Merge branch 'main' into fix/dated-provider-slug-zero-cost

Mark the blind except in _has_rate with noqa: BLE001, as the file does elsewhere for get_model_info.

.cargo/config.toml

@@ -17,3 +17,6 @@ rustflags = ["-C", "link-arg=-undefined", "-C", "link-arg=dynamic_lookup"]
 
 [target.aarch64-apple-darwin]
 rustflags = ["-C", "link-arg=-undefined", "-C", "link-arg=dynamic_lookup"]
+
+[env]
+SQLX_OFFLINE = "true"

.circleci/config.yml

@@ -486,6 +486,7 @@ jobs:
       - install_rust
       - run:
           name: Build the wheel
+          no_output_timeout: 30m
           environment:
             UV_HTTP_TIMEOUT: "300"
           command: |
@@ -3419,7 +3420,7 @@ workflows:
           name: integration-<< matrix.suite >>
           matrix:
             parameters:
-              suite: [management, accounting, database, providers, mcp, sdk, cost, browser]
+              suite: [management, accounting, database, providers, mcp, sdk, cost, security, browser]
       - integration_contracts:
           name: integration-extensions
           suite: extensions

.circleci/scripts/run_integration.sh

@@ -168,6 +168,7 @@ start_proxy() {
     "${database_env[@]}" REDIS_HOST="$REDIS_HOST" REDIS_PORT="$REDIS_PORT" \
     INTEGRATION_UPSTREAM_URL="$INTEGRATION_UPSTREAM_URL" \
     LITELLM_MASTER_KEY="$LITELLM_MASTER_KEY" LITELLM_SALT_KEY="$LITELLM_SALT_KEY" LITELLM_UI_PATH="$LITELLM_UI_PATH" PROXY_BASE_URL="http://127.0.0.1:$port" \
+    LITELLM_LICENSE="${LITELLM_LICENSE:-}" \
     LITELLM_MODE=PRODUCTION STORE_MODEL_IN_DB=True "${cost_map_env[@]}" \
     AWS_EC2_METADATA_DISABLED=true DO_NOT_TRACK=1 COVERAGE_FILE="$coverage_data" \
     "${proxy_command[@]}" --config tests/integration/proxy_config.yaml \
@@ -228,6 +229,7 @@ env -i PATH="$PATH" HOME="$HOME" PYTHONPATH="$PYTHONPATH" \
   INTEGRATION_UPSTREAM_URL="$INTEGRATION_UPSTREAM_URL" \
   INTEGRATION_WORKERS="${INTEGRATION_WORKERS:-1}" \
   INTEGRATION_MASTER_KEY="$INTEGRATION_MASTER_KEY" LITELLM_MODE=PRODUCTION \
+  LITELLM_LICENSE="${LITELLM_LICENSE:-}" \
   INTEGRATION_SEED="$INTEGRATION_SEED" \
   INTEGRATION_ORDER_SEED="$INTEGRATION_ORDER_SEED" \
   LITELLM_LOCAL_MODEL_COST_MAP=True AWS_EC2_METADATA_DISABLED=true DO_NOT_TRACK=1 \

.circleci/scripts/unit_selection.sh

@@ -89,6 +89,7 @@ legacy_paths() {
     proxy-db-auth-checks)
       echo tests/unit/proxy/auth/test_auth_checks.py
       echo tests/unit/proxy/auth/test_user_api_key_auth.py
+      echo tests/unit/proxy/test_credential_slot_registry.py
       echo tests/unit/proxy/test_deprecated_key_grace_period.py ;;
     proxy-db-budgets)
       echo tests/unit/proxy/auth/test_default_end_user_budget_simple.py
@@ -106,6 +107,7 @@ legacy_paths() {
       echo tests/unit/proxy/test_update_spend.py
       echo tests/unit/skills/test_skills_db.py ;;
     proxy-db-endpoints-and-responses)
+      echo tests/unit/proxy/engine
       echo tests/unit/proxy/auth/test_models_fallback_endpoint.py
       echo tests/unit/proxy/common_utils/test_check_batch_cost.py
       echo tests/unit/proxy/common_utils/test_check_responses_cost.py
@@ -147,7 +149,10 @@ legacy_paths() {
       echo tests/unit/proxy/test_proxy_server.py ;;
     proxy-db-proxy-utils) echo tests/unit/proxy/test_proxy_utils.py ;;
     proxy-extras) echo tests/unit/litellm_proxy_extras ;;
-    proxy-infra) echo tests/unit/gateway ;;
+    proxy-infra)
+      echo tests/unit/gateway
+      echo tests/unit/proxy/management_endpoints/test_roi_calculator_endpoints.py
+      echo tests/unit/proxy/roi_calculator ;;
     responses-caching-types)
       find tests/unit/responses -name 'test_*.py' -not -path 'tests/unit/responses/mcp/*'
       echo tests/unit/types ;;

.githooks/commit-msg

@@ -42,7 +42,7 @@ case "$subject" in
         ;;
 esac
 
-ALLOWED_TYPES="feat|fix|docs|style|refactor|perf|test|build|ci|chore|revert"
+ALLOWED_TYPES="feat|fix|docs|style|refactor|perf|test|build|ci|chore|revert|security"
 # Description must not start with an uppercase letter — kept in sync with the
 # subjectPattern in .github/workflows/conventional-commits.yml so the local
 # hook is the strictly tighter of the two gates. (Without this guard, a commit
@@ -61,7 +61,7 @@ cat >&2 <<EOF
   Expected: <type>(<scope>)!: <description>
             (description must start with a lowercase letter)
 
-  Allowed types: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert
+  Allowed types: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert, security
   Examples:
     feat(router): add weighted round-robin strategy
     fix(bedrock): decouple STS region from aws_region_name

.github/CODEOWNERS

@@ -1,10 +1,2 @@
-/ui/ @yuneng-berri @ryan-crabbe-berri
-/litellm/proxy/_experimental/out/ @yuneng-berri @ryan-crabbe-berri
-/ui/Dockerfile
-/ui/nginx.conf
-/ui/litellm-dashboard/src/lib/http/schema.d.ts
-/ui/litellm-dashboard/tsconfig.tsbuildinfo
 /model_prices_and_context_window.json @mateo-berri @ryan-crabbe-berri @kerry-berri
 /litellm/model_prices_and_context_window_backup.json @mateo-berri @ryan-crabbe-berri @kerry-berri
-/litellm-proxy-extras/litellm_proxy_extras/migrations/ @yuneng-berri @ryan-crabbe-berri
-/.github/CODEOWNERS @yuneng-berri

.github/actions/cache-cargo-build/action.yml

@@ -25,6 +25,7 @@ runs:
   using: composite
   steps:
     - name: Restore the Cargo registry and target directory
+      if: github.ref == 'refs/heads/main'
       uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
       with:
         path: |
@@ -34,3 +35,15 @@ runs:
         key: ${{ runner.os }}-maturin-${{ inputs.profile }}-${{ hashFiles('litellm-rust/Cargo.lock') }}
         restore-keys: |
           ${{ runner.os }}-maturin-${{ inputs.profile }}-
+
+    - name: Restore the Cargo registry and target directory
+      if: github.ref != 'refs/heads/main'
+      uses: actions/cache/restore@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+      with:
+        path: |
+          ~/.cargo/registry
+          ~/.cargo/git
+          litellm-rust/target
+        key: ${{ runner.os }}-maturin-${{ inputs.profile }}-${{ hashFiles('litellm-rust/Cargo.lock') }}
+        restore-keys: |
+          ${{ runner.os }}-maturin-${{ inputs.profile }}-

.github/actions/cache-prisma-binaries/action.yml

@@ -30,6 +30,7 @@ runs:
         echo "version=${version}" >> "$GITHUB_OUTPUT"
 
     - name: Restore Prisma binaries
+      if: github.ref == 'refs/heads/main'
       uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
       with:
         # ~/.cache/prisma-python holds the npm install tree prisma-client-py
@@ -38,3 +39,12 @@ runs:
           ~/.cache/prisma-python
           ~/.cache/prisma
         key: ${{ runner.os }}-prisma-binaries-${{ steps.version.outputs.version }}
+
+    - name: Restore Prisma binaries
+      if: github.ref != 'refs/heads/main'
+      uses: actions/cache/restore@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+      with:
+        path: |
+          ~/.cache/prisma-python
+          ~/.cache/prisma
+        key: ${{ runner.os }}-prisma-binaries-${{ steps.version.outputs.version }}

.github/actions/cache-uv-downloads/action.yml

@@ -0,0 +1,25 @@
+name: "Cache uv downloads"
+description: >-
+  Restore the uv download cache on every run and save it only from main, so pull
+  requests reuse main's cache instead of evicting it with their own copies.
+
+runs:
+  using: composite
+  steps:
+    - name: Restore and save the uv download cache
+      if: github.ref == 'refs/heads/main'
+      uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+      with:
+        path: ${{ env.UV_CACHE_DIR }}
+        key: ${{ runner.os }}-uv-downloads-py${{ env.UV_PYTHON }}-${{ hashFiles('uv.lock') }}
+        restore-keys: |
+          ${{ runner.os }}-uv-downloads-py${{ env.UV_PYTHON }}-
+
+    - name: Restore the uv download cache
+      if: github.ref != 'refs/heads/main'
+      uses: actions/cache/restore@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+      with:
+        path: ${{ env.UV_CACHE_DIR }}
+        key: ${{ runner.os }}-uv-downloads-py${{ env.UV_PYTHON }}-${{ hashFiles('uv.lock') }}
+        restore-keys: |
+          ${{ runner.os }}-uv-downloads-py${{ env.UV_PYTHON }}-

.github/actions/setup-uv-with-retries/action.yml

@@ -17,6 +17,7 @@ runs:
       uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
       with:
         version: ${{ inputs.version }}
+        save-cache: ${{ github.ref == 'refs/heads/main' }}
 
     - name: Wait before attempt 2
       if: steps.attempt-1.outcome == 'failure'
@@ -30,6 +31,7 @@ runs:
       uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
       with:
         version: ${{ inputs.version }}
+        save-cache: ${{ github.ref == 'refs/heads/main' }}
 
     - name: Wait before attempt 3
       if: steps.attempt-2.outcome == 'failure'
@@ -41,3 +43,4 @@ runs:
       uses: astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d # v10.0.1
       with:
         version: ${{ inputs.version }}
+        save-cache: ${{ github.ref == 'refs/heads/main' }}

.github/ci-coverage-allowlist.yml

@@ -111,3 +111,10 @@ dockerfiles:
       An example image under cookbook/ that is documentation rather than a shipped artifact
     paths:
       - cookbook/litellm-ollama-docker-image/Dockerfile
+  - reason: >-
+      The Rust gateway image compiles the whole workspace in release mode, which is too slow for
+      a per-pull-request job while the gateway binary is still being assembled; the Rust lint,
+      clippy, and compile jobs already cover the code it packages. Revisit when the gateway is
+      published
+    paths:
+      - litellm-rust/crates/gateway/Dockerfile

.github/merge-smoke-tests.json

@@ -3,8 +3,8 @@
     "CHAT-JSON": "tests/unit/llms/openai/test_openai.py::test_acompletion_returns_json_reply_over_injected_transport",
     "CHAT-TEXT-STREAM": "tests/unit/llms/openai/test_openai.py::test_acompletion_streams_text_deltas_over_injected_transport",
     "CHAT-TOOL-STREAM": "tests/unit/llms/openai/test_openai.py::test_acompletion_streams_tool_call_arguments_over_injected_transport",
-    "MODEL-ALLOW": "tests/test_litellm/proxy/auth/test_auth_checks.py::test_can_object_call_model_allows_listed_model_for_key",
-    "MODEL-DENY": "tests/test_litellm/proxy/auth/test_auth_checks.py::test_can_object_call_model_denials_return_forbidden[key-key_model_access_denied]",
+    "MODEL-ALLOW": "tests/unit/proxy/auth/test_auth_checks_object_access_and_lookup.py::test_can_object_call_model_allows_listed_model_for_key",
+    "MODEL-DENY": "tests/unit/proxy/auth/test_auth_checks_object_access_and_lookup.py::test_can_object_call_model_denials_return_forbidden[key-key_model_access_denied]",
     "COST-EXPLICIT": "tests/unit/test_cost_calculator.py::test_completion_cost_charges_explicit_per_token_rates_over_registered_ones",
     "COST-ZERO": "tests/unit/test_cost_calculator.py::test_completion_cost_is_zero_when_explicit_rates_are_zero",
     "LOG-CONTENT-ON": "tests/unit/litellm_core_utils/test_litellm_logging.py::test_standard_logging_payload_keeps_message_content_when_message_logging_is_on",

.github/scripts/verify_linux_native_wheel.py

@@ -214,7 +214,7 @@ def main(
     native_module: Final = load_native_module(native_path)
     native_module_loads: Final = native_module is not None
     panic_test_hook_absent: Final = native_module is not None and not hasattr(native_module, "_panic_for_test")
-    native_size_limit: Final = 40_000_000
+    native_size_limit: Final = 45_000_000
     native_size_within_limit: Final = native_member.file_size <= native_size_limit
     validations: Final = (
         (f"Python tag is {EXPECTED_PYTHON_TAG}", python_tag == EXPECTED_PYTHON_TAG),

.github/workflows/_test-unit-base.yml

@@ -132,12 +132,7 @@ jobs:
       - name: Cache uv dependencies
         if: steps.changes.outputs.decision != 'skip'
         timeout-minutes: 5
-        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
-        with:
-          path: ${{ env.UV_CACHE_DIR }}
-          key: ${{ runner.os }}-uv-downloads-py${{ env.UV_PYTHON }}-${{ hashFiles('uv.lock') }}
-          restore-keys: |
-            ${{ runner.os }}-uv-downloads-py${{ env.UV_PYTHON }}-
+        uses: ./.github/actions/cache-uv-downloads
 
       - name: Cache the Rust build
         if: steps.changes.outputs.decision != 'skip'
@@ -274,7 +269,7 @@ jobs:
       - name: Upload to Codecov
         id: codecov-upload
         continue-on-error: true
-        uses: codecov/codecov-action@75cd11691c0faa626561e295848008c8a7dddffe # v5.5.4
+        uses: codecov/codecov-action@0fb7174895f61a3b6b78fc075e0cd60383518dac # v5.5.5
         with:
           use_oidc: true
           directory: coverage-reports
@@ -285,7 +280,7 @@ jobs:
       - name: Upload to Codecov (retry)
         if: steps.codecov-upload.outcome == 'failure'
         continue-on-error: true
-        uses: codecov/codecov-action@75cd11691c0faa626561e295848008c8a7dddffe # v5.5.4
+        uses: codecov/codecov-action@0fb7174895f61a3b6b78fc075e0cd60383518dac # v5.5.5
         with:
           use_oidc: true
           directory: coverage-reports

.github/workflows/conventional-commits.yml

@@ -41,6 +41,7 @@ jobs:
             ci
             chore
             revert
+            security
           requireScope: false
           subjectPattern: ^(?![A-Z]).+$
           subjectPatternError: |

.github/workflows/create-rc-branch.yml

@@ -15,6 +15,8 @@ jobs:
     runs-on: ubuntu-latest
     permissions:
       contents: write
+    outputs:
+      version: ${{ steps.version.outputs.version }}
     steps:
       - name: Require main
         env:
@@ -64,3 +66,14 @@ jobs:
               sha: context.sha,
             });
             core.info(`Created branch ${branchName} at ${context.sha}`);
+
+  linear-release:
+    name: Move the Linear release to rc
+    needs: create-rc-branch
+    permissions:
+      contents: read
+    uses: ./.github/workflows/linear-release.yml
+    with:
+      rc_version: ${{ needs.create-rc-branch.outputs.version }}
+    secrets:
+      LINEAR_API_KEY: ${{ secrets.LINEAR_API_KEY }}

.github/workflows/lens-worker.yml

@@ -0,0 +1,66 @@
+name: Lens Worker Image
+
+on:
+  pull_request:
+    branches: [main, litellm_oss_branch, "litellm_**"]
+    paths:
+      - deploy/lens/**
+      - litellm/proxy/engine/**
+      - .github/workflows/lens-worker.yml
+  push:
+    branches: [main, litellm_agent_engine]
+    paths:
+      - deploy/lens/**
+      - litellm/proxy/engine/**
+      - .github/workflows/lens-worker.yml
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  lens-worker-image:
+    permissions:
+      contents: read
+      packages: write
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
+        with:
+          persist-credentials: false
+      - name: Build Lens worker
+        run: docker build -f deploy/lens/Dockerfile -t lens-worker:${{ github.sha }} .
+      - name: Verify standalone imports with a read-only filesystem
+        run: |
+          docker run --rm --network none --read-only --cap-drop ALL --tmpfs /tmp:rw,noexec,nosuid,size=1g \
+            --security-opt no-new-privileges --entrypoint python \
+            lens-worker:${{ github.sha }} -c '
+          import os
+          import engine.worker
+          from engine.trace_store import trace_store
+          assert os.getuid() == 65532
+          with trace_store() as store:
+              assert store.count() == 0
+          '
+      - name: Verify recovery after temporary storage fills
+        run: |
+          docker run --rm --network none --read-only --cap-drop ALL \
+            --tmpfs /tmp:rw,noexec,nosuid,size=64k --security-opt no-new-privileges \
+            -v "$PWD/tests/proxy_behavior/lens/worker_storage_smoke.py:/app/storage_smoke.py:ro" \
+            --entrypoint python lens-worker:${{ github.sha }} /app/storage_smoke.py
+      - name: Publish versioned Lens worker
+        if: github.event_name != 'pull_request' && github.repository == 'BerriAI/litellm'
+        env:
+          REGISTRY_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          REGISTRY_USER: ${{ github.actor }}
+          IMAGE: ghcr.io/berriai/litellm-lens-worker:sha-${{ github.sha }}
+        run: |
+          printf '%s' "$REGISTRY_TOKEN" | docker login ghcr.io -u "$REGISTRY_USER" --password-stdin
+          docker tag lens-worker:${{ github.sha }} "$IMAGE"
+          docker push "$IMAGE"
+          printf 'Lens worker image: `%s`\n' "$IMAGE" >> "$GITHUB_STEP_SUMMARY"

.github/workflows/linear-release.yml

@@ -0,0 +1,131 @@
+name: Linear Release
+
+on:
+  push:
+    branches:
+      - main
+      - "rc/**"
+  release:
+    types: [published]
+  workflow_call:
+    inputs:
+      rc_version:
+        description: "X.Y.0 release whose rc branch was just cut"
+        required: true
+        type: string
+    secrets:
+      LINEAR_API_KEY:
+        required: true
+
+permissions: {}
+
+jobs:
+  linear-release:
+    name: Linear Release
+    if: github.repository == 'BerriAI/litellm'
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+    steps:
+      - uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+
+      - name: Plan
+        id: plan
+        env:
+          EVENT: ${{ github.event_name }}
+          REF_NAME: ${{ github.ref_name }}
+          BEFORE: ${{ github.event.before }}
+          CREATED: ${{ github.event.created }}
+          RC_VERSION: ${{ inputs.rc_version }}
+          RELEASE_TAG: ${{ github.event.release.tag_name }}
+          PRERELEASE: ${{ github.event.release.prerelease }}
+        run: |
+          set -euo pipefail
+          sync_base="${BEFORE}"
+          if [ "${CREATED}" = "true" ]; then
+            sync_base=""
+          fi
+          if [ -n "${RC_VERSION}" ]; then
+            echo "version=${RC_VERSION}" >> "$GITHUB_OUTPUT"
+            echo "stage=rc" >> "$GITHUB_OUTPUT"
+          elif [ "${EVENT}" = "release" ]; then
+            if [ "${PRERELEASE}" = "true" ] || ! echo "${RELEASE_TAG}" | grep -qE '^v[0-9]+\.[0-9]+\.0$'; then
+              echo "::notice::${RELEASE_TAG} is not an X.Y.0 stable release; nothing to complete"
+              exit 0
+            fi
+            echo "version=${RELEASE_TAG#v}" >> "$GITHUB_OUTPUT"
+            echo "complete=true" >> "$GITHUB_OUTPUT"
+          elif [ "${REF_NAME}" = "main" ]; then
+            version="$(python3 .github/scripts/read_rc_version.py | cut -d= -f2)"
+            status=0
+            git ls-remote --exit-code --heads origin "rc/${version}" > /dev/null || status=$?
+            case "${status}" in
+              0)
+                IFS=. read -r major minor _ <<< "${version}"
+                version="${major}.$((minor + 1)).0"
+                ;;
+              2) ;;
+              *)
+                echo "::error::could not check whether rc/${version} exists (git ls-remote exit ${status})"
+                exit 1
+                ;;
+            esac
+            echo "version=${version}" >> "$GITHUB_OUTPUT"
+            echo "sync_base=${sync_base}" >> "$GITHUB_OUTPUT"
+            echo "main=true" >> "$GITHUB_OUTPUT"
+          else
+            echo "version=${REF_NAME#rc/}" >> "$GITHUB_OUTPUT"
+            echo "sync_base=${sync_base}" >> "$GITHUB_OUTPUT"
+            echo "stage=rc" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Sync commits into the release
+        if: steps.plan.outputs.sync_base != ''
+        uses: linear/linear-release-action@d4af10092984f9bc6d5efa075b242bdf01333463 # v0.18.0
+        with:
+          access_key: ${{ secrets.LINEAR_API_KEY }}
+          command: sync
+          name: LiteLLM ${{ steps.plan.outputs.version }}
+          version: ${{ steps.plan.outputs.version }}
+          base_ref: ${{ steps.plan.outputs.sync_base }}
+          cli_version: v0.18.0
+
+      - name: Keep the main stage unless the rc branch was cut during this run
+        id: main_stage
+        if: steps.plan.outputs.main == 'true'
+        env:
+          VERSION: ${{ steps.plan.outputs.version }}
+        run: |
+          set -euo pipefail
+          status=0
+          git ls-remote --exit-code --heads origin "rc/${VERSION}" > /dev/null || status=$?
+          case "${status}" in
+            0) echo "::notice::rc/${VERSION} was cut during this run; leaving the release in its rc stage" ;;
+            2) echo "stage=main" >> "$GITHUB_OUTPUT" ;;
+            *)
+              echo "::error::could not check whether rc/${VERSION} exists (git ls-remote exit ${status})"
+              exit 1
+              ;;
+          esac
+
+      - name: Move the release to its stage
+        if: steps.plan.outputs.stage != '' || steps.main_stage.outputs.stage != ''
+        uses: linear/linear-release-action@d4af10092984f9bc6d5efa075b242bdf01333463 # v0.18.0
+        with:
+          access_key: ${{ secrets.LINEAR_API_KEY }}
+          command: update
+          stage: ${{ steps.plan.outputs.stage || steps.main_stage.outputs.stage }}
+          version: ${{ steps.plan.outputs.version }}
+          cli_version: v0.18.0
+
+      - name: Complete the release
+        if: steps.plan.outputs.complete == 'true'
+        uses: linear/linear-release-action@d4af10092984f9bc6d5efa075b242bdf01333463 # v0.18.0
+        with:
+          access_key: ${{ secrets.LINEAR_API_KEY }}
+          command: complete
+          version: ${{ steps.plan.outputs.version }}
+          cli_version: v0.18.0

.github/workflows/mutation-test.yml

@@ -44,6 +44,7 @@ jobs:
           version: "0.10.9"
 
       - name: Cache uv dependencies
+        if: github.ref == 'refs/heads/main'
         uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
         with:
           path: |
@@ -53,6 +54,17 @@ jobs:
           restore-keys: |
             ${{ runner.os }}-uv-
 
+      - name: Cache uv dependencies
+        if: github.ref != 'refs/heads/main'
+        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: Cache the Rust build
         uses: ./.github/actions/cache-cargo-build
 

.github/workflows/test-code-quality.yml

@@ -44,6 +44,7 @@ jobs:
           version: "0.10.9"
 
       - name: Cache uv dependencies
+        if: github.ref == 'refs/heads/main'
         uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
         with:
           path: |
@@ -53,6 +54,17 @@ jobs:
           restore-keys: |
             ${{ runner.os }}-uv-
 
+      - name: Cache uv dependencies
+        if: github.ref != 'refs/heads/main'
+        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: Cache the Rust build
         uses: ./.github/actions/cache-cargo-build
 
@@ -80,6 +92,11 @@ jobs:
       - name: test_e2e_changed_gate
         run: uv run --no-sync pytest -q --noconftest -p no:cacheprovider -c /dev/null tests/code_coverage_tests/test_e2e_changed_gate.py tests/code_coverage_tests/test_e2e_idp_stack.py
 
+      - name: test_e2e_metadata
+        env:
+          PYTHONPATH: tests/e2e
+        run: uv run --no-sync pytest -q --noconftest -p no:cacheprovider -c /dev/null tests/code_coverage_tests/test_e2e_metadata.py tests/code_coverage_tests/test_e2e_junit_report.py
+
       - name: Check merge smoke harness
         run: uv run --no-sync pytest -q --noconftest -p no:cacheprovider -c /dev/null tests/code_coverage_tests/test_merge_smoke.py
 

.github/workflows/test-postgres.yml

@@ -24,6 +24,7 @@ jobs:
     timeout-minutes: ${{ matrix.job-timeout-minutes }}
     permissions:
       contents: read
+      id-token: write
 
     services:
       postgres:
@@ -94,7 +95,7 @@ jobs:
           version: "0.10.9"
 
       - name: Cache uv dependencies
-        if: steps.changes.outputs.decision != 'skip'
+        if: steps.changes.outputs.decision != 'skip' && github.ref == 'refs/heads/main'
         timeout-minutes: 5
         uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
         with:
@@ -105,6 +106,18 @@ jobs:
           restore-keys: |
             ${{ runner.os }}-uv-postgres-
 
+      - name: Cache uv dependencies
+        if: steps.changes.outputs.decision != 'skip' && github.ref != 'refs/heads/main'
+        timeout-minutes: 5
+        uses: actions/cache/restore@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
+        with:
+          path: |
+            ~/.cache/uv
+            .venv
+          key: ${{ runner.os }}-uv-postgres-${{ hashFiles('uv.lock') }}
+          restore-keys: |
+            ${{ runner.os }}-uv-postgres-
+
       - name: Install dependencies
         if: steps.changes.outputs.decision != 'skip'
         timeout-minutes: 12
@@ -134,9 +147,21 @@ jobs:
         env:
           TEST_PATH: ${{ matrix.test-path }}
           WORKERS: ${{ matrix.workers }}
+          PYTEST_ADDOPTS: ${{ matrix.shard == 'proxy-behavior' && '--cov=./litellm --cov-report=xml:coverage-lens-postgres.xml' || '' }}
         run: |
           if [ "${WORKERS}" = "0" ]; then
             uv run --no-sync pytest ${TEST_PATH:?} -vv --tb=short --durations=10
           else
             uv run --no-sync pytest ${TEST_PATH:?} -vv --tb=short --durations=10 -n "${WORKERS}"
           fi
+
+      - name: Upload Lens database coverage
+        if: steps.changes.outputs.decision != 'skip' && matrix.shard == 'proxy-behavior' && !cancelled()
+        uses: codecov/codecov-action@303a32d7a59b442fa8d48b6a1cc6825c09c847a5 # v7.1.1
+        with:
+          use_oidc: true
+          version: v11.3.1
+          root_dir: ${{ github.workspace }}
+          files: coverage-lens-postgres.xml
+          flags: lens-postgres
+          fail_ci_if_error: true

.github/workflows/test-redis-compat.yml

@@ -98,7 +98,7 @@ jobs:
 
       - name: Upload Redis coverage
         if: matrix.redis-version == '5.3.1'
-        uses: codecov/codecov-action@75cd11691c0faa626561e295848008c8a7dddffe # v5.5.4
+        uses: codecov/codecov-action@0fb7174895f61a3b6b78fc075e0cd60383518dac # v5.5.5
         with:
           use_oidc: true
           files: coverage-redis.xml

.github/workflows/test-rust.yml

@@ -83,6 +83,7 @@ jobs:
         with:
           workspaces: litellm-rust
           cache-on-failure: true
+          save-if: ${{ github.ref == 'refs/heads/main' }}
 
       - run: cargo clippy --workspace --all-targets --locked -- -D warnings
 
@@ -121,6 +122,7 @@ jobs:
         with:
           workspaces: litellm-rust
           cache-on-failure: true
+          save-if: ${{ github.ref == 'refs/heads/main' }}
 
       - run: cargo nextest run --workspace --locked
 
@@ -162,6 +164,7 @@ jobs:
         with:
           workspaces: litellm-rust
           cache-on-failure: true
+          save-if: ${{ github.ref == 'refs/heads/main' }}
 
       - run: uv build --wheel --out-dir dist
 

.github/workflows/test-terraform-provider.yml

@@ -77,6 +77,7 @@ jobs:
           version: "0.10.9"
 
       - name: Cache uv dependencies
+        if: github.ref == 'refs/heads/main'
         uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
         with:
           path: |
@@ -86,6 +87,17 @@ jobs:
           restore-keys: |
             ${{ runner.os }}-uv-
 
+      - name: Cache uv dependencies
+        if: github.ref != 'refs/heads/main'
+        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: Cache the Rust build
         uses: ./.github/actions/cache-cargo-build
 

.github/workflows/test-unit-documentation.yml

@@ -54,7 +54,7 @@ jobs:
           version: "0.10.9"
 
       - name: Cache uv dependencies
-        if: steps.changes.outputs.decision != 'skip'
+        if: steps.changes.outputs.decision != 'skip' && github.ref == 'refs/heads/main'
         uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
         with:
           path: |
@@ -64,6 +64,17 @@ jobs:
           restore-keys: |
             ${{ runner.os }}-uv-
 
+      - name: Cache uv dependencies
+        if: steps.changes.outputs.decision != 'skip' && github.ref != 'refs/heads/main'
+        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: Cache the Rust build
         if: steps.changes.outputs.decision != 'skip'
         uses: ./.github/actions/cache-cargo-build

.github/workflows/test-unit.yml

@@ -79,7 +79,9 @@ jobs:
 
           - shard: integrations
             artifact-name: integrations
-            test-path: ""
+            test-path: >-
+              tests/test_litellm/integrations
+              tests/test_litellm/tracing
             unit-flag: integrations
             workers: 2
             reruns: 3
@@ -117,10 +119,19 @@ jobs:
           - shard: proxy-auth
             artifact-name: proxy-auth
             test-path: >-
-              tests/test_litellm/proxy/auth
-              tests/test_litellm/proxy/hooks
-              tests/test_litellm/proxy/policy_engine
-              tests/test_litellm/proxy/client
+              tests/unit/proxy/auth
+              tests/unit/proxy/hooks
+              tests/unit/proxy/policy_engine
+              tests/unit/proxy/client
+              --ignore=tests/unit/proxy/auth/test_auth_checks.py
+              --ignore=tests/unit/proxy/auth/test_user_api_key_auth.py
+              --ignore=tests/unit/proxy/auth/test_default_end_user_budget_simple.py
+              --ignore=tests/unit/proxy/auth/test_jwt.py
+              --ignore=tests/unit/proxy/auth/test_models_fallback_endpoint.py
+              --ignore=tests/unit/proxy/auth/test_multipart_bypass_repro.py
+              --ignore=tests/unit/proxy/auth/test_proxy_routes.py
+              --ignore=tests/unit/proxy/hooks/test_banned_keyword_list.py
+              --ignore=tests/unit/proxy/hooks/test_unit_test_max_model_budget_limiter.py
             workers: 2
             reruns: 2
             timeout-minutes: 20
@@ -130,11 +141,15 @@ jobs:
             artifact-name: proxy-endpoints
             test-path: >-
               tests/test_litellm/proxy/analytics_endpoints
-              tests/test_litellm/proxy/management_endpoints
+              tests/unit/proxy/management_endpoints
               tests/test_litellm/proxy/list_api
               tests/test_litellm/proxy/memory
-              tests/test_litellm/proxy/guardrails
-              tests/test_litellm/proxy/management_helpers
+              tests/unit/proxy/guardrails
+              tests/unit/proxy/management_helpers
+              --ignore=tests/unit/proxy/management_endpoints/test_jwt_key_mapping.py
+              --ignore=tests/unit/proxy/management_endpoints/test_key_generate_prisma.py
+              --ignore=tests/unit/proxy/management_endpoints/test_roi_calculator_endpoints.py
+              --ignore=tests/unit/proxy/management_helpers/test_audit_logs_proxy.py
               tests/test_litellm/proxy/anthropic_endpoints
               tests/test_litellm/proxy/google_endpoints
               tests/test_litellm/proxy/openai_files_endpoint
@@ -188,6 +203,8 @@ jobs:
               tests/test_litellm/proxy/types_utils
               tests/test_litellm/proxy/logging_endpoints
               tests/test_litellm/proxy/test_*.py
+              tests/unit/proxy/test_proxy_server_endpoints_and_startup.py
+              tests/unit/proxy/test_proxy_utils_model_creation_and_error_logging.py
             unit-flag: proxy-infra
             workers: 4
             reruns: 2

Makefile

@@ -4,7 +4,7 @@
 .PHONY: help test test-unit test-unit-llms test-unit-proxy-guardrails test-unit-proxy-core test-unit-proxy-misc \
 	test-unit-integrations test-unit-core-utils test-unit-other test-unit-root \
 	test-proxy-unit-a test-proxy-unit-b test-integration test-unit-helm \
-	test-rust-extension \
+	test-rust-extension rust-sqlx-prepare \
 	info lint lint-inner lint-dev lint-checks format \
 	lint-basedpyright lint-e2e-basedpyright lint-basedpyright-budget-update lint-type-discipline lint-type-discipline-budget-update \
 	lint-ruff-budget lint-ruff-budget-update lint-budget-update lint-gate \
@@ -56,6 +56,7 @@ help:
 	@echo "  make test-integration   - Run integration tests"
 	@echo "  make test-unit-helm     - Run helm unit tests"
 	@echo "  make test-rust-extension - Build the Rust extension and run its public Python tests"
+	@echo "  make rust-sqlx-prepare  - Refresh litellm-rust/crates/db/.sqlx against a migrated Postgres container"
 	@echo ""
 	@echo "Heavy targets (check, lint) queue for LITELLM_GATE_SLOTS machine-wide"
 	@echo "slots (default 2; 0 disables) so parallel sessions don't thrash one machine."
@@ -306,6 +307,9 @@ test-rust-extension:
 	LITELLM_RUST=1 LITELLM_LOCAL_MODEL_COST_MAP=True \
 	"$$temporary/venv/bin/python" -I -m pytest --import-mode=importlib -m requires_rust_extension tests/test_litellm_rust
 
+rust-sqlx-prepare:
+	cd litellm-rust && cargo run -p litellm-db-testing --bin sqlx-prepare
+
 test: install-test-deps
 	$(UV_RUN) pytest tests/
 
@@ -317,13 +321,13 @@ test-unit-llms: install-test-deps
 	$(UV_RUN) pytest tests/unit/llms --tb=short -vv -n 4 --durations=20
 
 test-unit-proxy-guardrails: install-test-deps
-	$(UV_RUN) pytest tests/test_litellm/proxy/guardrails tests/test_litellm/proxy/management_endpoints tests/test_litellm/proxy/management_helpers --tb=short -vv -n 4 --durations=20
+	$(UV_RUN) pytest tests/unit/proxy/guardrails tests/unit/proxy/management_endpoints tests/unit/proxy/management_helpers --tb=short -vv -n 4 --durations=20
 
 test-unit-proxy-core: install-test-deps
-	$(UV_RUN) pytest tests/test_litellm/proxy/auth tests/test_litellm/proxy/client tests/test_litellm/proxy/db tests/test_litellm/proxy/hooks tests/test_litellm/proxy/policy_engine --tb=short -vv -n 4 --durations=20
+	$(UV_RUN) pytest tests/unit/proxy/auth tests/unit/proxy/client tests/test_litellm/proxy/db tests/unit/proxy/hooks tests/unit/proxy/policy_engine --tb=short -vv -n 4 --durations=20
 
 test-unit-proxy-misc: install-test-deps
-	$(UV_RUN) pytest tests/test_litellm/proxy/_experimental tests/test_litellm/proxy/agent_endpoints tests/test_litellm/proxy/anthropic_endpoints tests/test_litellm/proxy/common_utils tests/test_litellm/proxy/discovery_endpoints tests/test_litellm/proxy/experimental tests/test_litellm/proxy/google_endpoints tests/test_litellm/proxy/health_endpoints tests/test_litellm/proxy/image_endpoints tests/test_litellm/proxy/middleware tests/test_litellm/proxy/openai_files_endpoint tests/test_litellm/proxy/pass_through_endpoints tests/test_litellm/proxy/prompts tests/test_litellm/proxy/public_endpoints tests/test_litellm/proxy/response_api_endpoints tests/test_litellm/proxy/shutdown tests/test_litellm/proxy/spend_tracking tests/test_litellm/proxy/ui_crud_endpoints tests/test_litellm/proxy/vector_store_endpoints tests/test_litellm/proxy/test_*.py --tb=short -vv -n 4 --durations=20
+	$(UV_RUN) pytest tests/test_litellm/proxy/_experimental tests/test_litellm/proxy/agent_endpoints tests/test_litellm/proxy/anthropic_endpoints tests/test_litellm/proxy/common_utils tests/test_litellm/proxy/discovery_endpoints tests/test_litellm/proxy/experimental tests/test_litellm/proxy/google_endpoints tests/test_litellm/proxy/health_endpoints tests/test_litellm/proxy/image_endpoints tests/test_litellm/proxy/middleware tests/test_litellm/proxy/openai_files_endpoint tests/test_litellm/proxy/pass_through_endpoints tests/test_litellm/proxy/prompts tests/test_litellm/proxy/public_endpoints tests/test_litellm/proxy/response_api_endpoints tests/test_litellm/proxy/shutdown tests/test_litellm/proxy/spend_tracking tests/test_litellm/proxy/ui_crud_endpoints tests/test_litellm/proxy/vector_store_endpoints tests/test_litellm/proxy/test_*.py tests/unit/proxy/test_proxy_server_endpoints_and_startup.py tests/unit/proxy/test_proxy_utils_model_creation_and_error_logging.py tests/unit/proxy/_experimental/mcp_server/test_mcp_server_tool_calls_and_headers.py --tb=short -vv -n 4 --durations=20
 
 test-unit-integrations: install-test-deps
 	$(UV_RUN) pytest tests/unit/integrations --tb=short -vv -n 4 --durations=20

backend/routes/allowlist.py

@@ -81,6 +81,8 @@ BACKEND_PATH_PREFIXES: tuple[str, ...] = (
     # Spend / analytics
     "/spend/",
     "/analytics/",
+    "/engine/",
+    "/v1/traces",
     "/global/",
     "/user_agent",
     "/usage/",
@@ -144,6 +146,7 @@ BACKEND_EXACT_PATHS: frozenset[str] = frozenset(
     {
         "/",
         "/routes",
+        "/engine",
         "/openapi.json",
         "/docs",
         "/docs/oauth2-redirect",

basedpyright-code-budget.json

@@ -99,7 +99,7 @@
     "limit": 0
   },
   "reportUnknownArgumentType": {
-    "limit": 44358
+    "limit": 44802
   },
   "reportUnknownLambdaType": {
     "limit": 109

cookbook/litellm_proxy_server/mcp/README.md

@@ -1,37 +0,0 @@
-# Publish MCP servers in the AI Hub
-
-Set `litellm_settings.public_mcp_servers` to the concrete IDs of the servers you want listed in the public AI Hub. Pin `server_id` in each configuration entry so the publication list stays stable across deployments
-
-```yaml
-mcp_servers:
-  documentation:
-    server_id: documentation-mcp
-    url: https://mcp.example.com/mcp
-    transport: http
-    available_on_public_internet: true
-
-litellm_settings:
-  public_mcp_hub_strict_whitelist: true
-  public_mcp_servers:
-    - documentation-mcp
-```
-
-Use `documentation-mcp`, the `server_id`, in the publication list. The configuration key `documentation`, display names, and aliases are not publication IDs. Database-created servers use the ID returned by `/v1/mcp/server`
-
-The dashboard's **AI Hub > MCP Hub > Manage MCP Hub Visibility** dialog edits this same list. Its YAML example includes the selected server IDs. With database-backed configuration (`store_model_in_db: true`), a value declared in YAML is owned by that file: edit the file and reload, or remove that key from YAML to let the dashboard manage it in the database. File-backed deployments can save the list directly to their configuration file
-
-To remove all explicit entries, save an empty selection in the dialog or configure:
-
-```yaml
-litellm_settings:
-  public_mcp_hub_strict_whitelist: true
-  public_mcp_servers: []
-```
-
-## Hub listing and network access
-
-The **Hub listing** column in AI Hub identifies servers that appear in `/public/mcp_hub`. The dashboard derives this status from the current registry and publication settings. Setting `mcp_info.is_public` on a server does not publish it; that response field is derived metadata. `mcp_info.is_public_explicit` identifies registered servers included in the explicit publication list
-
-Gateway cards and server details show **All Networks** when `available_on_public_internet` is enabled or the server is explicitly published in `public_mcp_servers`. They show **Internal Only** when both are false. The per-server flag defaults to `true`; explicit publication overrides a disabled flag for compatibility. Older proxies that omit the metadata needed to determine access show **Unknown**. These labels describe allowed client IPs; authentication and tool permissions still apply
-
-The default `public_mcp_hub_strict_whitelist: true` lists only registered servers in `public_mcp_servers`. Legacy mode (`false`) additionally lists registered servers with `available_on_public_internet: true`. In legacy mode, clearing the explicit publication list leaves these automatically listed servers visible. Enable strict mode when the publication list should fully determine hub visibility

cookbook/misc/config.yaml

@@ -24,7 +24,7 @@ model_list:
   - model_name: sagemaker-completion-model
     litellm_params:
       model: sagemaker/berri-benchmarking-Llama-2-70b-chat-hf-4
-      input_cost_per_second: 0.000420  
+      cost_per_second: 0.000420
   - model_name: text-embedding-ada-002
     litellm_params: 
       model: azure/azure-embedding-model

cookbook/misc/test_responses_api.py

@@ -12,7 +12,7 @@ def encode_image(image_path):
 
 
 # Path to your image
-image_path = "litellm/proxy/logo.jpg"
+image_path = "litellm/proxy/logo.png"
 
 # Getting the Base64 string
 base64_image = encode_image(image_path)
@@ -27,7 +27,7 @@ response = client.responses.create(
                 {"type": "input_text", "text": "what color is the image"},
                 {
                     "type": "input_image",
-                    "image_url": f"data:image/jpeg;base64,{base64_image}",
+                    "image_url": f"data:image/png;base64,{base64_image}",
                 },
             ],
         }

deploy/lens/Dockerfile

@@ -0,0 +1,6 @@
+FROM python:3.12-slim
+WORKDIR /app
+RUN pip install --no-cache-dir httpx==0.28.1 pydantic==2.11.7
+COPY litellm/proxy/engine/__init__.py litellm/proxy/engine/models.py litellm/proxy/engine/trace_store.py litellm/proxy/engine/analysis.py litellm/proxy/engine/worker.py /app/engine/
+USER 65532:65532
+CMD ["python", "-m", "engine.worker"]

deploy/lens/Dockerfile.dockerignore

@@ -0,0 +1,8 @@
+**
+!litellm/
+!litellm/proxy/
+!litellm/proxy/engine/
+!litellm/proxy/engine/__init__.py
+!litellm/proxy/engine/models.py
+!litellm/proxy/engine/analysis.py
+!litellm/proxy/engine/worker.py

deploy/lens/README.md

@@ -0,0 +1,109 @@
+# Lens worker
+
+Lens reviews recorded activity and saves evidence-linked findings in the LiteLLM dashboard under Observability, Lens (`/ui/lens/`)
+
+## Start a worker
+
+Upgrade your existing LiteLLM proxy to a release that includes Lens with PostgreSQL, agent tracing (`general_settings.tracing: {store: clickhouse}`), and ClickHouse configured through `CLICKHOUSE_URL` and a separate SELECT-only `CLICKHOUSE_READER_URL`. Enable the ClickHouse callback and request/response logging to analyze LLM requests. Lens can only inspect content you actually retain
+
+In Lens, click **Set up analysis**, choose an existing virtual key or **Create worker key**, then **Generate setup command**. The LiteLLM address is filled in for you; change it only if the server running Docker needs a different network address. Copy the command and run it on your server. The dialog changes to **Analyzer connected** when the container checks in
+
+The command already contains the compatible worker image and one worker token. The selected virtual key stays on the proxy; its secret is never sent to the worker. No source checkout, environment file, or second LiteLLM deployment is needed. Keep the command private because it includes the token. The LiteLLM release provides the dashboard and APIs; the container only runs background analysis
+
+The dashboard and Compose file pin a verified worker image by digest. The image uses Linux amd64, and the generated command selects that platform. Worker image releases are independent of proxy releases: update the pinned image when changing their API contract. CI also publishes immutable commit tags for reproducible builds
+
+For deployments managed with Compose, download `compose.yaml` and provide `LITELLM_URL` and `LENS_WORKER_TOKEN` in an environment file. Its default image is already selected:
+
+```bash
+docker compose --env-file /path/to/lens.env -f compose.yaml up -d
+```
+
+Developers can build locally with `LENS_WORKER_IMAGE=litellm-lens-worker:local docker compose -f deploy/lens/compose.yaml -f deploy/lens/compose.build.yaml up -d --build`
+
+The generated command gives the worker 1 GiB of temporary memory-backed storage, shared across parallel reviews. Change `size=1g` in the Docker command or set `LENS_WORKER_TMP_SIZE` with Compose to fit your server and workload. A storage failure marks the scan as failed, cleans up temporary traces, and leaves the worker available for other scans; it does not silently truncate the review. Existing workers must be recreated with the new image and mount options
+
+The worker needs outbound HTTPS access to LiteLLM. It needs no inbound ports, provider keys, direct database access, or GPU. The proxy calls your selected model through its normal virtual-key authorization and inference pipeline; trace content reaches that model provider. Use a model with JSON output support and known token prices. One worker handles one scan at a time and can serve multiple lenses. For more throughput, start another worker with a separate credential
+
+If your deployment restricts `allowed_ips`, allow the worker's address. For workers behind a reverse proxy with `use_x_forwarded_for: true`, also configure `mcp_trusted_proxy_ranges` with that proxy's CIDRs and, when needed, `mcp_xff_num_trusted_hops`. Lens reuses these existing trusted-proxy settings. Forwarded addresses without an established trust boundary are rejected by the allowlist; accepting them would let a worker impersonate an allowed address
+
+V1 setup, manual runs, feedback, and worker credentials are restricted to proxy administrators. Proxy-admin viewers can inspect results. Regular user and team keys cannot access the Lens API. Worker credentials can serve the administrator’s lenses. Revoke it in the connection dialog when retiring a worker. Redeploy the worker alongside proxy upgrades so their API versions match
+
+## Configure a lens
+
+Choose agent runs, individual LLM requests, or both. The matching-activity preview updates as you choose an application (the recorded OpenTelemetry service.name) or, for request activity, a LiteLLM model group and add metadata conditions. It shows run names, timestamps, and trace IDs; open a run to inspect its original steps before starting analysis. Suggestions come from up to 100 recent executions and may not include every recorded attribute. You can enter other exact keys and values. Leave service and filters blank for all activity your account can access. Filters are exact key/value matches, combined with AND. Trace filters match span or resource attributes on the same span. Request filters match logged metadata, including caller metadata stored under `requester_metadata`; `tag=value` matches request tags. `swarm=research` works only if your instrumentation records that attribute
+
+Describe how the agent should behave and optionally add specific checks. Select the lookback window, team and metadata, then choose the percentage to review and an optional maximum. **100% with no maximum selects every matching run**. The preview pages through all matching activity and lets you select particular runs. Percentage sampling uses a stable hash order, rounds up, and applies the optional maximum after the percentage
+
+Choose your analysis model, parallelism and monthly budget. Parallelism controls simultaneous model calls, not the number of runs selected. New lenses run once by default. Turn on monitoring to repeat the same setup at a custom interval. **Run now** uses the same saved settings immediately, including the same lookback window and sampling. Every scan recalculates the window, so overlapping windows can review the same activity again. Duplicate a lens when you want a separate investigation without changing an existing monitor
+
+Pausing stops future scheduled scans; cancel the active scan separately if needed. The worker polls every 10 seconds; creating a lens or clicking Run now queues a scan, and due schedules are queued when the worker polls. Scans for the same lens never overlap, and its next interval starts after completion. Closing the browser does not stop the worker. Configuration edits apply to the next scan. A running scan retains its settings and selected execution IDs across retries
+
+## Read the results
+
+Needs attention shows issues, highest priority first. Patterns contains useful trends and successful behavior that may not need a fix. Each finding starts with a short explanation and a next step when useful. Expand the limitations for uncertainty and counterexamples. Evidence is grouped by run and collapsed until you need it; each quote opens the original step
+
+Use the batch selector or Scans tab to reopen previous results. Each batch keeps its own findings, settings, selected runs, coverage and cost. Older batches created before snapshot support remain available through accumulated findings. The Runs tab lists the selected batch's sample and can filter per-run observations, including runs without an observed issue and runs with insufficient evidence. These observations precede the final evidence investigation. Linked-run counts on findings include cited counterexamples, so they are not failure counts
+
+Choose **This is expected** and explain why to teach later scans about acceptable behavior. Feedback is kept with the lens and included in subsequent reviews. It does not alter historical evidence or exempt different problems
+
+## What a scan does
+
+The proxy selects executions received or updated within the configured lookback window, with a two-minute settling period. Older rows without receipt timestamps use execution end time. Overlapping scans do not increment a finding's occurrence count for the same execution ID
+
+A trace is spans sharing a trace ID within one team, not an automatically reconstructed conversation session. Requests are individual LLM calls. When both sources are enabled, requests correlated to a recorded span by response ID are excluded to reduce double counting
+
+The worker reviews the selected executions in parallel. It pages through their recorded spans and gives the first reviewer a catalog, task and outcome excerpts. The reviewer can read more original content to resolve uncertainties. Large catalogs and groups of observations are processed in bounded context windows, with every page available. Grouping retains supporting run IDs in code, so a pattern occurring thousands of times does not require a model to repeat thousands of IDs. Candidate investigators can page through supporting observations, other runs and original evidence
+
+There is no fixed total run, span, candidate or investigation-turn cutoff. Repeated or empty evidence requests stop a stalled investigation. Context windows, the configured budget, available model capacity and recorded evidence still bound practical work. The dashboard reports completed work and gaps. The investigator has no shell, browsing, code-editing or production-action tools
+
+Each model response must match a bounded JSON schema. A malformed response gets one repair attempt through the same budget controls; repeated invalid output fails the scan. Both the worker and proxy validate quoted evidence. Findings retain exact quotes and open the source trace or request. Resolve a finding after a fix, or dismiss it with a reason. A resolved finding reopens when new execution IDs support the same pattern; dismissed findings remain dismissed
+
+Coverage distinguishes eligible, sampled, reviewed, partial, and unassessable executions. Findings describe observations in the sample, not population-wide success rates or proven causes. A root span does not prove that a trace contains every expected span. Long, missing, redacted, or expired content limits the conclusions
+
+## Operations and limits
+
+PostgreSQL stores configurations, findings and all scan history, returned in pages of 50 jobs. Workers claim jobs with optimistic concurrency and a five-minute lease, renewed every 30 seconds. A disconnected job can be reclaimed up to three times. Cancellation stops subsequent work; a model call already in flight may finish and incur cost
+
+Before every model call, Lens reserves a conservative amount against the monthly lens budget. Successful calls reconcile to reported cost where pricing is available. Interrupted calls retain their reservation because the provider may have charged. A scan stops when the next reservation would exceed the limit, so it can stop with some budget remaining. Both the Lens budget and the selected virtual key’s budgets, model permissions, and rate limits apply. Analysis spend appears under that key in Virtual Keys and normal request logs, with Lens, scan, and worker IDs in request metadata. Analysis prompts and responses are redacted from spend logs; source traces and findings remain available through the administrator-only Lens API. Existing workers need a billing key assigned in **Set up analysis** before they can resume
+
+V1 requires ClickHouse for both sources. It does not reconstruct sessions from unrelated trace IDs, guarantee exhaustive reviews, cache all per-execution observations across scans, or automatically fix agent code. Trace contents can change as late spans arrive, even though a job's selected IDs are fixed. Findings should be reviewed by a person before acting on them
+
+
+## API access
+
+The UI and API use the same scan lifecycle. Authenticate with a proxy administrator credential for writes, or a proxy-admin viewer credential for reads. Worker credentials are only for worker operations
+
+```bash
+curl "$LITELLM_URL/engine" -H "Authorization: Bearer $LITELLM_API_KEY" \
+  -H 'Content-Type: application/json' -d '{
+    "name": "Research quality", "model": "your-model-alias",
+    "context": "Answer the requested question using cited, retrieved evidence.",
+    "source": "traces", "lookback_hours": 24,
+    "sample_percent": 100, "sample_size": null, "concurrency": 8,
+    "enabled": true, "interval_minutes": 1440, "monthly_budget": 50
+  }'
+
+curl "$LITELLM_URL/engine/$LENS_ID/runs" -X POST \
+  -H "Authorization: Bearer $LITELLM_API_KEY" -H 'Content-Type: application/json' -d '{}'
+
+curl "$LITELLM_URL/engine/$LENS_ID/runs?offset=0" -H "Authorization: Bearer $LITELLM_API_KEY"
+curl "$LITELLM_URL/engine/$LENS_ID/runs/$BATCH_ID" -H "Authorization: Bearer $LITELLM_API_KEY"
+```
+
+Creation queues the first batch. Posting to `/engine/{id}/runs` queues another, or returns the existing active batch. The run response contains its ID under `jobs[0].id`. Poll the batch URL for status, findings and assessments. List responses omit large result payloads; request a batch to retrieve them. Supply an optional complete `settings` object on the runs POST for a one-off override; the saved lens stays unchanged. Selection accepts `team_id`, exact `filters`, and opaque `execution_ids` returned by `/engine/preview/sample`. Preview accepts `offset` and `as_of` to keep the time window fixed while paging. Feedback uses `PATCH /engine/{id}/findings/{finding_id}` with `status` and `reason`
+
+## Quality evaluation
+
+Run the checked-in cases against a configured real model. Expected labels are used only for scoring, never passed to the model. Dev and held-out cases include missing outcomes, failed tools, recovery, handoffs, unsupported claims, repeated work, long evidence and prompt injection. The background option adds clean arithmetic traces to test rare-issue discovery at scale; those repeated synthetic cases do not establish accuracy on every production workload
+
+```bash
+python -m tests.proxy_behavior.lens.evaluate --api-base "$LITELLM_URL" \
+  --model your-model-alias --split all --background 1000 --concurrency 16 \
+  --output /tmp/lens-quality.json
+```
+
+Set `LITELLM_API_KEY` privately. This makes paid model calls. Inspect missed and unexpected per-run labels, final findings and coverage; do not equate a passing dataset with guaranteed detection on arbitrary traces
+
+The worker uses temporary disk space for trace content while reviewing it, and removes those files after each review. The Docker command supplies a writable temporary mount while keeping the application filesystem read-only
+
+To check that accepted behavior stays accepted without hiding new problems, run the evaluator with `--dataset tests/proxy_behavior/lens/feedback_cases.json`. Reports include elapsed time, model call count, reported cost when the proxy provides it, missed checks, unexpected checks, and inconclusive candidates

deploy/lens/compose.build.yaml

@@ -0,0 +1,6 @@
+services:
+  lens-worker:
+    build:
+      context: ../..
+      dockerfile: deploy/lens/Dockerfile
+    image: litellm-lens-worker:local

deploy/lens/compose.yaml

@@ -0,0 +1,12 @@
+services:
+  lens-worker:
+    image: ${LENS_WORKER_IMAGE:-ghcr.io/berriai/litellm-lens-worker@sha256:c41e932eaf3e4efbcaf8cc5027c7e93021e5b2823f21cb8785cd107e37b91c9a}
+    environment:
+      LITELLM_URL: ${LITELLM_URL:?Set the URL reachable from this container}
+      LENS_WORKER_TOKEN: ${LENS_WORKER_TOKEN:?Create a worker credential in the Lens UI}
+    restart: unless-stopped
+    read_only: true
+    tmpfs:
+      - /tmp:rw,noexec,nosuid,size=${LENS_WORKER_TMP_SIZE:-1g}
+    cap_drop: [ALL]
+    security_opt: [no-new-privileges:true]

docker/Dockerfile.non_root

@@ -103,7 +103,7 @@ ENV LITELLM_NON_ROOT=true
 
 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 && \
+    cp /app/litellm/proxy/logo.png /var/lib/litellm/assets/logo.png && \
     touch /var/lib/litellm/ui/.litellm_ui_ready
 
 RUN --mount=type=cache,target=/app/.cache/uv,id=litellm-uv-cache \

docker/docker-compose.tracing.yml

@@ -0,0 +1,62 @@
+name: litellm-tracing
+
+services:
+  litellm:
+    build:
+      context: ..
+      target: runtime
+    command: ["--config", "/app/tracing-config.yaml", "--port", "4000"]
+    environment:
+      LITELLM_MASTER_KEY: local-tracing-master-key
+      LITELLM_SALT_KEY: sk-local-tracing-salt-key
+      DATABASE_URL: postgresql://litellm:litellm@db:5432/litellm
+      STORE_MODEL_IN_DB: "True"
+      CLICKHOUSE_URL: http://default:local-tracing@clickhouse:8123
+      CLICKHOUSE_READER_URL: http://default:local-tracing@clickhouse:8123
+      CLICKHOUSE_DATABASE: litellm
+      OPENAI_API_KEY: ${OPENAI_API_KEY:-}
+    volumes:
+      - ./tracing-config.yaml:/app/tracing-config.yaml:ro
+    ports:
+      - "127.0.0.1:4002:4000"
+    depends_on:
+      db:
+        condition: service_healthy
+      clickhouse:
+        condition: service_healthy
+
+  db:
+    image: postgres:16
+    environment:
+      POSTGRES_DB: litellm
+      POSTGRES_USER: litellm
+      POSTGRES_PASSWORD: litellm
+    volumes:
+      - postgres_data:/var/lib/postgresql/data
+    ports:
+      - "127.0.0.1:15432:5432"
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U litellm -d litellm"]
+      interval: 5s
+      timeout: 5s
+      retries: 10
+
+  clickhouse:
+    image: clickhouse/clickhouse-server:26.9.6.6
+    environment:
+      CLICKHOUSE_USER: default
+      CLICKHOUSE_PASSWORD: local-tracing
+      CLICKHOUSE_DEFAULT_ACCESS_MANAGEMENT: "1"
+    volumes:
+      - clickhouse_data:/var/lib/clickhouse
+    ports:
+      - "127.0.0.1:18123:8123"
+    healthcheck:
+      test: ["CMD", "clickhouse-client", "--user", "default", "--password", "local-tracing", "--query", "SELECT 1"]
+      interval: 5s
+      timeout: 5s
+      retries: 20
+
+volumes:
+  postgres_data:
+  clickhouse_data:

docker/tracing-config.yaml

@@ -0,0 +1,10 @@
+model_list:
+  - model_name: gpt-6.1-sol
+    litellm_params:
+      model: openai/gpt-6.1-sol
+      api_key: os.environ/OPENAI_API_KEY
+
+general_settings:
+  master_key: os.environ/LITELLM_MASTER_KEY
+  tracing:
+    store: clickhouse

enterprise/litellm_enterprise/proxy/management_endpoints/project_endpoints.py

@@ -22,10 +22,8 @@ from litellm._uuid import uuid
 from litellm.proxy._types import *
 from litellm.proxy.auth.auth_checks import delete_cached_project_object
 from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
-from litellm.proxy.management_endpoints.common_utils import (
-    _is_user_team_admin,  # pyright: ignore[reportPrivateUsage]  # shared owner of team-admin membership
-    _set_object_metadata_field,
-)
+from litellm.proxy.management.teams.access import is_team_admin
+from litellm.proxy.management_endpoints.common_utils import _set_object_metadata_field
 from litellm.proxy.management_endpoints.team_admin_field_permissions import team_admin_may_manage_projects
 from litellm.proxy.management_helpers.utils import (
     management_endpoint_wrapper,
@@ -117,7 +115,7 @@ async def _check_user_permission_for_project(
         return False
 
     team: Final = LiteLLM_TeamTable.model_validate(team_row.model_dump())
-    return _is_user_team_admin(user_api_key_dict, team) or user_api_key_dict.user_id in (team.admins or [])
+    return is_team_admin(user_api_key_dict, team) or user_api_key_dict.user_id in (team.admins or [])
 
 
 async def _validate_team_exists(

enterprise/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "litellm-enterprise"
-version = "0.1.71"
+version = "0.1.72"
 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 = "0.1.71"
+version = "0.1.72"
 version_files = [
     "pyproject.toml:^version",
     "../pyproject.toml:litellm-enterprise==",

gateway/routes/allowlist.py

@@ -73,6 +73,7 @@ GATEWAY_PATH_PREFIXES: tuple[str, ...] = (
     "/v1/containers",
     "/containers",
     "/v1/evals",
+    "/v1/traces",
     "/v1/memory",
     "/queue/chat/",
     # Google data plane (v1beta is the Google AI Studio version)

litellm-proxy-extras/litellm_proxy_extras/migrations/20260921190000_agent_identity/migration.sql

@@ -0,0 +1,97 @@
+-- AlterTable
+ALTER TABLE "LiteLLM_AgentsTable" ADD COLUMN IF NOT EXISTS "enabled" BOOLEAN NOT NULL DEFAULT true,
+ADD COLUMN IF NOT EXISTS "execution_mode" TEXT NOT NULL DEFAULT 'autonomous',
+ADD COLUMN IF NOT EXISTS "identity_managed" BOOLEAN NOT NULL DEFAULT false;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_SpendLogs" ADD COLUMN IF NOT EXISTS "billing_agent_id" TEXT;
+
+-- CreateTable
+CREATE TABLE IF NOT EXISTS "LiteLLM_AgentIdentity" (
+    "agent_id" TEXT NOT NULL,
+    "active" BOOLEAN NOT NULL DEFAULT true,
+    "provider" TEXT NOT NULL,
+    "issuer" TEXT NOT NULL,
+    "tenant_id" TEXT NOT NULL,
+    "client_id" TEXT NOT NULL,
+    "service_principal_id" TEXT,
+    "required_roles" TEXT[] DEFAULT ARRAY[]::TEXT[],
+    "required_scopes" TEXT[] DEFAULT ARRAY['user_impersonation']::TEXT[],
+    "revision" TEXT NOT NULL,
+    "last_authenticated_at" TIMESTAMP(3),
+
+    CONSTRAINT "LiteLLM_AgentIdentity_pkey" PRIMARY KEY ("agent_id")
+);
+
+-- CreateTable
+CREATE TABLE IF NOT EXISTS "LiteLLM_RetiredAgentIdentity" (
+    "binding_id" TEXT NOT NULL,
+    "agent_id" TEXT,
+    "provider" TEXT NOT NULL,
+    "issuer" TEXT NOT NULL,
+    "tenant_id" TEXT NOT NULL,
+    "client_id" TEXT NOT NULL,
+
+    CONSTRAINT "LiteLLM_RetiredAgentIdentity_pkey" PRIMARY KEY ("binding_id")
+);
+
+-- CreateTable
+CREATE TABLE IF NOT EXISTS "LiteLLM_RetiredAgent" (
+    "original_agent_id" TEXT NOT NULL,
+    "retired_at" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
+
+    CONSTRAINT "LiteLLM_RetiredAgent_pkey" PRIMARY KEY ("original_agent_id")
+);
+
+-- CreateTable
+CREATE TABLE IF NOT EXISTS "LiteLLM_VerifiedSubject" (
+    "subject_id" TEXT NOT NULL,
+    "issuer" TEXT NOT NULL,
+    "tenant_id" TEXT NOT NULL,
+    "oid" TEXT NOT NULL,
+    "kind" TEXT NOT NULL DEFAULT 'human',
+    "user_id" TEXT,
+    "verified_via" TEXT NOT NULL DEFAULT 'sso_interactive',
+    "verified_at" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
+
+    CONSTRAINT "LiteLLM_VerifiedSubject_pkey" PRIMARY KEY ("subject_id")
+);
+
+-- CreateIndex
+CREATE UNIQUE INDEX IF NOT EXISTS "LiteLLM_AgentIdentity_provider_tenant_id_client_id_key" ON "LiteLLM_AgentIdentity"("provider", "tenant_id", "client_id");
+
+-- CreateIndex
+CREATE UNIQUE INDEX IF NOT EXISTS "LiteLLM_AgentIdentity_issuer_service_principal_id_key" ON "LiteLLM_AgentIdentity"("issuer", "service_principal_id");
+
+-- CreateIndex
+CREATE UNIQUE INDEX IF NOT EXISTS "LiteLLM_RetiredAgentIdentity_provider_tenant_id_client_id_key" ON "LiteLLM_RetiredAgentIdentity"("provider", "tenant_id", "client_id");
+
+-- CreateIndex
+CREATE INDEX IF NOT EXISTS "LiteLLM_VerifiedSubject_user_id_idx" ON "LiteLLM_VerifiedSubject"("user_id");
+
+-- CreateIndex
+CREATE UNIQUE INDEX IF NOT EXISTS "LiteLLM_VerifiedSubject_issuer_tenant_id_oid_key" ON "LiteLLM_VerifiedSubject"("issuer", "tenant_id", "oid");
+
+-- AddForeignKey
+DO $$
+BEGIN
+    IF NOT EXISTS (SELECT 1 FROM pg_constraint WHERE conname = 'LiteLLM_AgentIdentity_agent_id_fkey') THEN
+        ALTER TABLE "LiteLLM_AgentIdentity" ADD CONSTRAINT "LiteLLM_AgentIdentity_agent_id_fkey" FOREIGN KEY ("agent_id") REFERENCES "LiteLLM_AgentsTable"("agent_id") ON DELETE CASCADE ON UPDATE CASCADE;
+    END IF;
+END $$;
+
+-- AddForeignKey
+DO $$
+BEGIN
+    IF NOT EXISTS (SELECT 1 FROM pg_constraint WHERE conname = 'LiteLLM_RetiredAgentIdentity_agent_id_fkey') THEN
+        ALTER TABLE "LiteLLM_RetiredAgentIdentity" ADD CONSTRAINT "LiteLLM_RetiredAgentIdentity_agent_id_fkey" FOREIGN KEY ("agent_id") REFERENCES "LiteLLM_AgentsTable"("agent_id") ON DELETE SET NULL ON UPDATE CASCADE;
+    END IF;
+END $$;
+
+-- AddForeignKey
+DO $$
+BEGIN
+    IF NOT EXISTS (SELECT 1 FROM pg_constraint WHERE conname = 'LiteLLM_VerifiedSubject_user_id_fkey') THEN
+        ALTER TABLE "LiteLLM_VerifiedSubject" ADD CONSTRAINT "LiteLLM_VerifiedSubject_user_id_fkey" FOREIGN KEY ("user_id") REFERENCES "LiteLLM_UserTable"("user_id") ON DELETE CASCADE ON UPDATE CASCADE;
+    END IF;
+END $$;

litellm-proxy-extras/litellm_proxy_extras/migrations/20260925000000_add_mcp_pinned_tools/migration.sql

@@ -0,0 +1,2 @@
+-- AlterTable
+ALTER TABLE "LiteLLM_MCPServerTable" ADD COLUMN IF NOT EXISTS "pinned_tools" JSONB DEFAULT '{}';

litellm-proxy-extras/litellm_proxy_extras/migrations/20260928000000_add_daily_model_usage/migration.sql

@@ -0,0 +1,19 @@
+CREATE TABLE IF NOT EXISTS "LiteLLM_DailyModelUsage" (
+    "date" TEXT NOT NULL,
+    "model_group" TEXT NOT NULL,
+    "model" TEXT NOT NULL,
+    "custom_llm_provider" TEXT NOT NULL,
+    "task_type" TEXT NOT NULL,
+    "spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0,
+    "prompt_tokens" BIGINT NOT NULL DEFAULT 0,
+    "completion_tokens" BIGINT NOT NULL DEFAULT 0,
+    "request_count" BIGINT NOT NULL DEFAULT 0,
+    "successful_requests" BIGINT NOT NULL DEFAULT 0,
+    "failed_requests" BIGINT NOT NULL DEFAULT 0,
+    "created_at" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    "updated_at" TIMESTAMP(3) NOT NULL,
+    CONSTRAINT "LiteLLM_DailyModelUsage_pkey" PRIMARY KEY ("date", "model_group", "model", "custom_llm_provider", "task_type")
+);
+
+CREATE INDEX IF NOT EXISTS "LiteLLM_DailyModelUsage_date_idx" ON "LiteLLM_DailyModelUsage"("date");
+CREATE INDEX IF NOT EXISTS "LiteLLM_DailyModelUsage_model_group_idx" ON "LiteLLM_DailyModelUsage"("model_group");

litellm-proxy-extras/litellm_proxy_extras/migrations/20260930000000_agent_engine/migration.sql

@@ -0,0 +1,10 @@
+CREATE TABLE IF NOT EXISTS "LiteLLM_Engine" (
+    "id" TEXT NOT NULL PRIMARY KEY,
+    "version" INTEGER NOT NULL DEFAULT 0,
+    "data" JSONB NOT NULL
+);
+CREATE TABLE IF NOT EXISTS "LiteLLM_EngineWorker" (
+    "id" TEXT NOT NULL PRIMARY KEY,
+    "token_hash" TEXT NOT NULL UNIQUE,
+    "data" JSONB NOT NULL
+);

litellm-proxy-extras/litellm_proxy_extras/migrations/20261001000000_lens_run_history/migration.sql

@@ -0,0 +1,7 @@
+CREATE TABLE IF NOT EXISTS "LiteLLM_EngineRun" (
+    "id" TEXT NOT NULL PRIMARY KEY,
+    "engine_id" TEXT NOT NULL,
+    "created_at" TIMESTAMP(3) NOT NULL,
+    "data" JSONB NOT NULL
+);
+CREATE INDEX IF NOT EXISTS "LiteLLM_EngineRun_engine_id_created_at_idx" ON "LiteLLM_EngineRun"("engine_id", "created_at");

litellm-proxy-extras/litellm_proxy_extras/schema.prisma

@@ -78,6 +78,11 @@ model LiteLLM_AgentsTable {
   object_permission_id String?
   object_permission LiteLLM_ObjectPermissionTable? @relation(fields: [object_permission_id], references: [object_permission_id])
   spend      Float    @default(0.0)
+  identity_managed Boolean @default(false)
+  enabled Boolean @default(true)
+  execution_mode String @default("autonomous")
+  identity LiteLLM_AgentIdentity?
+  retired_identities LiteLLM_RetiredAgentIdentity[]
   tpm_limit  Int?
   rpm_limit  Int?
   session_tpm_limit Int?
@@ -88,6 +93,56 @@ model LiteLLM_AgentsTable {
   updated_by String
 }
 
+model LiteLLM_AgentIdentity {
+  agent_id String @id
+  active Boolean @default(true)
+  agent LiteLLM_AgentsTable @relation(fields: [agent_id], references: [agent_id], onDelete: Cascade)
+  provider String
+  issuer String
+  tenant_id String
+  client_id String
+  service_principal_id String?
+  required_roles String[] @default([])
+  required_scopes String[] @default(["user_impersonation"])
+  revision String @default(uuid())
+  last_authenticated_at DateTime?
+  @@unique([provider, tenant_id, client_id])
+  @@unique([issuer, service_principal_id])
+}
+
+model LiteLLM_RetiredAgentIdentity {
+  binding_id String @id @default(uuid())
+  agent_id String?
+  agent LiteLLM_AgentsTable? @relation(fields: [agent_id], references: [agent_id], onDelete: SetNull)
+  provider String
+  issuer String
+  tenant_id String
+  client_id String
+  @@unique([provider, tenant_id, client_id])
+}
+
+model LiteLLM_RetiredAgent {
+  original_agent_id String @id
+  retired_at DateTime @default(now())
+}
+
+model LiteLLM_VerifiedSubject {
+  subject_id String @id @default(uuid())
+  issuer String
+  tenant_id String
+  oid String
+  kind String @default("human")
+  user_id String?
+  user LiteLLM_UserTable? @relation(fields: [user_id], references: [user_id], onDelete: Cascade)
+  verified_via String @default("sso_interactive")
+  verified_at DateTime @default(now())
+  @@unique([issuer, tenant_id, oid])
+  @@index([user_id])
+}
+
+
+
+
 model LiteLLM_OrganizationTable {
 		organization_id String @id @default(uuid())
     organization_alias  String
@@ -241,6 +296,7 @@ model LiteLLM_DeletedTeamTable {
 
 // Track spend, rate limit, budget Users
 model LiteLLM_UserTable {
+    verified_subjects LiteLLM_VerifiedSubject[]
 		user_id    String @id
     user_alias String? 
     team_id    String?
@@ -322,6 +378,7 @@ model LiteLLM_MCPServerTable {
   allowed_tools String[]      @default([])
   tool_name_to_display_name Json? @default("{}")
   tool_name_to_description   Json? @default("{}")
+  pinned_tools Json? @default("{}")
   extra_headers String[]    @default([])
   static_headers Json?        @default("{}")
   // Admin-configured environment variables interpolated into static_headers
@@ -674,6 +731,7 @@ model LiteLLM_SpendLogs {
   session_id          String?
   status              String?
   mcp_namespaced_tool_name String? 
+  billing_agent_id    String?
   agent_id            String?
   proxy_server_request Json?     @default("{}")
   litellm_call_id     String?
@@ -1259,6 +1317,26 @@ model LiteLLM_DailyToolSpend {
   @@id([date, tool_name])
 }
 
+model LiteLLM_DailyModelUsage {
+  date                String
+  model_group         String
+  model               String
+  custom_llm_provider String
+  task_type           String
+  spend               Float    @default(0.0)
+  prompt_tokens       BigInt   @default(0)
+  completion_tokens   BigInt   @default(0)
+  request_count       BigInt   @default(0)
+  successful_requests BigInt   @default(0)
+  failed_requests     BigInt   @default(0)
+  created_at          DateTime @default(now())
+  updated_at          DateTime @updatedAt
+
+  @@id([date, model_group, model, custom_llm_provider, task_type])
+  @@index([date])
+  @@index([model_group])
+}
+
 // Gateway request counts recorded at the ASGI edge by
 // BillableRequestMetricsMiddleware. This is the source of truth for SGR
 // (successful gateway requests): it counts what the proxy actually answered,
@@ -1816,3 +1894,24 @@ model LiteLLM_WorkflowMessage {
   @@unique([run_id, sequence_number])
   @@index([run_id])
 }
+
+model LiteLLM_Engine {
+  id      String @id
+  version Int @default(0)
+  data    Json
+}
+
+model LiteLLM_EngineRun {
+  id         String @id
+  engine_id  String
+  created_at DateTime
+  data       Json
+
+  @@index([engine_id, created_at])
+}
+
+model LiteLLM_EngineWorker {
+  id         String @id
+  token_hash String @unique
+  data       Json
+}

litellm-proxy-extras/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "litellm-proxy-extras"
-version = "0.4.102"
+version = "0.4.103"
 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 = "0.4.102"
+version = "0.4.103"
 version_files = [
     "pyproject.toml:^version",
     "../pyproject.toml:litellm-proxy-extras==",

litellm-rust/.agents/skills/rust-tracing/SKILL.md

@@ -0,0 +1,22 @@
+---
+name: rust-tracing
+description: Add or change Rust diagnostic tracing in litellm-rust, including route spans, subscriber layers, and Python logger delivery
+---
+
+# Rust tracing
+
+Use upstream `tracing` throughout Rust, including `#[tracing::instrument]`, events, and span propagation. Centralize collection and delivery infrastructure in `crates/tracing`. Direct upstream imports still reach our configured subscriber; re-exporting macros does not control delivery. Do not introduce Rust `log` or `pyo3-log` for this path
+
+`litellm-tracing` owns shared subscriber layers, span field collection, and diagnostic processing. Keep adapters composable as `tracing_subscriber::Layer`s, with `Logger` providing host setup. Runtime-specific delivery belongs in the host bridge. The Python bridge delivers directly to the existing Python SDK logger, preserving its handlers, filtering, redaction, and request correlation. Keep Python dependencies out of `crates/tracing`
+
+Hosts configure subscribers. Keep Python execution scoped to its captured dispatch rather than installing a process-wide subscriber. Propagate both span context and dispatch across spawned work and returned streams
+
+In core, instrument execution shared by native calls and hosted machines. Use consistent route, model, provider, streaming, and outcome fields. Put status recording at shared provider boundaries instead of scattering basic logging through handlers. Keep upstream HTTP status separate from route success
+
+Use `skip_all` and explicitly selected fields. Basic tracing excludes bodies, credentials, headers, and raw error strings. Avoid automatic `ret` or `err` capture of sensitive values. Keep payload diagnostics separate and subject to existing redaction
+
+A returned stream retains its route span until exhaustion, error, or drop, with exactly one terminal outcome. Builder construction does not start a trace. Never hold a span entry guard across an await. Diagnostic tracing remains separate from lifecycle callbacks and `CustomLogger` dispatch
+
+Use `litellm_tracing::sink_layer` to compose a sink with other subscriber layers. It inherits span fields into events and emits span-close summaries with elapsed time. Test observable records, concurrent isolation, dynamic filtering, sensitive-field exclusion, and stream cancellation when changing this behavior
+
+Consult the [tracing API](https://docs.rs/tracing/latest/tracing/) and [subscriber layers](https://docs.rs/tracing-subscriber/latest/tracing_subscriber/layer/index.html) for implementation details

litellm-rust/AGENTS.md

@@ -1,5 +1,7 @@
 # Rust workspace rules
 
+For diagnostic tracing changes, follow [.agents/skills/rust-tracing/SKILL.md](.agents/skills/rust-tracing/SKILL.md)
+
 ## Test placement
 
 - Never create a `tests.rs` (or `test.rs`) file under `src/`, and never `#[path = "tests.rs"] mod tests;`
@@ -16,7 +18,9 @@ Use [`#[rstest]`](https://docs.rs/rstest/latest/rstest/attr.rstest.html) for new
 ## Error definitions
 
 - A crate's errors live in `src/error.rs`, defined with `thiserror`, and re-exported from `lib.rs`
+- Put message templates in the variant's `#[error(...)]` declaration. Callers pass only the small typed arguments needed to fill them, never `Error::Variant(format!(...))` or a preformatted message. Keep the smallest set of neutral variants that callers need to distinguish; different wording or providers do not justify new variants
 - Default to one top-level `Error` enum per crate, with one variant per failure mode and a `#[error(...)]` message on each. A failure mode is something a caller handles differently (phase, status code, retry, a message Python parity pins exactly); failures no caller tells apart share one variant and differ only in its message
+- Keep shared error enums minimal and provider-neutral. Provider names, credential types, configuration fields, and setup guidance belong in caller-supplied data, not dedicated variants or hardcoded shared messages. Reuse a variant for the same failure mode across providers, such as `MissingApiBase { provider: "Azure", guidance: "..." }`. An exact parity message does not justify a provider-specific variant when caller-supplied context can preserve it
 - Wrap a lower-level error as a variant with `#[from]` or `#[source]` instead of flattening it to a string
 - Exception: split into separate types when different functions fail in disjoint ways, especially when different callers see them. A shared enum would force every caller to match variants its function can never return
 - Name a split type after what went wrong (a unit struct is fine for a single failure mode), not after the function that returns it

litellm-rust/Cargo.toml

@@ -12,12 +12,18 @@ repository = "https://github.com/BerriAI/litellm"
 litellm-config = { path = "crates/config" }
 litellm-router = { path = "crates/router" }
 litellm-tracing = { path = "crates/tracing" }
+litellm-traces = { path = "crates/traces" }
 litellm-core = { path = "crates/core" }
+litellm-gateway-mcp = { path = "crates/gateway-mcp" }
 litellm-gateway = { path = "crates/gateway" }
 litellm-gateway-inference = { path = "crates/gateway-inference" }
 litellm-gateway-auth = { path = "crates/gateway-auth" }
+litellm-gateway-management = { path = "crates/gateway-management" }
+litellm-gateway-ui = { path = "crates/gateway-ui" }
 litellm-coroutine = { path = "crates/coroutine" }
 litellm-host = { path = "crates/host" }
+litellm-host-http = { path = "crates/host-http" }
+litellm-host-native = { path = "crates/host-native" }
 litellm-callbacks-legacy-python = { path = "crates/callbacks-legacy-python" }
 litellm-framing = { path = "crates/framer" }
 litellm-auth = { path = "crates/auth" }
@@ -34,8 +40,10 @@ litellm-secrets-azure = { path = "crates/secrets-azure" }
 litellm-secrets-cyberark = { path = "crates/secrets-cyberark" }
 litellm-http = { path = "crates/http" }
 litellm-llms = { path = "crates/llms" }
-litellm-types = { path = "crates/types" }
+litellm-llms-types = { path = "crates/llms-types" }
 litellm-core-utils = { path = "crates/core-utils" }
+litellm-db = { path = "crates/db" }
+litellm-db-testing = { path = "crates/db-testing" }
 litellm-cache = { path = "crates/cache" }
 litellm-cache-azure-blob = { path = "crates/cache-azure-blob" }
 litellm-cache-memory = { path = "crates/cache-memory" }
@@ -56,6 +64,8 @@ litellm-python-compat = { path = "crates/python-compat" }
 
 tracing = "0.1"
 axum = { version = "0.8.9", default-features = false, features = ["http1", "tokio", "multipart"] }
+axum-login = "0.18.0"
+tower-sessions = { version = "0.14.0", features = ["memory-store"] }
 bytes = "1"
 http = "1"
 google-cloud-auth = { version = "1.16.0", default-features = false }
@@ -65,6 +75,7 @@ proptest = "1.7.0"
 pyo3 = "0.29.2"
 pyo3-async-runtimes = { version = "0.29.0", features = ["tokio-runtime"] }
 rand = "0.8"
+macro_rules_attribute = "0.2.3"
 schemars = "1"
 reqwest = { version = "0.12", default-features = false, features = ["json", "multipart", "rustls-tls", "http2", "stream"] }
 qdrant-client = { version = "1.19.0", default-features = false }
@@ -80,6 +91,7 @@ serde = { version = "1.0", features = ["derive"] }
 serde_json = { version = "1.0", features = ["float_roundtrip"] }
 serde_with = { version = "=3.16.1", default-features = false, features = ["std", "macros"] }
 sha2 = "0.10"
+sqlx = { version = "0.9.0", default-features = false, features = ["json", "macros", "postgres", "runtime-tokio", "chrono", "tls-rustls-ring-native-roots"] }
 subtle = "2"
 thiserror = "2.0"
 tokenizers = { version = "0.23.1", default-features = false, features = ["onig"] }

litellm-rust/clippy.toml

@@ -1,4 +1,4 @@
-# The Tokio runtime is reached only through `host-python/src/execution.rs`, whose fork gate
+# The Tokio runtime is reached only through `host-python/src/runtime.rs`, whose fork gate
 # must see every entry. Going around it makes a fork-after-use hang instead of raising.
 disallowed-methods = [
     { path = "pyo3_async_runtimes::tokio::get_runtime", reason = "use litellm_host_python::run_sync / run_sync_value" },
@@ -12,6 +12,13 @@ disallowed-methods = [
     { path = "reqwest::ClientBuilder::danger_accept_invalid_certs", reason = "set HttpClientConfig::verify instead" },
     { path = "reqwest::ClientBuilder::identity", reason = "set HttpClientConfig::client_certificate instead" },
     { path = "reqwest::ClientBuilder::use_preconfigured_tls", reason = "HttpClientConfig owns the TLS configuration" },
+    { path = "sqlx::query", reason = "use sqlx::query! or query_file! so the SQL is checked against the migrated schema" },
+    { path = "sqlx::query_as", reason = "use sqlx::query_as! or query_file_as! so the SQL is checked against the migrated schema" },
+    { path = "sqlx::query_scalar", reason = "use sqlx::query_scalar! so the SQL is checked against the migrated schema" },
+    { path = "sqlx::query_with", reason = "use sqlx::query! or query_file! so the SQL is checked against the migrated schema" },
+    { path = "sqlx::query_as_with", reason = "use sqlx::query_as! or query_file_as! so the SQL is checked against the migrated schema" },
+    { path = "sqlx::query_scalar_with", reason = "use sqlx::query_scalar! so the SQL is checked against the migrated schema" },
+    { path = "sqlx::raw_sql", reason = "raw_sql is unchecked; use the checked query macros" },
 ]
 
 # Every outbound client comes from litellm_http::HttpClientPool so it honors the host's TLS,

litellm-rust/crates/auth-azure/src/native.rs

@@ -133,7 +133,7 @@ impl NativeAzureTokenAcquirer {
         let token = credential
             .get_token(&[scope.as_str()], None)
             .await
-            .map_err(|error| Error::AzureTokenAcquisition(error.to_string()))?;
+            .map_err(|error| Error::CredentialAcquisition(error.to_string().into()))?;
         let expires_on = u64::try_from(token.expires_on.unix_timestamp())
             .ok()
             .map(|seconds| UNIX_EPOCH + Duration::from_secs(seconds));
@@ -250,7 +250,12 @@ fn validate_authority(request: &NativeAzureRequest) -> Result<(), Error> {
     let Some(authority) = authority else {
         return Ok(());
     };
-    let url = url::Url::parse(authority.value()).map_err(|_| Error::InvalidAzureAuthority)?;
+    let url = url::Url::parse(authority.value()).map_err(|_| {
+        Error::InvalidConfiguration(
+            "Azure authority must be an HTTPS origin without credentials, query, or fragment"
+                .into(),
+        )
+    })?;
     if url.scheme() != "https"
         || url.host_str().is_none()
         || !url.username().is_empty()
@@ -259,7 +264,10 @@ fn validate_authority(request: &NativeAzureRequest) -> Result<(), Error> {
         || url.fragment().is_some()
         || !matches!(url.path(), "" | "/")
     {
-        return Err(Error::InvalidAzureAuthority);
+        return Err(Error::InvalidConfiguration(
+            "Azure authority must be an HTTPS origin without credentials, query, or fragment"
+                .into(),
+        ));
     }
     Ok(())
 }
@@ -368,7 +376,9 @@ fn trusted_source(sources: &[InputSource]) -> InputSource {
 }
 
 fn mixed_sources<T>() -> Result<T, Error> {
-    Err(Error::MixedAzureCredentialSources)
+    Err(Error::InvalidConfiguration(
+        "request-controlled Azure auth inputs cannot be combined with host credentials".into(),
+    ))
 }
 
 fn build_credential(
@@ -433,7 +443,12 @@ fn build_credential(
         NativeAzureRequest::DeveloperTools { .. } => DeveloperToolsCredential::new(None)
             .map(|credential| credential as Arc<dyn TokenCredential>),
     }
-    .map_err(|error| Error::AzureCredentialInitialization(error.to_string()))
+    .map_err(|error| {
+        Error::InvalidConfiguration(litellm_auth_types::ErrorDetail::failed(
+            "Azure credential initialization",
+            error,
+        ))
+    })
 }
 
 fn client_options(
@@ -638,7 +653,7 @@ mod tests {
         assert_eq!(transport.requests.lock().unwrap().len(), 6);
     }
 
-    #[test]
+    #[rstest::rstest]
     fn request_authority_requires_request_owned_client_secret_identity() {
         let error = ValidatedAzureRequest::new(sourced_client_secret(
             InputSource::Deployment,
@@ -647,10 +662,13 @@ mod tests {
         ))
         .unwrap_err();
 
-        assert!(matches!(
+        assert_eq!(
             error,
-            litellm_auth_types::Error::MixedAzureCredentialSources
-        ));
+            litellm_auth_types::Error::InvalidConfiguration(
+                "request-controlled Azure auth inputs cannot be combined with host credentials"
+                    .into()
+            )
+        );
     }
 
     #[test]
@@ -665,24 +683,24 @@ mod tests {
         assert_eq!(request.credential_source(), InputSource::Request);
     }
 
-    #[test]
-    fn authority_is_restricted_to_an_https_origin() {
-        for authority in [
-            "http://login.example",
-            "https://user@login.example",
-            "https://login.example/tenant",
-            "https://login.example?target=other",
-        ] {
-            let error = ValidatedAzureRequest::new(sourced_client_secret(
-                InputSource::Deployment,
-                InputSource::Deployment,
-                authority,
-            ))
-            .unwrap_err();
-            assert!(matches!(
-                error,
-                litellm_auth_types::Error::InvalidAzureAuthority
-            ));
-        }
+    #[rstest::rstest]
+    #[case::http("http://login.example")]
+    #[case::userinfo("https://user@login.example")]
+    #[case::path("https://login.example/tenant")]
+    #[case::query("https://login.example?target=other")]
+    fn authority_is_restricted_to_an_https_origin(#[case] authority: &str) {
+        let error = ValidatedAzureRequest::new(sourced_client_secret(
+            InputSource::Deployment,
+            InputSource::Deployment,
+            authority,
+        ))
+        .unwrap_err();
+        assert_eq!(
+            error,
+            litellm_auth_types::Error::InvalidConfiguration(
+                "Azure authority must be an HTTPS origin without credentials, query, or fragment"
+                    .into()
+            )
+        );
     }
 }

litellm-rust/crates/auth-azure/src/resolve.rs

@@ -91,7 +91,9 @@ impl AzureAuthService {
             AzureCredentialPlan::Caller(caller) => {
                 let credential = caller.acquire().await?;
                 if credential.secret().expose().is_empty() {
-                    return Err(Error::EmptyAzureToken);
+                    return Err(Error::EmptyCallerCredential(
+                        "Azure AD token provider returned an empty token",
+                    ));
                 }
                 Ok(Some(Sourced::new(credential, InputSource::Deployment)))
             }
@@ -104,7 +106,11 @@ impl AzureAuthService {
             } => {
                 let assertion = resolve_reference(inputs, env_lookup, reference.value())
                     .await?
-                    .ok_or(Error::UnresolvedOidcReference)?;
+                    .ok_or_else(|| {
+                        Error::CredentialAcquisition(
+                            "Azure OIDC reference did not resolve to a value".into(),
+                        )
+                    })?;
                 let request = ValidatedAzureRequest::new(NativeAzureRequest::ClientAssertion {
                     tenant_id,
                     client_id,
@@ -167,7 +173,7 @@ pub(crate) fn select_auth_plan(
                 .map(|selector| Sourced::new(selector, value.source()))
         })
         .transpose()
-        .map_err(|_| Error::InvalidAzureSelector)?;
+        .map_err(|_| Error::InvalidConfiguration("invalid Azure credential selector".into()))?;
     let federated_token_file = configured_string(
         &inputs.federated_token_file,
         AZURE_FEDERATED_TOKEN_FILE_ENV,
@@ -257,7 +263,9 @@ fn select_native_plan(
     let selection_source = selected.source();
 
     match selected.into_value() {
-        AzureCredentialType::ClientSecretCredential => Err(Error::MissingClientSecretFields),
+        AzureCredentialType::ClientSecretCredential => Err(Error::InvalidConfiguration(
+            "ClientSecretCredential requires tenant_id, client_id, and client_secret".into(),
+        )),
         AzureCredentialType::WorkloadIdentityCredential => {
             Ok(AzureCredentialPlan::Native(ValidatedAzureRequest::new(
                 workload_request(tenant_id, client_id, federated_token_file, scope, authority)?,
@@ -341,9 +349,17 @@ fn workload_request(
     authority: Option<Sourced<String>>,
 ) -> Result<NativeAzureRequest, Error> {
     Ok(NativeAzureRequest::WorkloadIdentity {
-        tenant_id: tenant_id.ok_or(Error::MissingWorkloadTenant)?,
-        client_id: client_id.ok_or(Error::MissingWorkloadClient)?,
-        token_file_path: token_file_path.ok_or(Error::MissingWorkloadTokenFile)?,
+        tenant_id: tenant_id.ok_or_else(|| {
+            Error::InvalidConfiguration("WorkloadIdentityCredential requires tenant_id".into())
+        })?,
+        client_id: client_id.ok_or_else(|| {
+            Error::InvalidConfiguration("WorkloadIdentityCredential requires client_id".into())
+        })?,
+        token_file_path: token_file_path.ok_or_else(|| {
+            Error::InvalidConfiguration(
+                "WorkloadIdentityCredential requires azure_federated_token_file".into(),
+            )
+        })?,
         scope,
         authority,
     })
@@ -394,10 +410,11 @@ async fn resolve_reference(
             .map_or(CredentialLookup::Missing, CredentialLookup::Found),
         CredentialRef::None => return Ok(None),
         CredentialRef::File(_) | CredentialRef::Request(_) | CredentialRef::Host(_) => {
-            let resolver = inputs
-                .credential_resolver
-                .as_ref()
-                .ok_or(Error::MissingHostResolver)?;
+            let resolver = inputs.credential_resolver.as_ref().ok_or_else(|| {
+                Error::InvalidConfiguration(
+                    "credential reference requires a host credential resolver".into(),
+                )
+            })?;
             resolver.resolve(reference).await?
         }
     };
@@ -415,7 +432,9 @@ fn oidc_reference(
     };
     let value = token.value().expose();
     if token.source() == InputSource::Request && value.starts_with("oidc/") {
-        return Err(Error::RequestAzureCredentialReference);
+        return Err(Error::InvalidConfiguration(
+            "request-controlled Azure credential references are not allowed".into(),
+        ));
     }
     if let Some(name) = value.strip_prefix("oidc/env/") {
         return non_empty_reference(name, "OIDC environment reference")
@@ -437,14 +456,20 @@ fn oidc_reference(
         )));
     }
     if value.starts_with("oidc/") {
-        return Err(Error::UnsupportedOidcReference);
+        return Err(Error::InvalidConfiguration(
+            "unsupported OIDC reference".into(),
+        ));
     }
     Ok(None)
 }
 
 fn non_empty_reference(value: &str, kind: &str) -> Result<String, Error> {
     if value.is_empty() {
-        return Err(Error::EmptyReference(kind.to_string()));
+        return Err(Error::InvalidConfiguration(
+            litellm_auth_types::ErrorDetail::Empty {
+                subject: kind.into(),
+            },
+        ));
     }
     Ok(value.to_string())
 }
@@ -493,7 +518,7 @@ mod tests {
                         expires_on: None,
                     })
                 } else {
-                    Err(Error::AzureTokenAcquisition(format!("{kind} failed")))
+                    Err(Error::CredentialAcquisition(kind.into()))
                 }
             })
         }
@@ -602,7 +627,7 @@ mod tests {
         assert!(error.to_string().contains("unsupported OIDC reference"));
     }
 
-    #[test]
+    #[rstest::rstest]
     fn request_oidc_reference_is_rejected_before_lookup() {
         let params = json!({
             "azure_ad_token": "oidc/env/ASSERTION",
@@ -624,7 +649,12 @@ mod tests {
         })
         .unwrap_err();
 
-        assert!(matches!(error, Error::RequestAzureCredentialReference));
+        assert_eq!(
+            error,
+            Error::InvalidConfiguration(
+                "request-controlled Azure credential references are not allowed".into()
+            )
+        );
     }
 
     #[tokio::test]
@@ -723,6 +753,7 @@ mod tests {
         assert_eq!(credential.value().secret().expose(), "caller-token");
     }
 
+    #[rstest::rstest]
     #[tokio::test]
     async fn empty_caller_token_is_rejected() {
         let error = AzureAuthService::default()
@@ -730,6 +761,9 @@ mod tests {
             .await
             .unwrap_err();
 
-        assert!(matches!(error, Error::EmptyAzureToken));
+        assert_eq!(
+            error,
+            Error::EmptyCallerCredential("Azure AD token provider returned an empty token")
+        );
     }
 }

litellm-rust/crates/auth-azure/src/types.rs

@@ -117,7 +117,12 @@ fn string_config(
         None => Ok(ConfigValue::Absent),
         Some(Value::Null) => Ok(ConfigValue::ExplicitNone(source)),
         Some(Value::String(value)) => Ok(ConfigValue::Value(Sourced::new(value.clone(), source))),
-        Some(_) => Err(Error::InvalidFieldType(name.to_string())),
+        Some(_) => Err(Error::InvalidConfiguration(
+            litellm_auth_types::ErrorDetail::InvalidType {
+                field: name.into(),
+                expected: "a string or null",
+            },
+        )),
     }
 }
 

litellm-rust/crates/auth-gcp/Cargo.toml

@@ -19,3 +19,6 @@ tokio.workspace = true
 gcp_auth = "0.12.7"
 google-cloud-auth = { workspace = true, optional = true }
 http = { workspace = true, optional = true }
+
+[dev-dependencies]
+rstest.workspace = true

litellm-rust/crates/auth-gcp/src/lib.rs

@@ -299,7 +299,7 @@ fn validate_request_credentials(configured: &str) -> Result<&str, Error> {
                 .map(str::to_string)
         });
     if token_uri.as_deref() != Some(GOOGLE_OAUTH_TOKEN_ENDPOINT) {
-        return Err(Error::RequestVertexTokenEndpoint);
+        return Err(Error::InvalidConfiguration("request-controlled Vertex credentials must use the canonical Google OAuth token endpoint".into()));
     }
     Ok(configured)
 }
@@ -376,10 +376,20 @@ fn optional_credentials(
                     .map(SecretValue::new)
                     .map(|value| Sourced::new(value, source))
                     .map(Some)
-                    .map_err(|error| Error::InvalidFieldType(format!("{}: {error}", names[0])));
+                    .map_err(|error| {
+                        Error::InvalidConfiguration(litellm_auth_types::ErrorDetail::failed(
+                            "credential serialization",
+                            error,
+                        ))
+                    });
             }
             Some(_) => {
-                return Err(Error::InvalidFieldType(names[0].to_string()));
+                return Err(Error::InvalidConfiguration(
+                    litellm_auth_types::ErrorDetail::InvalidType {
+                        field: names[0].into(),
+                        expected: "a string or null",
+                    },
+                ));
             }
         }
     }
@@ -397,7 +407,12 @@ fn optional_string(params: &Map<String, Value>, names: &[&str]) -> Result<Option
             Some(Value::String(value)) if value.trim().is_empty() => continue,
             Some(Value::String(value)) => return Ok(Some(value.clone())),
             Some(_) => {
-                return Err(Error::InvalidFieldType(names[0].to_string()));
+                return Err(Error::InvalidConfiguration(
+                    litellm_auth_types::ErrorDetail::InvalidType {
+                        field: names[0].into(),
+                        expected: "a string or null",
+                    },
+                ));
             }
         }
     }
@@ -411,7 +426,10 @@ fn non_empty_env(env_lookup: &dyn Fn(&str) -> Option<String>, name: &str) -> Opt
 }
 
 fn auth_acquisition_error(error: gcp_auth::Error) -> Error {
-    Error::VertexTokenAcquisition(error.to_string())
+    Error::CredentialAcquisition(litellm_auth_types::ErrorDetail::failed(
+        "Vertex AI credentials",
+        error,
+    ))
 }
 
 #[cfg(test)]
@@ -612,20 +630,15 @@ mod tests {
         );
     }
 
-    #[test]
-    fn request_credentials_require_canonical_token_endpoint() {
-        assert!(
-            validate_request_credentials(r#"{"token_uri":"https://oauth2.googleapis.com/token"}"#)
-                .is_ok()
-        );
-        assert!(matches!(
-            validate_request_credentials(r#"{"token_uri":"http://127.0.0.1/token"}"#),
-            Err(Error::RequestVertexTokenEndpoint)
-        ));
-        assert!(matches!(
-            validate_request_credentials("{}"),
-            Err(Error::RequestVertexTokenEndpoint)
-        ));
+    #[rstest::rstest]
+    #[case::canonical_endpoint(r#"{"token_uri":"https://oauth2.googleapis.com/token"}"#, true)]
+    #[case::noncanonical_endpoint(r#"{"token_uri":"http://127.0.0.1/token"}"#, false)]
+    #[case::missing_endpoint("{}", false)]
+    fn request_credentials_require_canonical_token_endpoint(
+        #[case] credentials: &str,
+        #[case] accepted: bool,
+    ) {
+        assert_eq!(validate_request_credentials(credentials).is_ok(), accepted);
     }
 
     #[tokio::test]

litellm-rust/crates/auth-types/Cargo.toml

@@ -12,4 +12,5 @@ thiserror.workspace = true
 veil.workspace = true
 
 [dev-dependencies]
+rstest.workspace = true
 tokio.workspace = true

litellm-rust/crates/auth-types/src/credential.rs

@@ -86,7 +86,9 @@ impl CredentialPlan {
             Self::Caller(caller) => {
                 let credential = caller.acquire().await?;
                 if credential.secret().expose().is_empty() {
-                    return Err(Error::EmptyCallerCredential);
+                    return Err(Error::EmptyCallerCredential(
+                        "credential caller returned an empty credential",
+                    ));
                 }
                 Ok(CredentialPlanResolution::Resolved(credential))
             }
@@ -147,10 +149,15 @@ mod tests {
 
     impl CredentialResolver for FailingResolver {
         fn resolve<'a>(&'a self, _reference: &'a CredentialRef) -> CredentialLookupFuture<'a> {
-            Box::pin(async { Err(Error::UnresolvedOidcReference) })
+            Box::pin(async {
+                Err(Error::CredentialAcquisition(
+                    "host credential lookup failed".into(),
+                ))
+            })
         }
     }
 
+    #[rstest::rstest]
     #[tokio::test]
     async fn acquisition_failure_is_terminal() {
         let resolver = CredentialResolverHandle::new(Arc::new(FailingResolver));
@@ -161,6 +168,9 @@ mod tests {
             .await
             .expect_err("acquisition errors cannot become fallback");
 
-        assert_eq!(error, Error::UnresolvedOidcReference);
+        assert_eq!(
+            error,
+            Error::CredentialAcquisition("host credential lookup failed".into())
+        );
     }
 }

litellm-rust/crates/auth-types/src/error.rs

@@ -2,84 +2,16 @@ use thiserror::Error as ThisError;
 
 #[derive(Clone, Debug, ThisError, PartialEq, Eq)]
 pub enum Error {
-    #[error("invalid authentication configuration: credential header already exists")]
-    ExistingCredentialHeader,
-    #[error(
-        "invalid authentication configuration: credential plan is not allowed by the provider auth policy"
-    )]
-    DisallowedCredentialPlan,
-    #[error("invalid authentication configuration: credential cannot be empty")]
-    EmptyCredential,
-    #[error("invalid authentication configuration: invalid Azure credential selector")]
-    InvalidAzureSelector,
-    #[error(
-        "invalid authentication configuration: ClientSecretCredential requires tenant_id, client_id, and client_secret"
-    )]
-    MissingClientSecretFields,
-    #[error("invalid authentication configuration: WorkloadIdentityCredential requires tenant_id")]
-    MissingWorkloadTenant,
-    #[error("invalid authentication configuration: WorkloadIdentityCredential requires client_id")]
-    MissingWorkloadClient,
-    #[error(
-        "invalid authentication configuration: WorkloadIdentityCredential requires azure_federated_token_file"
-    )]
-    MissingWorkloadTokenFile,
-    #[error(
-        "invalid authentication configuration: credential reference requires a host credential resolver"
-    )]
-    MissingHostResolver,
-    #[error(
-        "invalid authentication configuration: caller credential plan requires provider-specific inputs"
-    )]
-    MissingCallerInputs,
-    #[error("invalid authentication configuration: credential header {0} already exists")]
-    DuplicateHeader(&'static str),
-    #[error("invalid authentication configuration: {0} must be a string or null")]
-    InvalidFieldType(String),
-    #[error("invalid authentication configuration: unsupported OIDC reference")]
-    UnsupportedOidcReference,
-    #[error("invalid authentication configuration: {0} cannot be empty")]
-    EmptyReference(String),
-    #[error("invalid authentication configuration: Azure credential initialization failed: {0}")]
-    AzureCredentialInitialization(String),
-    #[error(
-        "invalid authentication configuration: Azure authority must be an HTTPS origin without credentials, query, or fragment"
-    )]
-    InvalidAzureAuthority,
-    #[error(
-        "invalid authentication configuration: request-controlled Azure auth inputs cannot be combined with host credentials"
-    )]
-    MixedAzureCredentialSources,
-    #[error(
-        "invalid authentication configuration: request-controlled Azure credential references are not allowed"
-    )]
-    RequestAzureCredentialReference,
-    #[error(
-        "invalid authentication configuration: host credentials cannot be sent to a request-controlled Azure endpoint"
-    )]
-    RequestAzureCredentialDestination,
-    #[error(
-        "invalid authentication configuration: credentials cannot be sent to a request-controlled Vertex AI endpoint"
-    )]
-    RequestVertexCredentialDestination,
-    #[error(
-        "invalid authentication configuration: request-controlled Vertex credentials must use the canonical Google OAuth token endpoint"
-    )]
-    RequestVertexTokenEndpoint,
+    #[error("invalid authentication configuration: {0}")]
+    InvalidConfiguration(#[source] ErrorDetail),
     #[error("credential acquisition failed: {0}")]
-    AzureTokenAcquisition(String),
-    #[error("credential acquisition failed: Vertex AI credentials: {0}")]
-    VertexTokenAcquisition(String),
+    CredentialAcquisition(#[source] ErrorDetail),
+    #[error("credential caller failed: {0}")]
+    EmptyCallerCredential(&'static str),
     #[error("{0}")]
     ProviderAuthentication(String),
     #[error("credential acquisition failed: {}", .0.iter().map(ToString::to_string).collect::<Vec<_>>().join("; "))]
     CredentialChain(Vec<Error>),
-    #[error("credential caller failed: credential caller returned an empty credential")]
-    EmptyCallerCredential,
-    #[error("credential caller failed: Azure AD token provider returned an empty token")]
-    EmptyAzureToken,
-    #[error("credential acquisition failed: Azure OIDC reference did not resolve to a value")]
-    UnresolvedOidcReference,
     #[error(
         "Missing {provider} API Key - Set `api_key` or the {environment_variable} environment variable"
     )]
@@ -87,34 +19,87 @@ pub enum Error {
         provider: &'static str,
         environment_variable: &'static str,
     },
-    #[error(
-        "Missing {provider} API Base - Set {environment_variable} environment variable or pass api_base parameter"
-    )]
+    #[error("Missing {provider} API Base - {guidance}")]
     MissingApiBase {
         provider: &'static str,
-        environment_variable: &'static str,
+        guidance: &'static str,
     },
-    #[error(
-        "Missing Azure API Base - Set `api_base` or the AZURE_API_BASE environment variable. Expected format: https://<resource-name>.services.ai.azure.com/anthropic"
-    )]
-    MissingAzureApiBase,
     #[error("invalid authentication header")]
     InvalidHeader,
 }
 
-#[cfg(test)]
-mod tests {
-    use super::Error;
+#[derive(Clone, Debug, PartialEq, Eq, thiserror::Error)]
+pub enum ErrorDetail {
+    #[error("{0}")]
+    Message(String),
+    #[error("{field} must be {expected}")]
+    InvalidType {
+        field: String,
+        expected: &'static str,
+    },
+    #[error("{subject} cannot be empty")]
+    Empty { subject: String },
+    #[error("credential header {0} already exists")]
+    DuplicateHeader(&'static str),
+    #[error("{operation} failed: {source}")]
+    Failed {
+        operation: &'static str,
+        #[source]
+        source: ErrorSource,
+    },
+}
 
-    #[test]
-    fn missing_api_key_names_provider_and_environment_variable() {
-        assert_eq!(
-            Error::MissingApiKey {
-                provider: "Anthropic",
-                environment_variable: "ANTHROPIC_API_KEY",
-            }
-            .to_string(),
-            "Missing Anthropic API Key - Set `api_key` or the ANTHROPIC_API_KEY environment variable"
-        );
+impl ErrorDetail {
+    pub fn failed(
+        operation: &'static str,
+        source: impl std::error::Error + Send + Sync + 'static,
+    ) -> Self {
+        Self::Failed {
+            operation,
+            source: ErrorSource::new(source),
+        }
     }
 }
+
+impl From<String> for ErrorDetail {
+    fn from(message: String) -> Self {
+        Self::Message(message)
+    }
+}
+
+impl From<&str> for ErrorDetail {
+    fn from(message: &str) -> Self {
+        Self::Message(message.into())
+    }
+}
+
+#[derive(Clone, Debug)]
+pub struct ErrorSource(std::sync::Arc<dyn std::error::Error + Send + Sync>);
+
+impl std::ops::Deref for ErrorSource {
+    type Target = dyn std::error::Error + Send + Sync;
+
+    fn deref(&self) -> &Self::Target {
+        self.0.as_ref()
+    }
+}
+
+impl std::fmt::Display for ErrorSource {
+    fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        std::fmt::Display::fmt(&self.0, formatter)
+    }
+}
+
+impl ErrorSource {
+    pub fn new(error: impl std::error::Error + Send + Sync + 'static) -> Self {
+        Self(std::sync::Arc::new(error))
+    }
+}
+
+impl PartialEq for ErrorSource {
+    fn eq(&self, other: &Self) -> bool {
+        std::sync::Arc::ptr_eq(&self.0, &other.0)
+    }
+}
+
+impl Eq for ErrorSource {}

litellm-rust/crates/auth-types/src/http.rs

@@ -21,13 +21,17 @@ pub fn apply_credential(
     placement: CredentialPlacement,
 ) -> Result<Vec<(String, String)>, Error> {
     if credential.trim().is_empty() {
-        return Err(Error::EmptyCredential);
+        return Err(Error::InvalidConfiguration(
+            "credential cannot be empty".into(),
+        ));
     }
     if headers
         .iter()
         .any(|(name, _)| name.eq_ignore_ascii_case(placement.header_name()))
     {
-        return Err(Error::DuplicateHeader(placement.header_name()));
+        return Err(Error::InvalidConfiguration(
+            crate::ErrorDetail::DuplicateHeader(placement.header_name()),
+        ));
     }
     let value = match placement {
         CredentialPlacement::Bearer => format!("Bearer {credential}"),

litellm-rust/crates/auth-types/src/lib.rs

@@ -50,7 +50,7 @@ pub use credential::{
     CredentialFileRef, CredentialLookup, CredentialLookupFuture, CredentialPlan,
     CredentialPlanResolution, CredentialRef, CredentialResolver, CredentialResolverHandle,
 };
-pub use error::Error;
+pub use error::{Error, ErrorDetail, ErrorSource};
 pub use http::CredentialPlacement;
 pub use policy::{CredentialPlanKind, CredentialRule, ExistingHeaderBehavior, ProviderAuthPolicy};
 pub use secret::SecretValue;

litellm-rust/crates/auth-types/src/policy.rs

@@ -47,14 +47,20 @@ impl ProviderAuthPolicy {
         if self.has_existing_credential(&headers) {
             return match self.existing_header_behavior {
                 ExistingHeaderBehavior::Preserve => Ok(headers),
-                ExistingHeaderBehavior::Reject => Err(Error::ExistingCredentialHeader),
+                ExistingHeaderBehavior::Reject => Err(Error::InvalidConfiguration(
+                    "credential header already exists".into(),
+                )),
             };
         }
         let rule = self
             .rules
             .iter()
             .find(|rule| rule.kind == kind)
-            .ok_or(Error::DisallowedCredentialPlan)?;
+            .ok_or_else(|| {
+                Error::InvalidConfiguration(
+                    "credential plan is not allowed by the provider auth policy".into(),
+                )
+            })?;
         apply_credential(headers, credential.secret().expose(), rule.placement)
     }
 }

litellm-rust/crates/auth-types/src/token.rs

@@ -39,7 +39,33 @@ impl TokenProviderHandle {
         Self(caller)
     }
 
+    pub fn from_callback<F, Fut>(acquire: F) -> Self
+    where
+        F: Fn() -> Fut + Send + Sync + 'static,
+        Fut: Future<Output = Result<ResolvedCredential, Error>> + Send + 'static,
+    {
+        Self::new(Arc::new(CallbackTokenProvider(acquire)))
+    }
+
     pub async fn acquire(&self) -> Result<ResolvedCredential, Error> {
         self.0.acquire().await
     }
 }
+
+struct CallbackTokenProvider<F>(F);
+
+impl<F> std::fmt::Debug for CallbackTokenProvider<F> {
+    fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        formatter.write_str("CallbackTokenProvider")
+    }
+}
+
+impl<F, Fut> TokenProvider for CallbackTokenProvider<F>
+where
+    F: Fn() -> Fut + Send + Sync,
+    Fut: Future<Output = Result<ResolvedCredential, Error>> + Send + 'static,
+{
+    fn acquire(&self) -> TokenFuture<'_> {
+        Box::pin((self.0)())
+    }
+}

litellm-rust/crates/auth-types/tests/error.rs

@@ -0,0 +1,74 @@
+use litellm_auth_types::Error;
+use rstest::rstest;
+
+#[rstest]
+#[case::api_key(
+    Error::MissingApiKey { provider: "Example", environment_variable: "EXAMPLE_API_KEY" },
+    "Missing Example API Key - Set `api_key` or the EXAMPLE_API_KEY environment variable"
+)]
+#[case::another_api_key(
+    Error::MissingApiKey { provider: "Custom", environment_variable: "CUSTOM_KEY" },
+    "Missing Custom API Key - Set `api_key` or the CUSTOM_KEY environment variable"
+)]
+#[case::api_base(
+    Error::MissingApiBase { provider: "Example", guidance: "Pass api_base" },
+    "Missing Example API Base - Pass api_base"
+)]
+#[case::another_api_base(
+    Error::MissingApiBase { provider: "Custom", guidance: "Set CUSTOM_ENDPOINT" },
+    "Missing Custom API Base - Set CUSTOM_ENDPOINT"
+)]
+#[case::configuration(
+    Error::InvalidConfiguration("credential selector is invalid".into()),
+    "invalid authentication configuration: credential selector is invalid"
+)]
+#[case::acquisition(
+    Error::CredentialAcquisition("token expired".into()),
+    "credential acquisition failed: token expired"
+)]
+#[case::caller(
+    Error::EmptyCallerCredential("empty token"),
+    "credential caller failed: empty token"
+)]
+#[case::provider(
+    Error::ProviderAuthentication("provider rejected credentials".into()),
+    "provider rejected credentials"
+)]
+#[case::chain(
+    Error::CredentialChain(vec![
+        Error::CredentialAcquisition("token expired".into()),
+        Error::EmptyCallerCredential("empty token"),
+    ]),
+    "credential acquisition failed: credential acquisition failed: token expired; credential caller failed: empty token"
+)]
+fn display_preserves_failure_phase_and_caller_context(
+    #[case] error: Error,
+    #[case] expected: &str,
+) {
+    assert_eq!(error.to_string(), expected);
+}
+
+#[rstest]
+#[case::configuration(true)]
+#[case::acquisition(false)]
+fn contextual_failures_keep_the_original_source(#[case] configuration: bool) {
+    use litellm_auth_types::ErrorDetail;
+
+    let detail = ErrorDetail::failed(
+        "test credential",
+        std::io::Error::from(std::io::ErrorKind::PermissionDenied),
+    );
+    let error = if configuration {
+        Error::InvalidConfiguration(detail)
+    } else {
+        Error::CredentialAcquisition(detail)
+    };
+    let source = std::iter::successors(Some(&error as &dyn std::error::Error), |error| {
+        error.source()
+    })
+    .find_map(|error| error.downcast_ref::<std::io::Error>())
+    .expect("the original credential error remains available");
+    assert_eq!(source.kind(), std::io::ErrorKind::PermissionDenied);
+    assert!(error.to_string().contains("test credential failed:"));
+    assert!(error.to_string().ends_with(&source.to_string()));
+}

litellm-rust/crates/auth-types/tests/token.rs

@@ -0,0 +1,104 @@
+use std::{
+    error::Error as StdError,
+    future::{Future, poll_fn},
+    sync::{
+        Arc,
+        atomic::{AtomicBool, AtomicUsize, Ordering},
+    },
+    task::Poll,
+    time::{Duration, SystemTime},
+};
+
+use litellm_auth_types::{
+    Error, ErrorDetail, ResolvedCredential, SecretValue, TokenProviderHandle,
+};
+use rstest::rstest;
+
+fn credential(index: usize, access_token: bool) -> ResolvedCredential {
+    let token = SecretValue::new(format!("credential-{index}"));
+    if access_token {
+        return ResolvedCredential::AccessToken {
+            token,
+            expires_on: Some(SystemTime::UNIX_EPOCH + Duration::from_secs(index as u64)),
+        };
+    }
+    ResolvedCredential::Static(token)
+}
+
+#[rstest]
+#[case::static_secret(false)]
+#[case::access_token(true)]
+#[tokio::test]
+async fn callbacks_acquire_fresh_credentials_on_demand(#[case] access_token: bool) {
+    let calls = Arc::new(AtomicUsize::new(0));
+    let callback_calls = calls.clone();
+    let provider = TokenProviderHandle::from_callback(move || {
+        let index = callback_calls.fetch_add(1, Ordering::SeqCst);
+        async move {
+            tokio::task::yield_now().await;
+            Ok(credential(index, access_token))
+        }
+    });
+    let cloned = provider.clone();
+
+    assert_eq!(calls.load(Ordering::SeqCst), 0);
+    assert_eq!(
+        provider.acquire().await.unwrap(),
+        credential(0, access_token)
+    );
+    assert_eq!(calls.load(Ordering::SeqCst), 1);
+    assert_eq!(cloned.acquire().await.unwrap(), credential(1, access_token));
+    assert_eq!(calls.load(Ordering::SeqCst), 2);
+}
+
+#[rstest]
+#[tokio::test]
+async fn callback_errors_preserve_the_original_source() {
+    let provider = TokenProviderHandle::from_callback(|| async {
+        Err(Error::CredentialAcquisition(ErrorDetail::failed(
+            "caller credential",
+            std::io::Error::from(std::io::ErrorKind::PermissionDenied),
+        )))
+    });
+
+    let error = provider.acquire().await.unwrap_err();
+    assert!(matches!(error, Error::CredentialAcquisition(_)));
+    let source = std::iter::successors(Some(&error as &(dyn StdError + 'static)), |error| {
+        (*error).source()
+    })
+    .find_map(|error| error.downcast_ref::<std::io::Error>())
+    .unwrap();
+    assert_eq!(source.kind(), std::io::ErrorKind::PermissionDenied);
+}
+
+struct Release(Arc<AtomicBool>);
+
+impl Drop for Release {
+    fn drop(&mut self) {
+        self.0.store(true, Ordering::SeqCst);
+    }
+}
+
+#[rstest]
+#[tokio::test]
+async fn cancelling_acquisition_drops_the_callback_future() {
+    let released = Arc::new(AtomicBool::new(false));
+    let callback_released = released.clone();
+    let provider = TokenProviderHandle::from_callback(move || {
+        let released = callback_released.clone();
+        async move {
+            let _release = Release(released);
+            std::future::pending().await
+        }
+    });
+
+    let mut acquisition = Box::pin(provider.acquire());
+    poll_fn(|context| {
+        assert!(acquisition.as_mut().poll(context).is_pending());
+        assert!(!released.load(Ordering::SeqCst));
+        Poll::Ready(())
+    })
+    .await;
+    drop(acquisition);
+    assert!(released.load(Ordering::SeqCst));
+}

litellm-rust/crates/cache-disk/Cargo.toml

@@ -9,7 +9,7 @@ repository.workspace = true
 litellm-cache.workspace = true
 py_literal = "0.4.0"
 rand.workspace = true
-rusqlite = { version = "0.40", features = ["bundled"] }
+rusqlite = { version = "0.39", features = ["bundled"] }
 serde-pickle = "1.2"
 serde_json.workspace = true
 tokio.workspace = true

litellm-rust/crates/cache-gcs/tests/cache.rs

@@ -56,6 +56,7 @@ async fn set_writes_encoded_object_and_headers(#[future(awt)] server: MockServer
 )]
 #[case::missing("missing", ResponseTemplate::new(404), Ok(None))]
 #[case::server_error("server-error", ResponseTemplate::new(500), Err(Error::Unavailable))]
+#[case::unauthorized("unauthorized", ResponseTemplate::new(401), Err(Error::Unavailable))]
 #[case::invalid(
     "invalid",
     ResponseTemplate::new(200).set_body_string("not json"),