDs -file_id1 = create_file("body-lotion.png") -file_id2 = create_file("incense-kit.png") - -response = client.responses.create( - model="${T}", - input=[ - { - "role": "user", - "content": [ - {"type": "input_text", "text": prompt}, - { - "type": "input_image", - "image_url": f"data:image/jpeg;base64,{base64_image1}", - }, - { - "type": "input_image", - "image_url": f"data:image/jpeg;base64,{base64_image2}", - }, - { - "type": "input_image", - "file_id": file_id1, - }, - { - "type": "input_image", - "file_id": file_id2, - } - ], - } - ], - tools=[{"type": "image_generation"}], -) - -# Process the response -image_generation_calls = [ - output - for output in response.output - if output.type == "image_generation_call" -] - -image_data = [output.result for output in image_generation_calls] - -if image_data: - image_base64 = image_data[0] - image_filename = f"edited_image_{int(time.time())}.png" - with open(image_filename, "wb") as f: - f.write(base64.b64decode(image_base64)) - print(f"Image saved to {image_filename}") -else: - # If no image is generated, there might be a text response with an explanation - text_response = [output.text for output in response.output if hasattr(output, 'text')] - if text_response: - print("No image generated. Model response:") - print("\\n".join(text_response)) - else: - print("No image data found in response.") - print("Full response for debugging:") - print(response) -`;break;case o.IMAGE_EDITS:t="azure"===A?` -import base64 -import os -import time -import json -from PIL import Image -import requests - -# Helper function to encode images to base64 -def encode_image(image_path): - with open(image_path, "rb") as image_file: - return base64.b64encode(image_file.read()).decode('utf-8') - -# The prompt entered by the user -prompt = "${C}" - -# Encode images to base64 -base64_image1 = encode_image("body-lotion.png") -base64_image2 = encode_image("soap.png") - -# Create file IDs -file_id1 = create_file("body-lotion.png") -file_id2 = create_file("incense-kit.png") - -response = client.responses.create( - model="${T}", - input=[ - { - "role": "user", - "content": [ - {"type": "input_text", "text": prompt}, - { - "type": "input_image", - "image_url": f"data:image/jpeg;base64,{base64_image1}", - }, - { - "type": "input_image", - "image_url": f"data:image/jpeg;base64,{base64_image2}", - }, - { - "type": "input_image", - "file_id": file_id1, - }, - { - "type": "input_image", - "file_id": file_id2, - } - ], - } - ], - tools=[{"type": "image_generation"}], -) - -# Process the response -image_generation_calls = [ - output - for output in response.output - if output.type == "image_generation_call" -] - -image_data = [output.result for output in image_generation_calls] - -if image_data: - image_base64 = image_data[0] - image_filename = f"edited_image_{int(time.time())}.png" - with open(image_filename, "wb") as f: - f.write(base64.b64decode(image_base64)) - print(f"Image saved to {image_filename}") -else: - # If no image is generated, there might be a text response with an explanation - text_response = [output.text for output in response.output if hasattr(output, 'text')] - if text_response: - print("No image generated. Model response:") - print("\\n".join(text_response)) - else: - print("No image data found in response.") - print("Full response for debugging:") - print(response) -`:` -import base64 -import os -import time - -# Helper function to encode images to base64 -def encode_image(image_path): - with open(image_path, "rb") as image_file: - return base64.b64encode(image_file.read()).decode('utf-8') - -# Helper function to create a file (simplified for this example) -def create_file(image_path): - # In a real implementation, this would upload the file to OpenAI - # For this example, we'll just return a placeholder ID - return f"file_{os.path.basename(image_path).replace('.', '_')}" - -# The prompt entered by the user -prompt = "${C}" - -# Encode images to base64 -base64_image1 = encode_image("body-lotion.png") -base64_image2 = encode_image("soap.png") - -# Create file IDs -file_id1 = create_file("body-lotion.png") -file_id2 = create_file("incense-kit.png") - -response = client.responses.create( - model="${T}", - input=[ - { - "role": "user", - "content": [ - {"type": "input_text", "text": prompt}, - { - "type": "input_image", - "image_url": f"data:image/jpeg;base64,{base64_image1}", - }, - { - "type": "input_image", - "image_url": f"data:image/jpeg;base64,{base64_image2}", - }, - { - "type": "input_image", - "file_id": file_id1, - }, - { - "type": "input_image", - "file_id": file_id2, - } - ], - } - ], - tools=[{"type": "image_generation"}], -) - -# Process the response -image_generation_calls = [ - output - for output in response.output - if output.type == "image_generation_call" -] - -image_data = [output.result for output in image_generation_calls] - -if image_data: - image_base64 = image_data[0] - image_filename = f"edited_image_{int(time.time())}.png" - with open(image_filename, "wb") as f: - f.write(base64.b64decode(image_base64)) - print(f"Image saved to {image_filename}") -else: - # If no image is generated, there might be a text response with an explanation - text_response = [output.text for output in response.output if hasattr(output, 'text')] - if text_response: - print("No image generated. Model response:") - print("\\n".join(text_response)) - else: - print("No image data found in response.") - print("Full response for debugging:") - print(response) -`;break;case o.EMBEDDINGS:t=` -response = client.embeddings.create( - input="${n||"Your string here"}", - model="${T}", - encoding_format="base64" # or "float" -) - -print(response.data[0].embedding) -`;break;case o.TRANSCRIPTION:t=` -# Open the audio file -audio_file = open("path/to/your/audio/file.mp3", "rb") - -# Make the transcription request -response = client.audio.transcriptions.create( - model="${T}", - file=audio_file${n?`, - prompt="${n.replace(/\\/g,"\\\\").replace(/"/g,'\\"')}"`:""} -) - -print(response.text) -`;break;case o.SPEECH:t=` -# Make the text-to-speech request -response = client.audio.speech.create( - model="${T}", - input="${n||"Your text to convert to speech here"}", - voice="${f}" # Options: alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer -) - -# Save the audio to a file -output_filename = "output_speech.mp3" -response.stream_to_file(output_filename) -print(f"Audio saved to {output_filename}") - -# Optional: Customize response format and speed -# response = client.audio.speech.create( -# model="${T}", -# input="${n||"Your text to convert to speech here"}", -# voice="alloy", -# response_format="mp3", # Options: mp3, opus, aac, flac, wav, pcm -# speed=1.0 # Range: 0.25 to 4.0 -# ) -# response.stream_to_file("output_speech.mp3") -`;break;default:t="\n# Code generation for this endpoint is not implemented yet."}return`${S} -${t}`}],190272)},652272,209261,e=>{"use strict";var t=e.i(843476),i=e.i(271645),a=e.i(447566),o=e.i(166406),r=e.i(492030),n=e.i(596239);let s=e=>"github"===e.source.source&&e.source.repo?`/plugin marketplace add ${e.source.repo}`:"url"===e.source.source&&e.source.url?`/plugin marketplace add ${e.source.url}`:`/plugin marketplace add ${e.name}`;e.s(["formatInstallCommand",0,s,"getCategoryBadgeColor",0,e=>{if(!e)return"gray";let t=e.toLowerCase();if(t.includes("development")||t.includes("dev"))return"blue";if(t.includes("productivity")||t.includes("workflow"))return"green";if(t.includes("learning")||t.includes("education"))return"purple";if(t.includes("security")||t.includes("safety"))return"red";if(t.includes("data")||t.includes("analytics"))return"orange";else if(t.includes("integration")||t.includes("api"))return"yellow";return"gray"},"isValidEmail",0,e=>!e||/^[^\s@]+@[^\s@]+\.[^\s@]+$/.test(e),"isValidSemanticVersion",0,e=>!e||/^\d+\.\d+\.\d+(-[a-zA-Z0-9.-]+)?(\+[a-zA-Z0-9.-]+)?$/.test(e),"isValidUrl",0,e=>{if(!e)return!0;try{return new URL(e),!0}catch{return!1}},"parseKeywords",0,e=>e&&""!==e.trim()?e.split(",").map(e=>e.trim()).filter(e=>""!==e):[],"validatePluginName",0,e=>!!e&&""!==e.trim()&&/^[a-z0-9-]+$/.test(e)],209261),e.s(["default",0,({skill:e,onBack:l})=>{let p,[d,c]=(0,i.useState)("overview"),[g,m]=(0,i.useState)(null),u=(e,t)=>{navigator.clipboard.writeText(e),m(t),setTimeout(()=>m(null),2e3)},f="github"===(p=e.source).source&&p.repo?`https://github.com/${p.repo}`:"git-subdir"===p.source&&p.url?p.path?`${p.url}/tree/main/${p.path}`:p.url:"url"===p.source&&p.url?p.url:null,_=s(e),h=[...e.category?[{property:"Category",value:e.category}]:[],...e.domain?[{property:"Domain",value:e.domain}]:[],...e.namespace?[{property:"Namespace",value:e.namespace}]:[],...e.version?[{property:"Version",value:e.version}]:[],...e.author?.name?[{property:"Author",value:e.author.name}]:[],...e.created_at?[{property:"Added",value:new Date(e.created_at).toLocaleDateString()}]:[]];return(0,t.jsxs)("div",{style:{padding:"24px 32px 24px 0"},children:[(0,t.jsxs)("div",{onClick:l,style:{display:"inline-flex",alignItems:"center",gap:6,color:"#5f6368",cursor:"pointer",fontSize:14,marginBottom:24},children:[(0,t.jsx)(a.ArrowLeftOutlined,{style:{fontSize:11}}),(0,t.jsx)("span",{children:"Skills"})]}),(0,t.jsxs)("div",{style:{marginBottom:8},children:[(0,t.jsx)("h1",{style:{fontSize:28,fontWeight:400,color:"#202124",margin:0,lineHeight:1.2},children:e.name}),e.description&&(0,t.jsx)("p",{style:{fontSize:14,color:"#5f6368",margin:"8px 0 0 0",lineHeight:1.6},children:e.description})]}),(0,t.jsx)("div",{style:{borderBottom:"1px solid #dadce0",marginBottom:28,marginTop:24},children:(0,t.jsx)("div",{style:{display:"flex",gap:0},children:[{key:"overview",label:"Overview"},{key:"usage",label:"How to Use"}].map(e=>(0,t.jsx)("div",{onClick:()=>c(e.key),style:{padding:"12px 20px",fontSize:14,color:d===e.key?"#1a73e8":"#5f6368",borderBottom:d===e.key?"3px solid #1a73e8":"3px solid transparent",cursor:"pointer",fontWeight:d===e.key?500:400,marginBottom:-1},children:e.label},e.key))})}),"overview"===d&&(0,t.jsxs)("div",{style:{display:"flex",gap:64},children:[(0,t.jsxs)("div",{style:{flex:1,minWidth:0},children:[(0,t.jsx)("h2",{style:{fontSize:18,fontWeight:400,color:"#202124",margin:"0 0 4px 0"},children:"Skill Details"}),(0,t.jsx)("p",{style:{fontSize:13,color:"#5f6368",margin:"0 0 16px 0"},children:"Metadata registered with this skill"}),(0,t.jsxs)("table",{style:{width:"100%",borderCollapse:"collapse",fontSize:14},children:[(0,t.jsx)("thead",{children:(0,t.jsxs)("tr",{style:{borderBottom:"1px solid #dadce0"},children:[(0,t.jsx)("th",{style:{textAlign:"left",padding:"12px 0",color:"#5f6368",fontWeight:500,width:160},children:"Property"}),(0,t.jsx)("th",{style:{textAlign:"left",padding:"12px 0",color:"#5f6368",fontWeight:500},children:e.name})]})}),(0,t.jsx)("tbody",{children:h.map((e,i)=>(0,t.jsxs)("tr",{style:{borderBottom:"1px solid #f1f3f4"},children:[(0,t.jsx)("td",{style:{padding:"12px 0",color:"#3c4043"},children:e.property}),(0,t.jsx)("td",{style:{padding:"12px 0",color:"#202124"},children:e.value})]},i))})]})]}),(0,t.jsxs)("div",{style:{width:240,flexShrink:0},children:[(0,t.jsxs)("div",{style:{marginBottom:24},children:[(0,t.jsx)("div",{style:{fontSize:12,color:"#5f6368",marginBottom:4},children:"Status"}),(0,t.jsx)("span",{style:{fontSize:12,padding:"3px 10px",borderRadius:12,backgroundColor:e.enabled?"#e6f4ea":"#f1f3f4",color:e.enabled?"#137333":"#5f6368",fontWeight:500},children:e.enabled?"Public":"Draft"})]}),f&&(0,t.jsxs)("div",{style:{marginBottom:24},children:[(0,t.jsx)("div",{style:{fontSize:12,color:"#5f6368",marginBottom:4},children:"Source"}),(0,t.jsxs)("a",{href:f,target:"_blank",rel:"noopener noreferrer",style:{fontSize:13,color:"#1a73e8",wordBreak:"break-all",display:"flex",alignItems:"center",gap:4},children:[f.replace("https://",""),(0,t.jsx)(n.LinkOutlined,{style:{fontSize:11,flexShrink:0}})]})]}),e.keywords&&e.keywords.length>0&&(0,t.jsxs)("div",{style:{marginBottom:24},children:[(0,t.jsx)("div",{style:{fontSize:12,color:"#5f6368",marginBottom:8},children:"Tags"}),(0,t.jsx)("div",{style:{display:"flex",flexWrap:"wrap",gap:6},children:e.keywords.map(e=>(0,t.jsx)("span",{style:{fontSize:12,padding:"4px 12px",borderRadius:16,border:"1px solid #dadce0",color:"#3c4043",backgroundColor:"#fff"},children:e},e))})]}),(0,t.jsxs)("div",{children:[(0,t.jsx)("div",{style:{fontSize:12,color:"#5f6368",marginBottom:4},children:"Skill ID"}),(0,t.jsx)("div",{style:{fontSize:12,fontFamily:"monospace",color:"#3c4043",wordBreak:"break-all"},children:e.id})]})]})]}),"usage"===d&&(0,t.jsxs)("div",{style:{maxWidth:640},children:[(0,t.jsx)("h2",{style:{fontSize:18,fontWeight:400,color:"#202124",margin:"0 0 8px 0"},children:"Using this skill"}),(0,t.jsx)("p",{style:{fontSize:14,color:"#5f6368",margin:"0 0 24px 0",lineHeight:1.6},children:"Once your proxy is set as a marketplace, enable this skill in Claude Code with one command:"}),(0,t.jsxs)("div",{style:{border:"1px solid #dadce0",borderRadius:8,overflow:"hidden",marginBottom:24},children:[(0,t.jsxs)("div",{style:{display:"flex",alignItems:"center",justifyContent:"space-between",padding:"10px 16px",backgroundColor:"#f8f9fa",borderBottom:"1px solid #dadce0"},children:[(0,t.jsx)("span",{style:{fontSize:13,color:"#3c4043",fontWeight:500},children:"Run in Claude Code"}),(0,t.jsxs)("button",{onClick:()=>u(_,"install"),style:{display:"flex",alignItems:"center",gap:4,fontSize:12,color:"install"===g?"#137333":"#1a73e8",background:"none",border:"none",cursor:"pointer",padding:0},children:["install"===g?(0,t.jsx)(r.CheckOutlined,{}):(0,t.jsx)(o.CopyOutlined,{}),"install"===g?"Copied":"Copy"]})]}),(0,t.jsx)("pre",{style:{margin:0,padding:"14px 16px",fontSize:14,fontFamily:"monospace",color:"#202124",backgroundColor:"#fff"},children:_})]}),(0,t.jsxs)("p",{style:{fontSize:13,color:"#5f6368",lineHeight:1.6,margin:0},children:["Don't have the marketplace configured yet?"," ",(0,t.jsx)("span",{onClick:()=>c("setup"),style:{color:"#1a73e8",cursor:"pointer"},children:"See one-time setup →"})]})]}),"setup"===d&&(0,t.jsxs)("div",{style:{maxWidth:640},children:[(0,t.jsx)("h2",{style:{fontSize:18,fontWeight:400,color:"#202124",margin:"0 0 8px 0"},children:"One-time marketplace setup"}),(0,t.jsxs)("p",{style:{fontSize:14,color:"#5f6368",margin:"0 0 24px 0",lineHeight:1.6},children:["Add this to ",(0,t.jsx)("code",{style:{fontSize:13,backgroundColor:"#f1f3f4",padding:"1px 6px",borderRadius:4},children:"~/.claude/settings.json"})," to point Claude Code at your proxy:"]}),(0,t.jsxs)("div",{style:{border:"1px solid #dadce0",borderRadius:8,overflow:"hidden"},children:[(0,t.jsxs)("div",{style:{display:"flex",alignItems:"center",justifyContent:"space-between",padding:"10px 16px",backgroundColor:"#f8f9fa",borderBottom:"1px solid #dadce0"},children:[(0,t.jsx)("span",{style:{fontSize:13,color:"#3c4043",fontWeight:500},children:"~/.claude/settings.json"}),(0,t.jsxs)("button",{onClick:()=>{u(JSON.stringify({extraKnownMarketplaces:{"my-org":{source:"url",url:`${window.location.origin}/claude-code/marketplace.json`}}},null,2),"settings")},style:{display:"flex",alignItems:"center",gap:4,fontSize:12,color:"settings"===g?"#137333":"#1a73e8",background:"none",border:"none",cursor:"pointer",padding:0},children:["settings"===g?(0,t.jsx)(r.CheckOutlined,{}):(0,t.jsx)(o.CopyOutlined,{}),"settings"===g?"Copied":"Copy"]})]}),(0,t.jsx)("pre",{style:{margin:0,padding:"14px 16px",fontSize:13,fontFamily:"monospace",color:"#202124",backgroundColor:"#fff"},children:JSON.stringify({extraKnownMarketplaces:{"my-org":{source:"url",url:`${window.location.origin}/claude-code/marketplace.json`}}},null,2)})]})]})]})}],652272)},879664,e=>{"use strict";let t=(0,e.i(475254).default)("info",[["circle",{cx:"12",cy:"12",r:"10",key:"1mglay"}],["path",{d:"M12 16v-4",key:"1dtifu"}],["path",{d:"M12 8h.01",key:"e9boi3"}]]);e.s(["default",()=>t])},916925,e=>{"use strict";var t,i=((t={}).A2A_Agent="A2A Agent",t.AI21="Ai21",t.AI21_CHAT="Ai21 Chat",t.AIML="AI/ML API",t.AIOHTTP_OPENAI="Aiohttp Openai",t.Anthropic="Anthropic",t.ANTHROPIC_TEXT="Anthropic Text",t.AssemblyAI="AssemblyAI",t.AUTO_ROUTER="Auto Router",t.Bedrock="Amazon Bedrock",t.BedrockMantle="Amazon Bedrock Mantle",t.SageMaker="AWS SageMaker",t.Azure="Azure",t.Azure_AI_Studio="Azure AI Foundry (Studio)",t.AZURE_TEXT="Azure Text",t.BASETEN="Baseten",t.BYTEZ="Bytez",t.Cerebras="Cerebras",t.CLARIFAI="Clarifai",t.CLOUDFLARE="Cloudflare",t.CODESTRAL="Codestral",t.Cohere="Cohere",t.COHERE_CHAT="Cohere Chat",t.COMETAPI="Cometapi",t.COMPACTIFAI="Compactifai",t.Cursor="Cursor",t.Dashscope="Dashscope",t.Databricks="Databricks (Qwen API)",t.DATAROBOT="Datarobot",t.DeepInfra="DeepInfra",t.Deepgram="Deepgram",t.Deepseek="Deepseek",t.DOCKER_MODEL_RUNNER="Docker Model Runner",t.DOTPROMPT="Dotprompt",t.ElevenLabs="ElevenLabs",t.EMPOWER="Empower",t.FalAI="Fal AI",t.FEATHERLESS_AI="Featherless Ai",t.FireworksAI="Fireworks AI",t.FRIENDLIAI="Friendliai",t.GALADRIEL="Galadriel",t.GITHUB_COPILOT="Github Copilot",t.Google_AI_Studio="Google AI Studio",t.GradientAI="GradientAI",t.Groq="Groq",t.HEROKU="Heroku",t.Hosted_Vllm="vllm",t.HUGGINGFACE="Huggingface",t.HYPERBOLIC="Hyperbolic",t.Infinity="Infinity",t.JinaAI="Jina AI",t.LAMBDA_AI="Lambda Ai",t.LEMONADE="Lemonade",t.LLAMAFILE="Llamafile",t.LM_STUDIO="Lm Studio",t.LLAMA="Meta Llama",t.MARITALK="Maritalk",t.MiniMax="MiniMax",t.MistralAI="Mistral AI",t.MOONSHOT="Moonshot",t.MORPH="Morph",t.NEBIUS="Nebius",t.NLP_CLOUD="Nlp Cloud",t.NOVITA="Novita",t.NSCALE="Nscale",t.NVIDIA_NIM="Nvidia Nim",t.Ollama="Ollama",t.OLLAMA_CHAT="Ollama Chat",t.OOBABOOGA="Oobabooga",t.OpenAI="OpenAI",t.OPENAI_LIKE="Openai Like",t.OpenAI_Compatible="OpenAI-Compatible Endpoints (Together AI, etc.)",t.OpenAI_Text="OpenAI Text Completion",t.OpenAI_Text_Compatible="OpenAI-Compatible Text Completion Models (Together AI, etc.)",t.Openrouter="Openrouter",t.Oracle="Oracle Cloud Infrastructure (OCI)",t.OVHCLOUD="Ovhcloud",t.Perplexity="Perplexity",t.PETALS="Petals",t.PG_VECTOR="Pg Vector",t.PREDIBASE="Predibase",t.RECRAFT="Recraft",t.REPLICATE="Replicate",t.RunwayML="RunwayML",t.SAGEMAKER_LEGACY="Sagemaker",t.Sambanova="Sambanova",t.SAP="SAP Generative AI Hub",t.Snowflake="Snowflake",t.TEXT_COMPLETION_CODESTRAL="Text-Completion-Codestral",t.TogetherAI="TogetherAI",t.TOPAZ="Topaz",t.Triton="Triton",t.V0="V0",t.VERCEL_AI_GATEWAY="Vercel Ai Gateway",t.Vertex_AI="Vertex AI (Anthropic, Gemini, etc.)",t.VERTEX_AI_BETA="Vertex Ai Beta",t.VLLM="Vllm",t.VolcEngine="VolcEngine",t.Voyage="Voyage AI",t.WANDB="Wandb",t.WATSONX="Watsonx",t.WATSONX_TEXT="Watsonx Text",t.xAI="xAI",t.XINFERENCE="Xinference",t.ZAI="Z.AI (Zhipu AI)",t);let a={A2A_Agent:"a2a_agent",AI21:"ai21",AI21_CHAT:"ai21_chat",AIML:"aiml",AIOHTTP_OPENAI:"aiohttp_openai",Anthropic:"anthropic",ANTHROPIC_TEXT:"anthropic_text",AssemblyAI:"assemblyai",AUTO_ROUTER:"auto_router",Azure:"azure",Azure_AI_Studio:"azure_ai",AZURE_TEXT:"azure_text",BASETEN:"baseten",Bedrock:"bedrock",BedrockMantle:"bedrock_mantle",BYTEZ:"bytez",Cerebras:"cerebras",CLARIFAI:"clarifai",CLOUDFLARE:"cloudflare",CODESTRAL:"codestral",Cohere:"cohere",COHERE_CHAT:"cohere_chat",COMETAPI:"cometapi",COMPACTIFAI:"compactifai",Cursor:"cursor",Dashscope:"dashscope",Databricks:"databricks",DATAROBOT:"datarobot",DeepInfra:"deepinfra",Deepgram:"deepgram",Deepseek:"deepseek",DOCKER_MODEL_RUNNER:"docker_model_runner",DOTPROMPT:"dotprompt",ElevenLabs:"elevenlabs",EMPOWER:"empower",FalAI:"fal_ai",FEATHERLESS_AI:"featherless_ai",FireworksAI:"fireworks_ai",FRIENDLIAI:"friendliai",GALADRIEL:"galadriel",GITHUB_COPILOT:"github_copilot",Google_AI_Studio:"gemini",GradientAI:"gradient_ai",Groq:"groq",HEROKU:"heroku",Hosted_Vllm:"hosted_vllm",HUGGINGFACE:"huggingface",HYPERBOLIC:"hyperbolic",Infinity:"infinity",JinaAI:"jina_ai",LAMBDA_AI:"lambda_ai",LEMONADE:"lemonade",LLAMAFILE:"llamafile",LLAMA:"meta_llama",LM_STUDIO:"lm_studio",MARITALK:"maritalk",MiniMax:"minimax",MistralAI:"mistral",MOONSHOT:"moonshot",MORPH:"morph",NEBIUS:"nebius",NLP_CLOUD:"nlp_cloud",NOVITA:"novita",NSCALE:"nscale",NVIDIA_NIM:"nvidia_nim",Ollama:"ollama",OLLAMA_CHAT:"ollama_chat",OOBABOOGA:"oobabooga",OpenAI:"openai",OPENAI_LIKE:"openai_like",OpenAI_Compatible:"openai",OpenAI_Text:"text-completion-openai",OpenAI_Text_Compatible:"text-completion-openai",Openrouter:"openrouter",Oracle:"oci",OVHCLOUD:"ovhcloud",Perplexity:"perplexity",PETALS:"petals",PG_VECTOR:"pg_vector",PREDIBASE:"predibase",RECRAFT:"recraft",REPLICATE:"replicate",RunwayML:"runwayml",SAGEMAKER_LEGACY:"sagemaker",SageMaker:"sagemaker_chat",Sambanova:"sambanova",SAP:"sap",Snowflake:"snowflake",TEXT_COMPLETION_CODESTRAL:"text-completion-codestral",TogetherAI:"together_ai",TOPAZ:"topaz",Triton:"triton",V0:"v0",VERCEL_AI_GATEWAY:"vercel_ai_gateway",Vertex_AI:"vertex_ai",VERTEX_AI_BETA:"vertex_ai_beta",VLLM:"vllm",VolcEngine:"volcengine",Voyage:"voyage",WANDB:"wandb",WATSONX:"watsonx",WATSONX_TEXT:"watsonx_text",xAI:"xai",XINFERENCE:"xinference",ZAI:"zai"},o="../ui/assets/logos/",r={"A2A Agent":`${o}a2a_agent.png`,Ai21:`${o}ai21.svg`,"Ai21 Chat":`${o}ai21.svg`,"AI/ML API":`${o}aiml_api.svg`,"Aiohttp Openai":`${o}openai_small.svg`,Anthropic:`${o}anthropic.svg`,"Anthropic Text":`${o}anthropic.svg`,AssemblyAI:`${o}assemblyai_small.png`,Azure:`${o}microsoft_azure.svg`,"Azure AI Foundry (Studio)":`${o}microsoft_azure.svg`,"Azure Text":`${o}microsoft_azure.svg`,Baseten:`${o}baseten.svg`,"Amazon Bedrock":`${o}bedrock.svg`,"Amazon Bedrock Mantle":`${o}bedrock.svg`,"AWS SageMaker":`${o}bedrock.svg`,Cerebras:`${o}cerebras.svg`,Cloudflare:`${o}cloudflare.svg`,Codestral:`${o}mistral.svg`,Cohere:`${o}cohere.svg`,"Cohere Chat":`${o}cohere.svg`,Cometapi:`${o}cometapi.svg`,Cursor:`${o}cursor.svg`,"Databricks (Qwen API)":`${o}databricks.svg`,Dashscope:`${o}dashscope.svg`,Deepseek:`${o}deepseek.svg`,Deepgram:`${o}deepgram.png`,DeepInfra:`${o}deepinfra.png`,ElevenLabs:`${o}elevenlabs.png`,"Fal AI":`${o}fal_ai.jpg`,"Featherless Ai":`${o}featherless.svg`,"Fireworks AI":`${o}fireworks.svg`,Friendliai:`${o}friendli.svg`,"Github Copilot":`${o}github_copilot.svg`,"Google AI Studio":`${o}google.svg`,GradientAI:`${o}gradientai.svg`,Groq:`${o}groq.svg`,vllm:`${o}vllm.png`,Huggingface:`${o}huggingface.svg`,Hyperbolic:`${o}hyperbolic.svg`,Infinity:`${o}infinity.png`,"Jina AI":`${o}jina.png`,"Lambda Ai":`${o}lambda.svg`,"Lm Studio":`${o}lmstudio.svg`,"Meta Llama":`${o}meta_llama.svg`,MiniMax:`${o}minimax.svg`,"Mistral AI":`${o}mistral.svg`,Moonshot:`${o}moonshot.svg`,Morph:`${o}morph.svg`,Nebius:`${o}nebius.svg`,Novita:`${o}novita.svg`,"Nvidia Nim":`${o}nvidia_nim.svg`,Ollama:`${o}ollama.svg`,"Ollama Chat":`${o}ollama.svg`,Oobabooga:`${o}openai_small.svg`,OpenAI:`${o}openai_small.svg`,"Openai Like":`${o}openai_small.svg`,"OpenAI Text Completion":`${o}openai_small.svg`,"OpenAI-Compatible Text Completion Models (Together AI, etc.)":`${o}openai_small.svg`,"OpenAI-Compatible Endpoints (Together AI, etc.)":`${o}openai_small.svg`,Openrouter:`${o}openrouter.svg`,"Oracle Cloud Infrastructure (OCI)":`${o}oracle.svg`,Perplexity:`${o}perplexity-ai.svg`,Recraft:`${o}recraft.svg`,Replicate:`${o}replicate.svg`,RunwayML:`${o}runwayml.png`,Sagemaker:`${o}bedrock.svg`,Sambanova:`${o}sambanova.svg`,"SAP Generative AI Hub":`${o}sap.png`,Snowflake:`${o}snowflake.svg`,"Text-Completion-Codestral":`${o}mistral.svg`,TogetherAI:`${o}togetherai.svg`,Topaz:`${o}topaz.svg`,Triton:`${o}nvidia_triton.png`,V0:`${o}v0.svg`,"Vercel Ai Gateway":`${o}vercel.svg`,"Vertex AI (Anthropic, Gemini, etc.)":`${o}google.svg`,"Vertex Ai Beta":`${o}google.svg`,Vllm:`${o}vllm.png`,VolcEngine:`${o}volcengine.png`,"Voyage AI":`${o}voyage.webp`,Watsonx:`${o}watsonx.svg`,"Watsonx Text":`${o}watsonx.svg`,xAI:`${o}xai.svg`,Xinference:`${o}xinference.svg`};e.s(["Providers",()=>i,"getPlaceholder",0,e=>{if("AI/ML API"===e)return"aiml/flux-pro/v1.1";if("Vertex AI (Anthropic, Gemini, etc.)"===e)return"gemini-pro";if("Anthropic"==e)return"claude-3-opus";if("Amazon Bedrock"==e)return"claude-3-opus";if("AWS SageMaker"==e)return"sagemaker/jumpstart-dft-meta-textgeneration-llama-2-7b";else if("Google AI Studio"==e)return"gemini-pro";else if("Azure AI Foundry (Studio)"==e)return"azure_ai/command-r-plus";else if("Azure"==e)return"my-deployment";else if("Oracle Cloud Infrastructure (OCI)"==e)return"oci/xai.grok-4";else if("Snowflake"==e)return"snowflake/mistral-7b";else if("Voyage AI"==e)return"voyage/";else if("Jina AI"==e)return"jina_ai/";else if("VolcEngine"==e)return"volcengine/";else if("DeepInfra"==e)return"deepinfra/";else if("Fal AI"==e)return"fal_ai/fal-ai/flux-pro/v1.1-ultra";else if("RunwayML"==e)return"runwayml/gen4_turbo";else if("Watsonx"===e)return"watsonx/ibm/granite-3-3-8b-instruct";else if("Cursor"===e)return"cursor/claude-4-sonnet";else if("Z.AI (Zhipu AI)"===e)return"zai/glm-4.5";else return"gpt-3.5-turbo"},"getProviderLogoAndName",0,e=>{if(!e)return{logo:"",displayName:"-"};if("gemini"===e.toLowerCase()){let e="Google AI Studio";return{logo:r[e],displayName:e}}let t=Object.keys(a).find(t=>a[t].toLowerCase()===e.toLowerCase());if(!t)return{logo:"",displayName:e};let o=i[t];return{logo:r[o],displayName:o}},"getProviderModels",0,(e,t)=>{console.log(`Provider key: ${e}`);let i=a[e];console.log(`Provider mapped to: ${i}`);let o=[];return e&&"object"==typeof t&&(Object.entries(t).forEach(([e,t])=>{if(null!==t&&"object"==typeof t&&"litellm_provider"in t){let a=t.litellm_provider;(a===i||"string"==typeof a&&a.includes(i))&&o.push(e)}}),"Cohere"==e&&(console.log("Adding cohere chat models"),Object.entries(t).forEach(([e,t])=>{null!==t&&"object"==typeof t&&"litellm_provider"in t&&"cohere_chat"===t.litellm_provider&&o.push(e)})),"AWS SageMaker"==e&&(console.log("Adding sagemaker chat models"),Object.entries(t).forEach(([e,t])=>{null!==t&&"object"==typeof t&&"litellm_provider"in t&&"sagemaker_chat"===t.litellm_provider&&o.push(e)}))),o},"providerLogoMap",0,r,"provider_map",0,a])},798496,e=>{"use strict";var t=e.i(843476),i=e.i(152990),a=e.i(682830),o=e.i(271645),r=e.i(269200),n=e.i(427612),s=e.i(64848),l=e.i(942232),p=e.i(496020),d=e.i(977572),c=e.i(94629),g=e.i(360820),m=e.i(871943);function u({data:e=[],columns:u,isLoading:f=!1,defaultSorting:_=[],pagination:h,onPaginationChange:A,enablePagination:x=!1,onRowClick:b}){let[y,I]=o.default.useState(_),[v]=o.default.useState("onChange"),[C,E]=o.default.useState({}),[O,T]=o.default.useState({}),S=(0,i.useReactTable)({data:e,columns:u,state:{sorting:y,columnSizing:C,columnVisibility:O,...x&&h?{pagination:h}:{}},columnResizeMode:v,onSortingChange:I,onColumnSizingChange:E,onColumnVisibilityChange:T,...x&&A?{onPaginationChange:A}:{},getCoreRowModel:(0,a.getCoreRowModel)(),getSortedRowModel:(0,a.getSortedRowModel)(),...x?{getPaginationRowModel:(0,a.getPaginationRowModel)()}:{},enableSorting:!0,enableColumnResizing:!0,defaultColumn:{minSize:40,maxSize:500}});return(0,t.jsx)("div",{className:"rounded-lg custom-border relative",children:(0,t.jsx)("div",{className:"overflow-x-auto",children:(0,t.jsx)("div",{className:"relative min-w-full",children:(0,t.jsxs)(r.Table,{className:"[&_td]:py-2 [&_th]:py-2",style:{width:S.getTotalSize(),minWidth:"100%",tableLayout:"fixed"},children:[(0,t.jsx)(n.TableHead,{children:S.getHeaderGroups().map(e=>(0,t.jsx)(p.TableRow,{children:e.headers.map(e=>(0,t.jsxs)(s.TableHeaderCell,{className:`py-1 h-8 relative ${"actions"===e.id?"sticky right-0 bg-white shadow-[-4px_0_8px_-6px_rgba(0,0,0,0.1)] w-[120px] ml-8":""} ${e.column.columnDef.meta?.className||""}`,style:{width:"actions"===e.id?120:e.getSize(),position:"actions"===e.id?"sticky":"relative",right:"actions"===e.id?0:"auto"},onClick:e.column.getCanSort()?e.column.getToggleSortingHandler():void 0,children:[(0,t.jsxs)("div",{className:"flex items-center justify-between gap-2",children:[(0,t.jsx)("div",{className:"flex items-center",children:e.isPlaceholder?null:(0,i.flexRender)(e.column.columnDef.header,e.getContext())}),"actions"!==e.id&&e.column.getCanSort()&&(0,t.jsx)("div",{className:"w-4",children:e.column.getIsSorted()?({asc:(0,t.jsx)(g.ChevronUpIcon,{className:"h-4 w-4 text-blue-500"}),desc:(0,t.jsx)(m.ChevronDownIcon,{className:"h-4 w-4 text-blue-500"})})[e.column.getIsSorted()]:(0,t.jsx)(c.SwitchVerticalIcon,{className:"h-4 w-4 text-gray-400"})})]}),e.column.getCanResize()&&(0,t.jsx)("div",{onMouseDown:e.getResizeHandler(),onTouchStart:e.getResizeHandler(),className:`absolute right-0 top-0 h-full w-2 cursor-col-resize select-none touch-none ${e.column.getIsResizing()?"bg-blue-500":"hover:bg-blue-200"}`})]},e.id))},e.id))}),(0,t.jsx)(l.TableBody,{children:f?(0,t.jsx)(p.TableRow,{children:(0,t.jsx)(d.TableCell,{colSpan:u.length,className:"h-8 text-center",children:(0,t.jsx)("div",{className:"text-center text-gray-500",children:(0,t.jsx)("p",{children:"🚅 Loading models..."})})})}):S.getRowModel().rows.length>0?S.getRowModel().rows.map(e=>(0,t.jsx)(p.TableRow,{onClick:()=>b?.(e.original),className:b?"cursor-pointer hover:bg-gray-50":"",children:e.getVisibleCells().map(e=>(0,t.jsx)(d.TableCell,{className:`py-0.5 overflow-hidden ${"actions"===e.column.id?"sticky right-0 bg-white shadow-[-4px_0_8px_-6px_rgba(0,0,0,0.1)] w-[120px] ml-8":""} ${e.column.columnDef.meta?.className||""}`,style:{width:"actions"===e.column.id?120:e.column.getSize(),position:"actions"===e.column.id?"sticky":"relative",right:"actions"===e.column.id?0:"auto"},children:(0,i.flexRender)(e.column.columnDef.cell,e.getContext())},e.id))},e.id)):(0,t.jsx)(p.TableRow,{children:(0,t.jsx)(d.TableCell,{colSpan:u.length,className:"h-8 text-center",children:(0,t.jsx)("div",{className:"text-center text-gray-500",children:(0,t.jsx)("p",{children:"No models found"})})})})})]})})})})}e.s(["ModelDataTable",()=>u])}]); \ No newline at end of file diff --git a/litellm/proxy/_experimental/out/_next/static/chunks/06b2ea8c776c2e9b.js b/litellm/proxy/_experimental/out/_next/static/chunks/06b2ea8c776c2e9b.js deleted file mode 100644 index fe571604a7f..00000000000 --- a/litellm/proxy/_experimental/out/_next/static/chunks/06b2ea8c776c2e9b.js +++ /dev/null @@ -1 +0,0 @@ -(globalThis.TURBOPACK||(globalThis.TURBOPACK=[])).push(["object"==typeof document?document.currentScript:void 0,233538,e=>{"use strict";function t(e){let t=e.parentElement,r=null;for(;t&&!(t instanceof HTMLFieldSetElement);)t instanceof HTMLLegendElement&&(r=t),t=t.parentElement;let l=(null==t?void 0:t.getAttribute("disabled"))==="";return!(l&&function(e){if(!e)return!1;let t=e.previousElementSibling;for(;null!==t;){if(t instanceof HTMLLegendElement)return!1;t=t.previousElementSibling}return!0}(r))&&l}e.s(["isDisabledReactIssue7711",()=>t])},220508,e=>{"use strict";var t=e.i(271645);let r=t.forwardRef(function(e,r){return t.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",fill:"none",viewBox:"0 0 24 24",strokeWidth:2,stroke:"currentColor","aria-hidden":"true",ref:r},e),t.createElement("path",{strokeLinecap:"round",strokeLinejoin:"round",d:"M9 12l2 2 4-4m6 2a9 9 0 11-18 0 9 9 0 0118 0z"}))});e.s(["CheckCircleIcon",0,r],220508)},503269,214520,601893,694421,140721,942803,35889,722678,e=>{"use strict";var t=e.i(271645),r=e.i(914189);function l(e,l,s){let[a,n]=(0,t.useState)(s),i=void 0!==e,o=(0,t.useRef)(i),c=(0,t.useRef)(!1),d=(0,t.useRef)(!1);return!i||o.current||c.current?i||!o.current||d.current||(d.current=!0,o.current=i,console.error("A component is changing from controlled to uncontrolled. This may be caused by the value changing from a defined value to undefined, which should not happen.")):(c.current=!0,o.current=i,console.error("A component is changing from uncontrolled to controlled. This may be caused by the value changing from undefined to a defined value, which should not happen.")),[i?e:a,(0,r.useEvent)(e=>(i||n(e),null==l?void 0:l(e)))]}function s(e){let[r]=(0,t.useState)(e);return r}e.s(["useControllable",()=>l],503269),e.s(["useDefaultValue",()=>s],214520);let a=(0,t.createContext)(void 0);function n(){return(0,t.useContext)(a)}e.s(["useDisabled",()=>n],601893);var i=e.i(174080),o=e.i(746725);function c(e={},t=null,r=[]){for(let[l,s]of Object.entries(e))!function e(t,r,l){if(Array.isArray(l))for(let[s,a]of l.entries())e(t,d(r,s.toString()),a);else l instanceof Date?t.push([r,l.toISOString()]):"boolean"==typeof l?t.push([r,l?"1":"0"]):"string"==typeof l?t.push([r,l]):"number"==typeof l?t.push([r,`${l}`]):null==l?t.push([r,""]):c(l,r,t)}(r,d(t,l),s);return r}function d(e,t){return e?e+"["+t+"]":t}function u(e){var t,r;let l=null!=(t=null==e?void 0:e.form)?t:e.closest("form");if(l){for(let t of l.elements)if(t!==e&&("INPUT"===t.tagName&&"submit"===t.type||"BUTTON"===t.tagName&&"submit"===t.type||"INPUT"===t.nodeName&&"image"===t.type))return void t.click();null==(r=l.requestSubmit)||r.call(l)}}e.s(["attemptSubmit",()=>u,"objectToFormEntries",()=>c],694421);var m=e.i(700020),f=e.i(2788);let h=(0,t.createContext)(null);function g({children:e}){let r=(0,t.useContext)(h);if(!r)return t.default.createElement(t.default.Fragment,null,e);let{target:l}=r;return l?(0,i.createPortal)(t.default.createElement(t.default.Fragment,null,e),l):null}function p({data:e,form:r,disabled:l,onReset:s,overrides:a}){let[n,i]=(0,t.useState)(null),d=(0,o.useDisposables)();return(0,t.useEffect)(()=>{if(s&&n)return d.addEventListener(n,"reset",s)},[n,r,s]),t.default.createElement(g,null,t.default.createElement(x,{setForm:i,formId:r}),c(e).map(([e,s])=>t.default.createElement(f.Hidden,{features:f.HiddenFeatures.Hidden,...(0,m.compact)({key:e,as:"input",type:"hidden",hidden:!0,readOnly:!0,form:r,disabled:l,name:e,value:s,...a})})))}function x({setForm:e,formId:r}){return(0,t.useEffect)(()=>{if(r){let t=document.getElementById(r);t&&e(t)}},[e,r]),r?null:t.default.createElement(f.Hidden,{features:f.HiddenFeatures.Hidden,as:"input",type:"hidden",hidden:!0,readOnly:!0,ref:t=>{if(!t)return;let r=t.closest("form");r&&e(r)}})}e.s(["FormFields",()=>p],140721);let b=(0,t.createContext)(void 0);function y(){return(0,t.useContext)(b)}e.s(["useProvidedId",()=>y],942803);var v=e.i(835696),j=e.i(294316);let k=(0,t.createContext)(null);function N(){var e,r;return null!=(r=null==(e=(0,t.useContext)(k))?void 0:e.value)?r:void 0}function w(){let[e,l]=(0,t.useState)([]);return[e.length>0?e.join(" "):void 0,(0,t.useMemo)(()=>function(e){let s=(0,r.useEvent)(e=>(l(t=>[...t,e]),()=>l(t=>{let r=t.slice(),l=r.indexOf(e);return -1!==l&&r.splice(l,1),r}))),a=(0,t.useMemo)(()=>({register:s,slot:e.slot,name:e.name,props:e.props,value:e.value}),[s,e.slot,e.name,e.props,e.value]);return t.default.createElement(k.Provider,{value:a},e.children)},[l])]}k.displayName="DescriptionContext";let S=Object.assign((0,m.forwardRefWithAs)(function(e,r){let l=(0,t.useId)(),s=n(),{id:a=`headlessui-description-${l}`,...i}=e,o=function e(){let r=(0,t.useContext)(k);if(null===r){let t=Error("You used a component, but it is not inside a relevant parent.");throw Error.captureStackTrace&&Error.captureStackTrace(t,e),t}return r}(),c=(0,j.useSyncRefs)(r);(0,v.useIsoMorphicEffect)(()=>o.register(a),[a,o.register]);let d=s||!1,u=(0,t.useMemo)(()=>({...o.slot,disabled:d}),[o.slot,d]),f={ref:c,...o.props,id:a};return(0,m.useRender)()({ourProps:f,theirProps:i,slot:u,defaultTag:"p",name:o.name||"Description"})}),{});e.s(["Description",()=>S,"useDescribedBy",()=>N,"useDescriptions",()=>w],35889);let C=(0,t.createContext)(null);function M(e){var r,l,s;let a=null!=(l=null==(r=(0,t.useContext)(C))?void 0:r.value)?l:void 0;return(null!=(s=null==e?void 0:e.length)?s:0)>0?[a,...e].filter(Boolean).join(" "):a}function O({inherit:e=!1}={}){let l=M(),[s,a]=(0,t.useState)([]),n=e?[l,...s].filter(Boolean):s;return[n.length>0?n.join(" "):void 0,(0,t.useMemo)(()=>function(e){let l=(0,r.useEvent)(e=>(a(t=>[...t,e]),()=>a(t=>{let r=t.slice(),l=r.indexOf(e);return -1!==l&&r.splice(l,1),r}))),s=(0,t.useMemo)(()=>({register:l,slot:e.slot,name:e.name,props:e.props,value:e.value}),[l,e.slot,e.name,e.props,e.value]);return t.default.createElement(C.Provider,{value:s},e.children)},[a])]}C.displayName="LabelContext";let E=Object.assign((0,m.forwardRefWithAs)(function(e,l){var s;let a=(0,t.useId)(),i=function e(){let r=(0,t.useContext)(C);if(null===r){let t=Error("You used a