scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-14 23:21:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 156

chore: merge litellm_internal_staging into litellm_fix_priority_tpm_enforcement

Browse source 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
This commit is contained in:
shivam 2026-07-28 19:53:54 +00:00
commit 53cc5eb8f7
1341 changed files with 24058 additions and 12180 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
.github/workflows/test-linting.yml vendored
View file

@ -104,6 +104,7 @@ jobs:
- name: Check basedpyright budget (delta vs base)
env:
BASE_SHA: ${{ github.event.pull_request.base.sha }}
NODE_OPTIONS: --max-old-space-size=12288
run: |
(uv run --no-sync basedpyright --outputjson || true) | uv run --no-sync python scripts/type_check_gate.py --base "$BASE_SHA"

10
.github/workflows/test-litellm-ui-lint.yml vendored
View file

@ -29,7 +29,15 @@ jobs:
id: changed
env:
BASE_SHA: ${{ github.event.pull_request.base.sha }}
HEAD_SHA: ${{ github.event.pull_request.head.sha }}
run: |
# base.sha is the base branch tip from when the PR was opened, while
# actions/checkout leaves HEAD on a merge of the PR into the *current*
# base tip. "$BASE_SHA"...HEAD therefore spans every base-branch commit
# landed since, so a PR that touches no UI file still gets linted
# against hundreds of other people's files. Diff the PR head against its
# own merge base instead, which is exactly what this PR changed.
merge_base=$(git merge-base "$BASE_SHA" "$HEAD_SHA")
: > "$RUNNER_TEMP/prettier_files.txt"
: > "$RUNNER_TEMP/eslint_files.txt"
while IFS= read -r f; do
@ -41,7 +49,7 @@ jobs:
*.json | *.css | *.scss | *.md | *.mdx | *.yml | *.yaml | *.html)
printf '%s\n' "$f" >> "$RUNNER_TEMP/prettier_files.txt" ;;
esac
done < <(git diff --name-only --diff-filter=ACMR --relative "$BASE_SHA"...HEAD -- .)
done < <(git diff --name-only --diff-filter=ACMR --relative "$merge_base" "$HEAD_SHA" -- .)
if [ -s "$RUNNER_TEMP/prettier_files.txt" ] || [ -s "$RUNNER_TEMP/eslint_files.txt" ]; then
echo "has_files=true" >> "$GITHUB_OUTPUT"
else

151
.github/workflows/test_server_root_path.yml vendored
View file

@ -1,151 +0,0 @@
name: Test Proxy SERVER_ROOT_PATH Routing
permissions:
contents: read
on:
pull_request:
branches:
- main
- litellm_internal_staging
- litellm_oss_staging
- "litellm_**"
jobs:
test-server-root-path:
runs-on: ubuntu-latest
timeout-minutes: 30
strategy:
fail-fast: false
matrix:
root_path: ["/api/v1", "/llmproxy"]
steps:
- name: Checkout repository
uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
with:
persist-credentials: false
- name: Free up disk space
run: |
sudo rm -rf /usr/local/lib/android /usr/share/dotnet /opt/ghc /usr/local/share/boost
sudo apt-get clean
df -h /
- name: Set up Docker Buildx
uses: docker/setup-buildx-action@8d2750c68a42422c14e847fe6c8ac0403b4cbd6f # v3.12.0
- name: Build Docker image
uses: docker/build-push-action@0adf9959216b96bec444f325f1e493d4aa344497 # v6.14.0
with:
context: .
file: ./docker/Dockerfile.non_root
tags: litellm-test:${{ github.sha }}
load: true
push: false
- name: Start LiteLLM container with SERVER_ROOT_PATH
run: |
docker run -d \
--name litellm-test \
-p 4000:4000 \
-e SERVER_ROOT_PATH="${{ matrix.root_path }}" \
-e LITELLM_MASTER_KEY="sk-1234" \
litellm-test:${{ github.sha }} \
--detailed_debug
- name: Wait for container to be healthy
run: |
echo "Waiting for LiteLLM to start..."
max_attempts=30
attempt=0
while [ $attempt -lt $max_attempts ]; do
if docker logs litellm-test 2>&1 | grep -q "Uvicorn running"; then
echo "LiteLLM started successfully"
break
fi
attempt=$((attempt + 1))
echo "Attempt $attempt/$max_attempts - waiting for server to start..."
sleep 2
done
if [ $attempt -eq $max_attempts ]; then
echo "Server failed to start within timeout"
docker logs litellm-test
exit 1
fi
sleep 5
- name: Show container logs
if: always()
run: docker logs litellm-test
- name: Test UI endpoint with root path
run: |
ROOT_PATH="${{ matrix.root_path }}"
echo "Testing UI at: http://localhost:4000${ROOT_PATH}/ui/"
for i in 1 2 3; do
content=$(curl -sL --max-time 5 -H "Authorization: Bearer sk-1234" "http://localhost:4000${ROOT_PATH}/ui/")
if echo "$content" | grep -q -E "(html|<!DOCTYPE|<head|<body)"; then
echo "UI page contains valid HTML content"
exit 0
fi
echo "Attempt $i/3 - no valid HTML, retrying in 5s..."
sleep 5
done
echo "UI page does not contain expected HTML content"
echo "Response: $content"
docker logs litellm-test
exit 1
- name: Setup Node for Playwright
uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
with:
node-version: "20"
- name: Install e2e deps and Chromium
working-directory: tests/e2e/ui
run: |
retry() {
local attempt=1
local max_attempts=4
until "$@"; do
if [ "$attempt" -ge "$max_attempts" ]; then
echo "Command failed after $attempt attempts: $*"
return 1
fi
echo "Attempt $attempt failed: $*. Retrying in $((attempt * 15))s..."
sleep $((attempt * 15))
attempt=$((attempt + 1))
done
}
npm config set fetch-retries 5
npm config set fetch-retry-mintimeout 20000
npm config set fetch-retry-maxtimeout 120000
retry npm ci
retry npx playwright install --with-deps chromium
- name: Run SERVER_ROOT_PATH redirect e2e
working-directory: tests/e2e/ui
env:
SERVER_ROOT_PATH: ${{ matrix.root_path }}
run: npx playwright test --config=serverRootPath.config.ts
- name: Upload Playwright artifacts on failure
if: failure()
uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
with:
name: playwright-trace-${{ strategy.job-index }}
path: tests/e2e/ui/test-results/
retention-days: 7
- name: Cleanup
if: always()
run: |
docker stop litellm-test || true
docker rm litellm-test || true

4
backend/routes/allowlist.py
View file

@ -70,6 +70,10 @@ BACKEND_PATH_PREFIXES: tuple[str, ...] = (
"/project/",
"/memory/",
"/mcp/",
# Control plane (see the List Endpoints + Tables standard). Every resource
# eventually moves under this prefix, so allowlist it once rather than
# per-resource.
"/management/v1/",
# Spend / analytics
"/spend/",
"/analytics/",

24
basedpyright-code-budget.json
View file

@ -1,9 +1,9 @@
{
"reportAny": {
"limit": 37484
"limit": 34906
},
"reportArgumentType": {
"limit": 2704
"limit": 2701
},
"reportAssignmentType": {
"limit": 330
@ -12,7 +12,7 @@
"limit": 516
},
"reportCallIssue": {
"limit": 124
"limit": 123
},
"reportConstantRedefinition": {
"limit": 59
@ -24,7 +24,7 @@
"limit": 42
},
"reportExplicitAny": {
"limit": 10389
"limit": 10230
},
"reportFunctionMemberAccess": {
"limit": 11
@ -54,10 +54,10 @@
"limit": 0
},
"reportMissingParameterType": {
"limit": 5900
"limit": 5893
},
"reportMissingTypeArgument": {
"limit": 15903
"limit": 15886
},
"reportMissingTypeStubs": {
"limit": 41
@ -99,31 +99,31 @@
"limit": 0
},
"reportUnknownArgumentType": {
"limit": 45894
"limit": 45870
},
"reportUnknownLambdaType": {
"limit": 113
},
"reportUnknownMemberType": {
"limit": 40539
"limit": 40525
},
"reportUnknownParameterType": {
"limit": 20403
"limit": 20384
},
"reportUnknownVariableType": {
"limit": 32141
"limit": 32099
},
"reportUnnecessaryCast": {
"limit": 177
},
"reportUnnecessaryComparison": {
"limit": 1025
"limit": 1023
},
"reportUnnecessaryContains": {
"limit": 7
},
"reportUnnecessaryIsInstance": {
"limit": 1209
"limit": 1206
},
"reportUntypedBaseClass": {
"limit": 165

46
db_scripts/backfill_daily_tool_spend.sql Normal file
View file

@ -0,0 +1,46 @@
-- One-shot backfill of the LiteLLM_DailyToolSpend rollup from the per-request
-- LiteLLM_SpendLogToolIndex x LiteLLM_SpendLogs tables.
--
-- This is an opt-in, manual operation. New deployments do not need it: the
-- rollup is written at request time from the moment the release is deployed.
-- Run it only if you want the Cost Optimization "Spend by tool" card to show
-- history from before the deploy, and only once.
--
-- IMPORTANT caveats before running:
--
-- 1. Pre-deploy index rows may include tools that were merely DECLARED in a
-- request body but never invoked (the release this ships with stops
-- recording those). For agentic clients that declare many tools per
-- request, backfilled history attributes each request's full spend to
-- every declared tool, overstating per-tool spend. Post-deploy rows do not
-- have this problem. If your traffic is mostly such clients, consider not
-- backfilling.
--
-- 2. Coverage is bounded by spend-log retention: rows older than
-- maximum_spend_logs_retention_period are already gone.
--
-- 3. Replace the cutover timestamp below with the time you deployed the
-- release, so backfilled per-request rows cannot double-count on top of
-- rollup rows the new writer already created. ON CONFLICT DO NOTHING is a
-- second guard for (date, tool_name) buckets the writer already touched:
-- such buckets keep the writer's numbers and skip the backfill's.
--
-- Usage:
-- psql "$DATABASE_URL" -v cutover="'2026-07-25T00:00:00Z'" -f db_scripts/backfill_daily_tool_spend.sql
SET TIME ZONE 'UTC';
INSERT INTO "LiteLLM_DailyToolSpend" (date, tool_name, spend, total_tokens, request_count, created_at, updated_at)
SELECT
to_char(ti.start_time, 'YYYY-MM-DD') AS date,
ti.tool_name,
COALESCE(SUM(sl.spend), 0) AS spend,
COALESCE(SUM(sl.total_tokens), 0) AS total_tokens,
COUNT(*) AS request_count,
now() AS created_at,
now() AS updated_at
FROM "LiteLLM_SpendLogToolIndex" ti
JOIN "LiteLLM_SpendLogs" sl ON sl.request_id = ti.request_id
WHERE ti.start_time < :cutover::timestamptz
GROUP BY 1, 2
ON CONFLICT (date, tool_name) DO NOTHING;

38
enterprise/litellm_enterprise/proxy/hooks/managed_files.py
View file

@ -316,26 +316,34 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints):
where_clause: Dict[str, Any] = {"file_purpose": "batch", **owner_filter}
if after:
where_clause["id"] = {"gt": after}
cursor_row = (
await self.prisma_client.db.litellm_managedobjecttable.find_first(
where={**where_clause, "unified_object_id": after}
)
)
if cursor_row is None:
raise HTTPException(
status_code=400,
detail=f"Invalid 'after' cursor: no batch found with id '{after}'.",
)
fetch_limit = limit or 20
if target_model_names:
# Oversample so post-fetch model-name filtering still has enough rows.
fetch_limit = max(fetch_limit * 3, 100)
page_size = limit or 20
cursor_args: Dict[str, Any] = (
{"cursor": {"unified_object_id": after}, "skip": 1} if after else {}
)
batches = await self.prisma_client.db.litellm_managedobjecttable.find_many(
where=where_clause,
take=fetch_limit,
order={"created_at": "desc"},
take=page_size + 1,
order=[{"created_at": "desc"}, {"unified_object_id": "desc"}],
**cursor_args,
)
batch_objects: List[LiteLLMBatch] = []
for batch in batches:
try:
# Stop once we have enough after filtering
if len(batch_objects) >= (limit or 20):
break
has_more = len(batches) > page_size
batch_objects: List[LiteLLMBatch] = []
for batch in batches[:page_size]:
try:
batch_data = (
json.loads(batch.file_object)
if isinstance(batch.file_object, str)
@ -351,9 +359,7 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints):
)
continue
return build_list_page(
batch_objects, has_more=len(batch_objects) == (limit or 20)
)
return build_list_page(batch_objects, has_more=has_more)
async def get_user_created_file_ids(
self, user_api_key_dict: UserAPIKeyAuth, model_object_ids: List[str]

191
enterprise/litellm_enterprise/proxy/management_endpoints/project_endpoints.py
View file

@ -11,7 +11,8 @@ Endpoints for /project operations
#### PROJECT MANAGEMENT ####
import json
from typing import List, Optional, Union
from collections.abc import Mapping, Sequence
from typing import TYPE_CHECKING
from fastapi import APIRouter, Depends, HTTPException, Request
@ -25,15 +26,24 @@ from litellm.proxy.management_helpers.utils import (
)
from litellm.proxy.utils import PrismaClient, handle_exception_on_proxy
if TYPE_CHECKING:
from prisma import models as prisma_models
from prisma.actions import LiteLLM_TeamTableActions
router = APIRouter()
def _team_table(prisma_client: PrismaClient) -> "LiteLLM_TeamTableActions[prisma_models.LiteLLM_TeamTable]":
team_table: LiteLLM_TeamTableActions[prisma_models.LiteLLM_TeamTable] = prisma_client.db.litellm_teamtable
return team_table
async def _check_user_permission_for_project(
user_api_key_dict: UserAPIKeyAuth,
team_id: Optional[str],
team_id: str | None,
prisma_client: PrismaClient,
require_admin: bool = False,
team_object: Optional[LiteLLM_TeamTable] = None,
team_object: LiteLLM_TeamTable | None = None,
) -> bool:
"""
Check if user has permission to manage a project.
@ -57,9 +67,7 @@ async def _check_user_permission_for_project(
team = team_object
if team is None:
team = await prisma_client.db.litellm_teamtable.find_unique(
where={"team_id": team_id}
)
team = await _team_table(prisma_client).find_unique(where={"team_id": team_id})
if team and team.admins:
return user_api_key_dict.user_id in team.admins
@ -70,9 +78,9 @@ async def _check_user_permission_for_project(
async def _validate_team_exists(
team_id: str,
prisma_client: PrismaClient,
):
) -> "prisma_models.LiteLLM_TeamTable":
"""Validate that a team exists. Returns the team row."""
team = await prisma_client.db.litellm_teamtable.find_unique(
team = await _team_table(prisma_client).find_unique(
where={"team_id": team_id},
)
@ -89,7 +97,7 @@ async def _validate_team_exists(
def _check_team_project_limits(
team_object: LiteLLM_TeamTable,
data: Union[NewProjectRequest, UpdateProjectRequest],
data: NewProjectRequest | UpdateProjectRequest,
) -> None:
"""
Check that project limits respect its parent Team's limits.
@ -108,16 +116,12 @@ def _check_team_project_limits(
if data.max_budget is not None and data.max_budget < 0:
raise HTTPException(
status_code=400,
detail={
"error": f"max_budget cannot be negative. Received: {data.max_budget}"
},
detail={"error": f"max_budget cannot be negative. Received: {data.max_budget}"},
)
if data.soft_budget is not None and data.soft_budget < 0:
raise HTTPException(
status_code=400,
detail={
"error": f"soft_budget cannot be negative. Received: {data.soft_budget}"
},
detail={"error": f"soft_budget cannot be negative. Received: {data.soft_budget}"},
)
# --- soft_budget < max_budget ---
@ -131,7 +135,7 @@ def _check_team_project_limits(
)
# --- Validate project models are a subset of team models ---
project_models = getattr(data, "models", None)
project_models = data.models
team_models = team_object.models or []
if project_models and len(team_models) > 0:
# If team has 'all-proxy-models', skip validation as it allows all models
@ -148,11 +152,7 @@ def _check_team_project_limits(
# --- Validate project max_budget <= team max_budget ---
# Team stores budget fields directly (max_budget, tpm_limit, rpm_limit)
# unlike Project which uses a separate LiteLLM_BudgetTable relation
if (
data.max_budget is not None
and team_object.max_budget is not None
and data.max_budget > team_object.max_budget
):
if data.max_budget is not None and team_object.max_budget is not None and data.max_budget > team_object.max_budget:
raise HTTPException(
status_code=400,
detail={
@ -161,11 +161,7 @@ def _check_team_project_limits(
)
# --- Validate project tpm_limit <= team tpm_limit ---
if (
data.tpm_limit is not None
and team_object.tpm_limit is not None
and data.tpm_limit > team_object.tpm_limit
):
if data.tpm_limit is not None and team_object.tpm_limit is not None and data.tpm_limit > team_object.tpm_limit:
raise HTTPException(
status_code=400,
detail={
@ -174,11 +170,7 @@ def _check_team_project_limits(
)
# --- Validate project rpm_limit <= team rpm_limit ---
if (
data.rpm_limit is not None
and team_object.rpm_limit is not None
and data.rpm_limit > team_object.rpm_limit
):
if data.rpm_limit is not None and team_object.rpm_limit is not None and data.rpm_limit > team_object.rpm_limit:
raise HTTPException(
status_code=400,
detail={
@ -189,19 +181,19 @@ def _check_team_project_limits(
async def _create_budget_for_project(
data: NewProjectRequest,
user_id: Optional[str],
user_id: str | None,
litellm_proxy_admin_name: str,
prisma_client: PrismaClient,
) -> str:
"""Create a budget for the project and return budget_id."""
budget_params = LiteLLM_BudgetTable.model_fields.keys()
_json_data = data.json(exclude_none=True)
_json_data: Mapping[str, object] = data.json(exclude_none=True)
_budget_data = {k: v for k, v in _json_data.items() if k in budget_params}
budget_row = LiteLLM_BudgetTable(**_budget_data)
budget_row = LiteLLM_BudgetTable.model_validate(_budget_data)
new_budget = prisma_client.jsonify_object(budget_row.json(exclude_none=True))
_budget = await prisma_client.db.litellm_budgettable.create(
_budget: prisma_models.LiteLLM_BudgetTable = await prisma_client.db.litellm_budgettable.create(
data={
**new_budget,
"created_by": user_id or litellm_proxy_admin_name,
@ -214,8 +206,8 @@ async def _create_budget_for_project(
async def _set_project_object_permission(
data: NewProjectRequest,
prisma_client: Optional[PrismaClient],
) -> Optional[str]:
prisma_client: PrismaClient | None,
) -> str | None:
"""
Creates the LiteLLM_ObjectPermissionTable record for the project.
Returns the object_permission_id if created, otherwise None.
@ -224,7 +216,7 @@ async def _set_project_object_permission(
return None
if data.object_permission is not None:
created_object_permission = (
created_object_permission: prisma_models.LiteLLM_ObjectPermissionTable = (
await prisma_client.db.litellm_objectpermissiontable.create(
data=data.object_permission.model_dump(exclude_none=True),
)
@ -344,8 +336,7 @@ async def new_project(
raise HTTPException(
status_code=403,
detail={
"error": "Only premium users can add tags to projects. "
+ CommonProxyErrors.not_premium_user.value
"error": "Only premium users can add tags to projects. " + CommonProxyErrors.not_premium_user.value
},
)
@ -353,8 +344,7 @@ async def new_project(
raise HTTPException(
status_code=403,
detail={
"error": "Project management is an enterprise feature. "
+ CommonProxyErrors.not_premium_user.value
"error": "Project management is an enterprise feature. " + CommonProxyErrors.not_premium_user.value
},
)
@ -375,13 +365,11 @@ async def new_project(
)
# Validate team exists and get team object with budget
team_object = await _validate_team_exists(
team_id=data.team_id, prisma_client=prisma_client
)
team_object = await _validate_team_exists(team_id=data.team_id, prisma_client=prisma_client)
# Validate project limits against team limits
_check_team_project_limits(
team_object=LiteLLM_TeamTable(**team_object.model_dump()),
team_object=LiteLLM_TeamTable.model_validate(team_object.model_dump()),
data=data,
)
@ -391,7 +379,7 @@ async def new_project(
user_api_key_dict=user_api_key_dict,
team_id=data.team_id,
prisma_client=prisma_client,
team_object=LiteLLM_TeamTable(**team_object.model_dump()),
team_object=LiteLLM_TeamTable.model_validate(team_object.model_dump()),
)
if not has_permission:
@ -449,17 +437,13 @@ async def new_project(
value=getattr(data, field),
)
new_project_row = prisma_client.jsonify_object(
project_row.json(exclude_none=True)
)
new_project_row = prisma_client.jsonify_object(project_row.json(exclude_none=True))
# Remove budget fields (following organization_endpoints.py pattern)
new_project_row = _remove_budget_fields_from_project_data(new_project_row)
verbose_proxy_logger.info(
f"new_project_row: {json.dumps(new_project_row, indent=2)}"
)
response = await prisma_client.db.litellm_projecttable.create(
verbose_proxy_logger.info(f"new_project_row: {json.dumps(new_project_row, indent=2)}")
response: prisma_models.LiteLLM_ProjectTable = await prisma_client.db.litellm_projecttable.create(
data={
**new_project_row, # type: ignore
},
@ -469,9 +453,7 @@ async def new_project(
return response
except Exception as e:
verbose_proxy_logger.exception(
"litellm.proxy.management_endpoints.project_endpoints.new_project(): Exception occured - {}".format(
str(e)
)
"litellm.proxy.management_endpoints.project_endpoints.new_project(): Exception occured - {}".format(str(e))
)
raise handle_exception_on_proxy(e)
@ -539,8 +521,7 @@ async def update_project(
raise HTTPException(
status_code=403,
detail={
"error": "Only premium users can add tags to projects. "
+ CommonProxyErrors.not_premium_user.value
"error": "Only premium users can add tags to projects. " + CommonProxyErrors.not_premium_user.value
},
)
@ -548,8 +529,7 @@ async def update_project(
raise HTTPException(
status_code=403,
detail={
"error": "Project management is an enterprise feature. "
+ CommonProxyErrors.not_premium_user.value
"error": "Project management is an enterprise feature. " + CommonProxyErrors.not_premium_user.value
},
)
@ -576,9 +556,9 @@ async def update_project(
)
# Fetch existing project
existing_project = await prisma_client.db.litellm_projecttable.find_unique(
where={"project_id": data.project_id}
)
existing_project: (
prisma_models.LiteLLM_ProjectTable | None
) = await prisma_client.db.litellm_projecttable.find_unique(where={"project_id": data.project_id})
if existing_project is None:
raise ProxyException(
@ -595,9 +575,7 @@ async def update_project(
target_team_id = data.team_id or existing_project.team_id
target_team_obj = None
if target_team_id is not None:
target_team_obj = await _validate_team_exists(
team_id=target_team_id, prisma_client=prisma_client
)
target_team_obj = await _validate_team_exists(team_id=target_team_id, prisma_client=prisma_client)
has_permission = await _check_user_permission_for_project(
user_api_key_dict=user_api_key_dict,
@ -620,32 +598,26 @@ async def update_project(
team_id=data.team_id,
prisma_client=prisma_client,
team_object=(
LiteLLM_TeamTable(**target_team_obj.model_dump())
if target_team_obj
else None
LiteLLM_TeamTable.model_validate(target_team_obj.model_dump()) if target_team_obj else None
),
)
if not can_assign_to_target:
raise HTTPException(
status_code=403,
detail={
"error": "Cannot reassign project to a team you are not an admin of"
},
detail={"error": "Cannot reassign project to a team you are not an admin of"},
)
# Validate project limits against team limits
if target_team_obj is not None:
_check_team_project_limits(
team_object=LiteLLM_TeamTable(**target_team_obj.model_dump()),
team_object=LiteLLM_TeamTable.model_validate(target_team_obj.model_dump()),
data=data,
)
# Prepare update data
update_data = data.json(exclude_none=True, exclude={"project_id"})
update_data = prisma_client.jsonify_object(update_data)
update_data["updated_by"] = (
user_api_key_dict.user_id or litellm_proxy_admin_name
)
update_data["updated_by"] = user_api_key_dict.user_id or litellm_proxy_admin_name
# Handle budget updates
budget_fields = LiteLLM_BudgetTable.model_fields.keys()
@ -671,21 +643,17 @@ async def update_project(
if existing_project.object_permission_id:
# Update existing permission
await prisma_client.db.litellm_objectpermissiontable.update(
where={
"object_permission_id": existing_project.object_permission_id
},
where={"object_permission_id": existing_project.object_permission_id},
data=object_permission_data,
)
else:
# Create new permission
created_permission = (
created_permission: prisma_models.LiteLLM_ObjectPermissionTable = (
await prisma_client.db.litellm_objectpermissiontable.create(
data=object_permission_data,
)
)
update_data["object_permission_id"] = (
created_permission.object_permission_id
)
update_data["object_permission_id"] = created_permission.object_permission_id
# Handle metadata fields
for field in LiteLLM_ManagementEndpoint_MetadataFields:
@ -698,7 +666,7 @@ async def update_project(
update_data = _remove_budget_fields_from_project_data(update_data)
# Update project
updated_project = await prisma_client.db.litellm_projecttable.update(
updated_project: prisma_models.LiteLLM_ProjectTable | None = await prisma_client.db.litellm_projecttable.update(
where={"project_id": data.project_id},
data=update_data,
include={"litellm_budget_table": True, "object_permission": True},
@ -718,7 +686,7 @@ async def update_project(
"/project/delete",
tags=["project management"],
dependencies=[Depends(user_api_key_auth)],
response_model=List[LiteLLM_ProjectTable],
response_model=list[LiteLLM_ProjectTable],
)
@management_endpoint_wrapper
async def delete_project(
@ -749,8 +717,7 @@ async def delete_project(
raise HTTPException(
status_code=403,
detail={
"error": "Project management is an enterprise feature. "
+ CommonProxyErrors.not_premium_user.value
"error": "Project management is an enterprise feature. " + CommonProxyErrors.not_premium_user.value
},
)
@ -778,9 +745,7 @@ async def delete_project(
for project_id in data.project_ids:
# Check if project exists
existing_project = await prisma_client.db.litellm_projecttable.find_unique(
where={"project_id": project_id}
)
existing_project = await prisma_client.db.litellm_projecttable.find_unique(where={"project_id": project_id})
if existing_project is None:
raise ProxyException(
@ -791,11 +756,9 @@ async def delete_project(
)
# Check if there are any keys associated with this project
associated_keys = (
await prisma_client.db.litellm_verificationtoken.find_many(
where={"project_id": project_id}
)
)
associated_keys: Sequence[
prisma_models.LiteLLM_VerificationToken
] = await prisma_client.db.litellm_verificationtoken.find_many(where={"project_id": project_id})
if len(associated_keys) > 0:
raise ProxyException(
@ -806,9 +769,9 @@ async def delete_project(
)
# Delete the project
deleted_project = await prisma_client.db.litellm_projecttable.delete(
where={"project_id": project_id}
)
deleted_project: (
prisma_models.LiteLLM_ProjectTable | None
) = await prisma_client.db.litellm_projecttable.delete(where={"project_id": project_id})
deleted_projects.append(deleted_project)
@ -854,7 +817,7 @@ async def project_info(
)
# Fetch project
project = await prisma_client.db.litellm_projecttable.find_unique(
project: prisma_models.LiteLLM_ProjectTable | None = await prisma_client.db.litellm_projecttable.find_unique(
where={"project_id": project_id},
include={"litellm_budget_table": True, "object_permission": True},
)
@ -872,17 +835,11 @@ async def project_info(
is_team_member = False
if project.team_id and user_api_key_dict.user_id:
team = await prisma_client.db.litellm_teamtable.find_unique(
where={"team_id": project.team_id}
)
team = await _team_table(prisma_client).find_unique(where={"team_id": project.team_id})
if team:
caller_user_id = user_api_key_dict.user_id
for m in team.members_with_roles or []:
m_user_id = (
m.get("user_id")
if isinstance(m, dict)
else getattr(m, "user_id", None)
)
m_user_id = m.get("user_id") if isinstance(m, dict) else getattr(m, "user_id", None)
if m_user_id == caller_user_id:
is_team_member = True
break
@ -896,9 +853,7 @@ async def project_info(
return project
except Exception as e:
verbose_proxy_logger.exception(
"litellm.proxy.management_endpoints.project_endpoints.project_info(): Exception occured - {}".format(
str(e)
)
"litellm.proxy.management_endpoints.project_endpoints.project_info(): Exception occured - {}".format(str(e))
)
raise handle_exception_on_proxy(e)
@ -907,7 +862,7 @@ async def project_info(
"/project/list",
tags=["project management"],
dependencies=[Depends(user_api_key_auth)],
response_model=List[LiteLLM_ProjectTable],
response_model=list[LiteLLM_ProjectTable],
)
async def list_projects(
user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth),
@ -932,21 +887,19 @@ async def list_projects(
# If proxy admin, get all projects
if user_api_key_dict.user_role == LitellmUserRoles.PROXY_ADMIN:
projects = await prisma_client.db.litellm_projecttable.find_many(
projects: Sequence[
prisma_models.LiteLLM_ProjectTable
] = await prisma_client.db.litellm_projecttable.find_many(
include={"litellm_budget_table": True, "object_permission": True}
)
else:
# Look up the user's team memberships via the reverse-index on
# LiteLLM_UserTable.teams (maintained by team_member_add alongside
# members_with_roles). This avoids a full scan of all team rows.
user_record = await prisma_client.db.litellm_usertable.find_unique(
user_record: prisma_models.LiteLLM_UserTable | None = await prisma_client.db.litellm_usertable.find_unique(
where={"user_id": user_api_key_dict.user_id},
)
user_team_ids = (
user_record.teams
if user_record is not None and user_record.teams
else []
)
user_team_ids: Sequence[str] = user_record.teams if user_record is not None and user_record.teams else []
projects = await prisma_client.db.litellm_projecttable.find_many(
where={"team_id": {"in": user_team_ids}},

4
enterprise/pyproject.toml
View file

@ -1,6 +1,6 @@
[project]
name = "litellm-enterprise"
version = "0.1.51"
version = "0.1.52"
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 = "0.1.51"
version = "0.1.52"
version_files = [
"pyproject.toml:^version",
"../pyproject.toml:litellm-enterprise==",

1
gateway/routes/allowlist.py
View file

@ -54,6 +54,7 @@ GATEWAY_PATH_PREFIXES: tuple[str, ...] = (
"/messages",
"/v1/skills",
"/v1/a2a/",
"/a2a/",
# LiteLLM-native LLM surface
"/v1/rerank",
"/v2/rerank",

2
helm/litellm/templates/ingress.yaml
View file

@ -19,7 +19,7 @@
"/v1/fine-tuning" "/fine-tuning" "/v1/responses" "/responses" "/v1/threads" "/threads"
"/v1/assistants" "/assistants" "/v1/vector_stores" "/vector_stores" "/v1/indexes"
"/v1/models" "/models" "/openai" "/engines"
"/v1/messages" "/messages" "/v1/skills" "/v1/a2a"
"/v1/messages" "/messages" "/v1/skills" "/v1/a2a" "/a2a"
"/v1/rerank" "/v2/rerank" "/rerank" "/v1/ocr" "/ocr" "/v1/rag" "/rag"
"/v1/video" "/v1/videos" "/video" "/videos" "/v1/search" "/search"
"/v1/containers" "/containers" "/v1/evals" "/v1/memory" "/queue/chat"

2
litellm-proxy-extras/litellm_proxy_extras/migrations/20260724000000_add_spend_log_tool_index_start_time_idx/migration.sql Normal file
View file

@ -0,0 +1,2 @@
-- CreateIndex
CREATE INDEX IF NOT EXISTS "LiteLLM_SpendLogToolIndex_start_time_idx" ON "LiteLLM_SpendLogToolIndex"("start_time");

12
litellm-proxy-extras/litellm_proxy_extras/migrations/20260725000000_add_daily_tool_spend/migration.sql Normal file
View file

@ -0,0 +1,12 @@
-- CreateTable
CREATE TABLE IF NOT EXISTS "LiteLLM_DailyToolSpend" (
"date" TEXT NOT NULL,
"tool_name" TEXT NOT NULL,
"spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0,
"total_tokens" BIGINT NOT NULL DEFAULT 0,
"request_count" BIGINT NOT NULL DEFAULT 0,
"created_at" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
"updated_at" TIMESTAMP(3) NOT NULL,
CONSTRAINT "LiteLLM_DailyToolSpend_pkey" PRIMARY KEY ("date","tool_name")
);

14
litellm-proxy-extras/litellm_proxy_extras/schema.prisma
View file

@ -1094,6 +1094,20 @@ model LiteLLM_SpendLogToolIndex {
@@id([request_id, tool_name])
@@index([tool_name, start_time])
@@index([start_time])
}
// Daily tool spend rollup (one row per tool per day) – the Cost Optimization card reads this, never SpendLogs
model LiteLLM_DailyToolSpend {
date String
tool_name String
spend Float @default(0.0)
total_tokens BigInt @default(0)
request_count BigInt @default(0)
created_at DateTime @default(now())
updated_at DateTime @updatedAt
@@id([date, tool_name])
}
// Prompt table for storing prompt configurations

4
litellm-proxy-extras/pyproject.toml
View file

@ -1,6 +1,6 @@
[project]
name = "litellm-proxy-extras"
version = "0.4.80"
version = "0.4.81"
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 = "0.4.80"
version = "0.4.81"
version_files = [
"pyproject.toml:^version",
"../pyproject.toml:litellm-proxy-extras==",

39
litellm/completion_extras/litellm_responses_transformation/handler.py
View file

@ -209,7 +209,15 @@ class ResponsesToCompletionBridgeHandler:
json_mode=kwargs.get("json_mode"),
)
elif isinstance(result, ModelResponse):
return result
if not stream:
return result
return self._completed_response_as_stream(
response=result,
model=model,
custom_llm_provider=custom_llm_provider,
logging_obj=logging_obj,
json_mode=kwargs.get("json_mode"),
)
elif not stream:
responses_api_response = self._collect_response_from_stream(result)
return self.transformation_handler.transform_response(
@ -299,7 +307,15 @@ class ResponsesToCompletionBridgeHandler:
json_mode=kwargs.get("json_mode"),
)
elif isinstance(result, ModelResponse):
return result
if not stream:
return result
return self._completed_response_as_stream(
response=result,
model=model,
custom_llm_provider=custom_llm_provider,
logging_obj=logging_obj,
json_mode=kwargs.get("json_mode"),
)
elif not stream:
responses_api_response = await self._collect_response_from_stream_async(result)
return self.transformation_handler.transform_response(
@ -331,6 +347,25 @@ class ResponsesToCompletionBridgeHandler:
)
return self._apply_post_stream_processing(streamwrapper, model, custom_llm_provider)
def _completed_response_as_stream(
self,
response: "ModelResponse",
model: str,
custom_llm_provider: str,
logging_obj: "LiteLLMLoggingObj",
json_mode: bool | None,
) -> "CustomStreamWrapper":
from litellm.litellm_core_utils.streaming_handler import CustomStreamWrapper
from litellm.llms.base_llm.base_model_iterator import MockResponseIterator
streamwrapper = CustomStreamWrapper(
completion_stream=MockResponseIterator(model_response=response, json_mode=json_mode),
model=model,
custom_llm_provider=custom_llm_provider,
logging_obj=logging_obj,
)
return self._apply_post_stream_processing(streamwrapper, model, custom_llm_provider)
@staticmethod
def _apply_post_stream_processing(
stream: "CustomStreamWrapper",

12
litellm/completion_extras/litellm_responses_transformation/transformation.py
View file

@ -1077,6 +1077,7 @@ class LiteLLMResponsesTransformationHandler(CompletionTransformationBridge):
class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator):
def __init__(self, streaming_response, sync_stream: bool, json_mode: Optional[bool] = False):
super().__init__(streaming_response, sync_stream, json_mode)
self._chat_completion_id: str | None = None
def _handle_string_chunk(
self, str_line: Union[str, "BaseModel"]
@ -1384,4 +1385,13 @@ class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator):
ModelResponseStream: OpenAI-formatted streaming chunk
"""
verbose_logger.debug(f"Chat provider: transform_streaming_response called with chunk: {chunk}")
return OpenAiResponsesToChatCompletionStreamIterator.translate_responses_chunk_to_openai_stream(chunk)
return self._with_stream_scoped_id(
OpenAiResponsesToChatCompletionStreamIterator.translate_responses_chunk_to_openai_stream(chunk)
)
def _with_stream_scoped_id(self, chunk: "ModelResponseStream") -> "ModelResponseStream":
if self._chat_completion_id is None:
self._chat_completion_id = chunk.id
else:
chunk.id = self._chat_completion_id
return chunk

3
litellm/constants.py
View file

@ -1420,6 +1420,8 @@ LITELLM_EXPIRED_UI_SESSION_KEY_CLEANUP_BATCH_SIZE = int(
os.getenv("LITELLM_EXPIRED_UI_SESSION_KEY_CLEANUP_BATCH_SIZE", 1000)
)
LITELLM_PROXY_ADMIN_NAME = "default_user_id"
LITELLM_PROXY_BUDGET_NAME = "litellm-proxy-budget"
GLOBAL_PROXY_SPEND_CACHE_KEY = f"{LITELLM_PROXY_ADMIN_NAME}:spend"
########################### CLI SSO AUTHENTICATION CONSTANTS ###########################
LITELLM_CLI_SOURCE_IDENTIFIER = "litellm-cli"
@ -1457,6 +1459,7 @@ SPEND_LOG_CLEANUP_MAX_CONSECUTIVE_BATCH_FAILURES = int(os.getenv("SPEND_LOG_CLEA
SPEND_LOG_CLEANUP_BATCH_FAILURE_BACKOFF_SECONDS = float(
os.getenv("SPEND_LOG_CLEANUP_BATCH_FAILURE_BACKOFF_SECONDS", 0.5)
)
TOOL_SPEND_TOP_TOOLS = 100
SPEND_LOG_PARTITION_INTERVAL = os.getenv("SPEND_LOG_PARTITION_INTERVAL", "day")
SPEND_LOG_PARTITION_PRECREATE_AHEAD = int(os.getenv("SPEND_LOG_PARTITION_PRECREATE_AHEAD", 7))
SPEND_LOG_QUEUE_SIZE_THRESHOLD = int(os.getenv("SPEND_LOG_QUEUE_SIZE_THRESHOLD", 100))

17
litellm/integrations/custom_guardrail.py
View file

@ -69,6 +69,8 @@ from litellm.exceptions import (
# proxy's metadata sanitizer.
_PRE_CALL_EXECUTED_TOKEN = secrets.token_hex(16)
_GUARDRAIL_BLOCK_STATUS_CODES = frozenset({400, 403, 422})
_guardrail_self_recorded: contextvars.ContextVar[bool] = contextvars.ContextVar(
"litellm_guardrail_self_recorded", default=False
)
@ -1055,8 +1057,15 @@ class CustomGuardrail(CustomLogger):
- GuardrailRaisedException (generic guardrail API, tool permission)
- BlockedPiiEntityError (Presidio PII detection)
- SensitiveDataRouteException (sensitive-data reroute to on-premise model)
- HTTPException with status 400 (content policy violation)
- HTTPException with a block-signalling status (400, 403, 422)
- ModifyResponseException (passthrough mode violation)
Only the statuses guardrails use in-tree to signal a deliberate rejection
count as an intervention: 400 (content policy), 403 (e.g. akto) and 422
(e.g. llm_as_a_judge). Other 4xx codes are commonly propagated from an
upstream guardrail provider response (401 bad key, 408 timeout, 429 rate
limit, or a raw upstream status), which are technical failures, not
blocks, so they stay guardrail_failed_to_respond.
"""
if isinstance(e, ModifyResponseException):
return True
@ -1069,7 +1078,11 @@ class CustomGuardrail(CustomLogger):
),
):
return True
if HTTPException is not None and isinstance(e, HTTPException) and e.status_code == 400:
if (
HTTPException is not None
and isinstance(e, HTTPException)
and e.status_code in _GUARDRAIL_BLOCK_STATUS_CODES
):
return True
return False

20
litellm/integrations/langsmith.py
View file

@ -133,6 +133,15 @@ class LangsmithLogger(CustomBatchLogger):
"dotted_order": metadata.get("dotted_order", None),
}
def _redact_metadata(self, metadata: dict) -> dict:
# helper is shallow; also scrub nested requester_metadata since
# LangSmith forwards the whole dict into the run
redacted = redact_user_api_key_info(metadata=dict(metadata))
nested = redacted.get("requester_metadata")
if isinstance(nested, dict):
redacted["requester_metadata"] = redact_user_api_key_info(metadata=nested)
return redacted
def _build_extra_metadata(self, metadata: Dict):
extra_metadata = dict(metadata)
requester_metadata = extra_metadata.get("requester_metadata")
@ -141,13 +150,7 @@ class LangsmithLogger(CustomBatchLogger):
if key in requester_metadata and key not in extra_metadata:
extra_metadata[key] = requester_metadata[key]
# helper is shallow; also scrub nested requester_metadata since
# LangSmith forwards the whole dict into `extra`
extra_metadata = redact_user_api_key_info(metadata=extra_metadata)
nested = extra_metadata.get("requester_metadata")
if isinstance(nested, dict):
extra_metadata["requester_metadata"] = redact_user_api_key_info(metadata=nested)
return extra_metadata
return self._redact_metadata(extra_metadata)
def _build_outputs_with_usage(self, payload: StandardLoggingPayload) -> Dict[str, Any]:
response = payload["response"]
@ -200,12 +203,13 @@ class LangsmithLogger(CustomBatchLogger):
metadata = payload["metadata"]
extra_metadata = self._build_extra_metadata(dict(metadata))
inputs = {**payload, "metadata": self._redact_metadata(dict(metadata))}
outputs = self._build_outputs_with_usage(payload)
data = {
"name": fields["run_name"],
"run_type": "llm",
"inputs": payload,
"inputs": inputs,
"outputs": outputs,
"session_name": fields["project_name"],
"start_time": payload["startTime"],

11
litellm/integrations/otel/logger.py
View file

@ -17,6 +17,7 @@ from litellm.integrations.otel.model.baggage import promoted_baggage
from litellm.integrations.otel.model.config import OpenTelemetryV2Config
from litellm.integrations.otel.plumbing.context import (
is_recordable_span,
mcp_message_transport_span,
request_root_span,
resolve_mcp_span_context,
resolve_parent_context,
@ -641,8 +642,14 @@ class OpenTelemetryV2(CustomLogger):
endpoint, auth failure), so the failed request carries the same error keys
a failed LLM call does. v1's ``OpenTelemetry`` implemented this same hook;
v2 lost it when it stopped subclassing ``OpenTelemetry``, which is the
LIT-4179 regression for pre-call failures."""
span = request_root_span() or user_api_key_dict.parent_otel_span
LIT-4179 regression for pre-call failures.
An MCP message is handled on the session's task, where the request-root
anchor is whatever request opened the session, so prefer the transport the
gateway published for this specific message. Without that, a failed tool
call aimed its error at the ``initialize`` request's finished span and the
SDK dropped it, leaving the POST that actually failed unmarked."""
span = mcp_message_transport_span() or request_root_span() or user_api_key_dict.parent_otel_span
if span is None or not is_recordable_span(span):
return None
stamp_error(span, _span_error_from_exception(original_exception, traceback_str=traceback_str))

78
litellm/integrations/otel/plumbing/context.py
View file

@ -79,47 +79,67 @@ def reset_mcp_message_trace_carrier(token: "Token[Mapping[str, str] | None]") ->
_mcp_message_trace_carrier.reset(token)
# The transport span of the HTTP request carrying the CURRENT MCP message, as a
# plain ``SpanContext`` so it can cross a task boundary.
# The transport span of the HTTP request carrying the CURRENT MCP message.
#
# ``_request_root_span`` above cannot be used for MCP: a *stateful* streamable-HTTP
# session runs every message on the single task spawned by that session's
# ``initialize`` POST, so the ContextVar the ASGI request task writes at auth time
# is frozen at ``initialize`` there and never sees the later ``tools/call`` POSTs.
# Reading it from the message handler would parent every tool call in the session
# to the first request's (already ended) server span. The gateway instead resolves
# the current message's transport span on the request task and hands it over the
# same way it hands over per-request auth, and the handler publishes it here for
# the span emitter to pick up.
_mcp_message_transport_span_context: "ContextVar[SpanContext | None]" = ContextVar(
"litellm_otel_mcp_message_transport_span_context", default=None
# Reading it from the message handler parents every tool call in the session to the
# first request's server span and aims that call's ``error.*`` at it — a span that
# ended long ago, so the SDK drops the write and the failure reaches no request at
# all. The gateway instead resolves the current message's transport span on the
# request task and hands it over the same way it hands over per-request auth, and
# the handler publishes it here for the span emitter and the failure hook.
_mcp_message_transport_span: "ContextVar[Span | None]" = ContextVar(
"litellm_otel_mcp_message_transport_span", default=None
)
def set_mcp_message_transport_span_context(
span_context: "SpanContext | None",
) -> "Token[SpanContext | None]":
def set_mcp_message_transport_span(span: object) -> "Token[Span | None]":
"""Publish the transport span of the request carrying the current MCP message.
Also re-anchors the request root, so everything else the message emits or stamps
— the identity attributes seeded onto the server span, a guardrail span, a
proxy-level failure — lands on this request instead of on the one that opened
the session. The MCP SDK dispatches each message on its own task, so the anchor
is scoped to this message; the handler re-publishes it for the next one either
way. Only a transport still open for writes is anchored: replacing the anchor
with a request that already answered would just move the dropped writes from one
finished span to another.
Takes ``object`` because the gateway reads it back out of the ASGI scope, whose
values are untyped; anything that is not a usable span is stored as ``None``
rather than trusted.
Returns the reset token; the caller must reset it once the message is handled
so the transport never leaks to the next message on the same session task.
"""
return _mcp_message_transport_span_context.set(span_context)
transport = span if isinstance(span, Span) and is_recordable_span(span) else None
if transport is not None and transport.is_recording():
set_request_root_span(transport)
return _mcp_message_transport_span.set(transport)
def reset_mcp_message_transport_span_context(token: "Token[SpanContext | None]") -> None:
_mcp_message_transport_span_context.reset(token)
def reset_mcp_message_transport_span(token: "Token[Span | None]") -> None:
_mcp_message_transport_span.reset(token)
def request_root_span_context() -> "SpanContext | None":
"""The anchored request root span's context, safe to hand to another task.
def mcp_message_transport_span() -> "Span | None":
"""The published transport span, only while it is still open for writes.
A ``SpanContext`` is an immutable value, unlike the live ``Span``, so passing it
across the MCP session-task boundary cannot keep a finished span alive or invite
writes to it from the wrong request.
Recording — not merely valid — is the bar here because this span is the target
of ``error.*`` stamping from another task, and the publisher's validity check
cannot speak for a span that has since ended. A finished span keeps a valid
context forever, so it would otherwise be handed back for a write the SDK then
refuses. The POST carrying a ``tools/call`` stays open until the result is
written, so it is recording for the life of the call; a notification POST can
answer first, and this returns ``None`` for it rather than writing into the void.
"""
span = request_root_span()
return span.get_span_context() if span is not None else None
span = _mcp_message_transport_span.get()
if span is None or not span.is_recording():
return None
return span
def _mcp_transport_span_context() -> "SpanContext | None":
@ -127,12 +147,16 @@ def _mcp_transport_span_context() -> "SpanContext | None":
Prefers the transport the gateway published for this specific message; falls
back to the ambient request anchor for paths that emit an MCP span on the
request task itself (the REST MCP endpoints, the SDK).
request task itself (the REST MCP endpoints, the SDK). Parenting and linking
only need the immutable context, and unlike ``mcp_message_transport_span`` they
stay correct against a transport that has already finished, so this does not
require the span to still be recording.
"""
published = _mcp_message_transport_span_context.get()
if published is not None and published.is_valid:
return published
return request_root_span_context()
published = _mcp_message_transport_span.get()
if published is not None:
return published.get_span_context()
span = request_root_span()
return span.get_span_context() if span is not None else None
def set_request_baggage(values: Mapping[str, str], context: Context | None = None) -> Context:

62
litellm/integrations/prometheus.py
View file

@ -16,6 +16,7 @@ from typing import (
Dict,
List,
Literal,
Mapping,
Optional,
Sequence,
Tuple,
@ -1449,6 +1450,8 @@ class PrometheusLogger(CustomLogger):
prompt_details = usage_object.get("prompt_tokens_details") or {}
completion_details = usage_object.get("completion_tokens_details") or {}
cache_creation_detail_tokens = PrometheusLogger._resolve_cache_write_tokens(prompt_details)
detail_metrics: List[Tuple[Any, DEFINED_PROMETHEUS_METRICS, Any]] = [
(
self.litellm_input_cached_tokens_metric,
@ -1458,7 +1461,7 @@ class PrometheusLogger(CustomLogger):
(
self.litellm_input_cache_creation_tokens_metric,
"litellm_input_cache_creation_tokens_metric",
(prompt_details.get("cache_creation_tokens") if isinstance(prompt_details, dict) else None),
cache_creation_detail_tokens,
),
(
self.litellm_input_audio_tokens_metric,
@ -1597,27 +1600,12 @@ class PrometheusLogger(CustomLogger):
)
# Provider prompt caching metrics are independent of LiteLLM cache_hit.
provider_cache_read_tokens = 0
provider_cache_creation_tokens = 0
usage_obj = (standard_logging_payload.get("metadata", {}) or {}).get("usage_object")
if isinstance(usage_obj, dict):
# Prefer explicit provider cache fields when available.
_read = usage_obj.get("cache_read_input_tokens")
_write = usage_obj.get("cache_creation_input_tokens")
if isinstance(_read, int):
provider_cache_read_tokens = _read
if isinstance(_write, int):
provider_cache_creation_tokens = _write
# Fallback to prompt_tokens_details.cached_tokens (common normalization point).
# Only fallback when the explicit field is genuinely absent (None).
if _read is None:
prompt_details = usage_obj.get("prompt_tokens_details")
if isinstance(prompt_details, dict):
cached_tokens = prompt_details.get("cached_tokens")
if isinstance(cached_tokens, int):
provider_cache_read_tokens = cached_tokens
(
provider_cache_read_tokens,
provider_cache_creation_tokens,
) = PrometheusLogger._resolve_provider_cache_tokens(usage_obj)
if provider_cache_read_tokens > 0:
PrometheusLogger._inc_labeled_counter(
@ -1639,6 +1627,40 @@ class PrometheusLogger(CustomLogger):
amount=float(provider_cache_creation_tokens),
)
@staticmethod
def _resolve_provider_cache_tokens(usage_obj: Mapping[str, object]) -> tuple[int, int]:
# Prefer explicit provider cache fields when available.
_read = usage_obj.get("cache_read_input_tokens")
_write = usage_obj.get("cache_creation_input_tokens")
provider_cache_read_tokens = _read if isinstance(_read, int) else 0
provider_cache_creation_tokens = _write if isinstance(_write, int) else 0
# Fallback to prompt_tokens_details (common normalization point).
# Only fallback when the explicit field is genuinely absent (None).
prompt_details = usage_obj.get("prompt_tokens_details")
if _read is None and isinstance(prompt_details, dict):
cached_tokens = prompt_details.get("cached_tokens")
if isinstance(cached_tokens, int):
provider_cache_read_tokens = cached_tokens
if _write is None:
write_tokens = PrometheusLogger._resolve_cache_write_tokens(prompt_details)
if write_tokens is not None:
provider_cache_creation_tokens = write_tokens
return provider_cache_read_tokens, provider_cache_creation_tokens
@staticmethod
def _resolve_cache_write_tokens(prompt_details: object) -> int | None:
if not isinstance(prompt_details, dict):
return None
for key in ("cache_write_tokens", "cache_creation_tokens"):
value = prompt_details.get(key)
if isinstance(value, int) and not isinstance(value, bool):
return value
return None
def _increment_mcp_tool_call_metrics(
self,
standard_logging_payload: StandardLoggingPayload,

21
litellm/litellm_core_utils/litellm_logging.py
View file

@ -1878,7 +1878,14 @@ class Logging(LiteLLMLoggingBaseClass):
elif standard_logging_object is not None:
self.model_call_details["standard_logging_object"] = standard_logging_object
else:
self.model_call_details["response_cost"] = None
# Streaming reaches here before its cost is known, so the cost
# is seeded to None, but only when nothing has already
# established one. A stream that assembles into a response
# object recomputes the cost right after this; a pass-through
# stream cannot (its body is opaque) and carries the cost its
# upstream reported in the response headers, which an
# unconditional reset would discard.
self.model_call_details.setdefault("response_cost", None)
result = self._transform_usage_objects(result=result)
@ -5547,18 +5554,6 @@ def scrub_sensitive_keys_in_metadata(litellm_params: Optional[dict]):
litellm_params["_langfuse_masking_function"] = masking_fn
litellm_params["metadata"] = metadata
## check user_api_key_metadata for sensitive logging keys
cleaned_user_api_key_metadata = {}
if "user_api_key_metadata" in metadata and isinstance(metadata["user_api_key_metadata"], dict):
for k, v in metadata["user_api_key_metadata"].items():
if k == "logging": # prevent logging user logging keys
cleaned_user_api_key_metadata[k] = "scrubbed_by_litellm_for_sensitive_keys"
else:
cleaned_user_api_key_metadata[k] = v
metadata["user_api_key_metadata"] = cleaned_user_api_key_metadata
litellm_params["metadata"] = metadata
return litellm_params

32
litellm/litellm_core_utils/prompt_templates/factory.py
View file

@ -6,6 +6,7 @@ import mimetypes
import re
import xml.etree.ElementTree as ET
from enum import Enum
from collections.abc import Mapping
from typing import Any, Dict, List, Optional, Set, Tuple, TypedDict, Union, cast, overload
from jinja2.sandbox import ImmutableSandboxedEnvironment
@ -5350,7 +5351,9 @@ def prompt_factory(
def get_attribute_or_key(tool_or_function, attribute, default=None):
if hasattr(tool_or_function, attribute):
return getattr(tool_or_function, attribute)
return tool_or_function.get(attribute, default)
if isinstance(tool_or_function, Mapping):
return tool_or_function.get(attribute, default)
return default
class NormalizedToolCall(TypedDict):
@ -5379,14 +5382,18 @@ def _parse_tool_call_arguments(raw: Any, tool_name: Optional[str], context: str)
return parsed if isinstance(parsed, dict) else {}
def _tool_calls_from_chat_completion_response(response: Any) -> list[NormalizedToolCall]:
def _tool_calls_from_chat_completion_response(
response: Any, include_all_choices: bool = False
) -> list[NormalizedToolCall]:
choices = get_attribute_or_key(response, "choices", None)
if not (isinstance(choices, list) and choices):
return []
message = get_attribute_or_key(choices[0], "message", None)
tool_calls = get_attribute_or_key(message, "tool_calls", None) if message else None
if not isinstance(tool_calls, list):
return []
tool_calls: list[Any] = []
for choice in choices if include_all_choices else choices[:1]:
message = get_attribute_or_key(choice, "message", None)
choice_tool_calls = get_attribute_or_key(message, "tool_calls", None) if message else None
if isinstance(choice_tool_calls, list):
tool_calls.extend(choice_tool_calls)
result: list[NormalizedToolCall] = []
for tc in tool_calls:
fn = get_attribute_or_key(tc, "function", None)
@ -5449,7 +5456,7 @@ def _tool_calls_from_anthropic_messages_response(response: Any) -> list[Normaliz
return result
def get_tool_calls_from_response(response: Any) -> list[NormalizedToolCall]:
def get_tool_calls_from_response(response: Any, include_all_choices: bool = False) -> list[NormalizedToolCall]:
"""
Extract tool/function calls from a response object into a normalized
``{"id", "name", "arguments"}`` shape, regardless of which API surface
@ -5457,11 +5464,20 @@ def get_tool_calls_from_response(response: Any) -> list[NormalizedToolCall]:
the Responses API (``output`` items of type ``function_call``), or the
Anthropic Messages API (``content`` blocks of type ``tool_use``).
``include_all_choices`` decides the chat-completions scope: the default
reads only ``choices[0]``, which is what consumers that act on THE reply
(e.g. guardrails rebuilding the primary assistant message) want; usage
accounting passes True because every choice of an ``n>1`` request costs
money and its tool calls really ran. The other surfaces have a single
output, so the flag has no effect on them.
Callers that only care about a specific tool should filter the result by
``name`` themselves -- this returns every tool call found.
"""
chat_tool_calls = _tool_calls_from_chat_completion_response(response, include_all_choices=include_all_choices)
if chat_tool_calls:
return chat_tool_calls
for extractor in (
_tool_calls_from_chat_completion_response,
_tool_calls_from_responses_api_response,
_tool_calls_from_anthropic_messages_response,
):

39
litellm/llms/anthropic/experimental_pass_through/messages/interceptors/advisor.py
View file

@ -21,6 +21,7 @@ import litellm
import litellm.constants as _c
from litellm.litellm_core_utils.url_utils import validate_url
from litellm.llms.anthropic.common_utils import strip_advisor_blocks_from_messages
from litellm.router_utils.cooldown_handlers import mark_advisor_orchestration_failure
from litellm.types.llms.anthropic_messages.anthropic_response import (
AnthropicMessagesResponse,
)
@ -124,30 +125,36 @@ class AdvisorOrchestrationHandler(MessagesInterceptor):
iteration += 1
if iteration > max_uses:
raise AdvisorMaxIterationsError(
max_iterations_error = AdvisorMaxIterationsError(
f"Advisor orchestration loop exceeded max_uses={max_uses}. "
"Increase max_uses in the advisor tool definition or cap the request."
)
mark_advisor_orchestration_failure(max_iterations_error)
raise max_iterations_error
# --- Build advisor context ---
advisor_messages = _build_advisor_context(current_messages, executor_response, advisor_use_block)
# --- Advisor sub-call (always non-streaming, no tools) ---
advisor_response: AnthropicMessagesResponse = await _call_messages_handler(
model=advisor_model,
messages=advisor_messages,
tools=None,
stream=False,
max_tokens=max_tokens,
custom_llm_provider=None, # let litellm resolve from model name
metadata={
**metadata_base,
"advisor_sub_call": True,
"parent_request_id": parent_request_id,
},
api_key=advisor_api_key,
api_base=advisor_api_base,
)
try:
advisor_response: AnthropicMessagesResponse = await _call_messages_handler(
model=advisor_model,
messages=advisor_messages,
tools=None,
stream=False,
max_tokens=max_tokens,
custom_llm_provider=None, # let litellm resolve from model name
metadata={
**metadata_base,
"advisor_sub_call": True,
"parent_request_id": parent_request_id,
},
api_key=advisor_api_key,
api_base=advisor_api_base,
)
except Exception as advisor_sub_call_exception:
mark_advisor_orchestration_failure(advisor_sub_call_exception)
raise
advisor_text = _extract_response_text(advisor_response)

157
litellm/llms/azure/batches/handler.py
View file

@ -2,7 +2,8 @@
Azure Batches API Handler
"""
from typing import Any, Coroutine, Optional, Union, cast
from collections.abc import Coroutine
from typing import cast
import httpx
from openai import AsyncOpenAI, OpenAI
@ -33,32 +34,30 @@ class AzureBatchesAPI(BaseAzureLLM):
async def acreate_batch(
self,
create_batch_data: CreateBatchRequest,
azure_client: Union[AsyncAzureOpenAI, AsyncOpenAI],
azure_client: AsyncAzureOpenAI | AsyncOpenAI,
) -> LiteLLMBatch:
response = await azure_client.batches.create(**create_batch_data) # type: ignore[arg-type]
return LiteLLMBatch(**response.model_dump())
return LiteLLMBatch.model_validate(response.model_dump())
def create_batch(
self,
_is_async: bool,
create_batch_data: CreateBatchRequest,
api_key: Optional[str],
api_base: Optional[str],
api_version: Optional[str],
timeout: Union[float, httpx.Timeout],
max_retries: Optional[int],
client: Optional[Union[AzureOpenAI, AsyncAzureOpenAI, OpenAI, AsyncOpenAI]] = None,
litellm_params: Optional[dict] = None,
) -> Union[LiteLLMBatch, Coroutine[Any, Any, LiteLLMBatch]]:
azure_client: Optional[Union[AzureOpenAI, AsyncAzureOpenAI, OpenAI, AsyncOpenAI]] = (
self.get_azure_openai_client(
api_key=api_key,
api_base=api_base,
api_version=api_version,
client=client,
_is_async=_is_async,
litellm_params=litellm_params or {},
)
api_key: str | None,
api_base: str | None,
api_version: str | None,
timeout: float | httpx.Timeout,
max_retries: int | None,
client: AzureOpenAI | AsyncAzureOpenAI | OpenAI | AsyncOpenAI | None = None,
litellm_params: dict | None = None,
) -> LiteLLMBatch | Coroutine[object, object, LiteLLMBatch]:
azure_client: AzureOpenAI | AsyncAzureOpenAI | OpenAI | AsyncOpenAI | None = self.get_azure_openai_client(
api_key=api_key,
api_base=api_base,
api_version=api_version,
client=client,
_is_async=_is_async,
litellm_params=litellm_params or {},
)
if azure_client is None:
raise ValueError(
@ -73,38 +72,36 @@ class AzureBatchesAPI(BaseAzureLLM):
return self.acreate_batch( # type: ignore
create_batch_data=create_batch_data, azure_client=azure_client
)
response = cast(Union[AzureOpenAI, OpenAI], azure_client).batches.create(**create_batch_data) # type: ignore[arg-type]
return LiteLLMBatch(**response.model_dump())
response = cast(AzureOpenAI | OpenAI, azure_client).batches.create(**create_batch_data) # type: ignore[arg-type]
return LiteLLMBatch.model_validate(response.model_dump())
async def aretrieve_batch(
self,
retrieve_batch_data: RetrieveBatchRequest,
client: Union[AsyncAzureOpenAI, AsyncOpenAI],
client: AsyncAzureOpenAI | AsyncOpenAI,
) -> LiteLLMBatch:
response = await client.batches.retrieve(**retrieve_batch_data) # type: ignore[arg-type]
return LiteLLMBatch(**response.model_dump())
return LiteLLMBatch.model_validate(response.model_dump())
def retrieve_batch(
self,
_is_async: bool,
retrieve_batch_data: RetrieveBatchRequest,
api_key: Optional[str],
api_base: Optional[str],
api_version: Optional[str],
timeout: Union[float, httpx.Timeout],
max_retries: Optional[int],
client: Optional[Union[AzureOpenAI, AsyncAzureOpenAI, OpenAI, AsyncOpenAI]] = None,
litellm_params: Optional[dict] = None,
api_key: str | None,
api_base: str | None,
api_version: str | None,
timeout: float | httpx.Timeout,
max_retries: int | None,
client: AzureOpenAI | AsyncAzureOpenAI | OpenAI | AsyncOpenAI | None = None,
litellm_params: dict | None = None,
):
azure_client: Optional[Union[AzureOpenAI, AsyncAzureOpenAI, OpenAI, AsyncOpenAI]] = (
self.get_azure_openai_client(
api_key=api_key,
api_base=api_base,
api_version=api_version,
client=client,
_is_async=_is_async,
litellm_params=litellm_params or {},
)
azure_client: AzureOpenAI | AsyncAzureOpenAI | OpenAI | AsyncOpenAI | None = self.get_azure_openai_client(
api_key=api_key,
api_base=api_base,
api_version=api_version,
client=client,
_is_async=_is_async,
litellm_params=litellm_params or {},
)
if azure_client is None:
raise ValueError(
@ -119,38 +116,36 @@ class AzureBatchesAPI(BaseAzureLLM):
return self.aretrieve_batch( # type: ignore
retrieve_batch_data=retrieve_batch_data, client=azure_client
)
response = cast(Union[AzureOpenAI, OpenAI], azure_client).batches.retrieve(**retrieve_batch_data)
return LiteLLMBatch(**response.model_dump())
response = cast(AzureOpenAI | OpenAI, azure_client).batches.retrieve(**retrieve_batch_data)
return LiteLLMBatch.model_validate(response.model_dump())
async def acancel_batch(
self,
cancel_batch_data: CancelBatchRequest,
client: Union[AsyncAzureOpenAI, AsyncOpenAI],
client: AsyncAzureOpenAI | AsyncOpenAI,
) -> LiteLLMBatch:
response = await client.batches.cancel(**cancel_batch_data)
return LiteLLMBatch(**response.model_dump())
return LiteLLMBatch.model_validate(response.model_dump())
def cancel_batch(
self,
_is_async: bool,
cancel_batch_data: CancelBatchRequest,
api_key: Optional[str],
api_base: Optional[str],
api_version: Optional[str],
timeout: Union[float, httpx.Timeout],
max_retries: Optional[int],
client: Optional[Union[AzureOpenAI, AsyncAzureOpenAI, OpenAI, AsyncOpenAI]] = None,
litellm_params: Optional[dict] = None,
api_key: str | None,
api_base: str | None,
api_version: str | None,
timeout: float | httpx.Timeout,
max_retries: int | None,
client: AzureOpenAI | AsyncAzureOpenAI | OpenAI | AsyncOpenAI | None = None,
litellm_params: dict | None = None,
):
azure_client: Optional[Union[AzureOpenAI, AsyncAzureOpenAI, OpenAI, AsyncOpenAI]] = (
self.get_azure_openai_client(
api_key=api_key,
api_base=api_base,
api_version=api_version,
client=client,
_is_async=_is_async,
litellm_params=litellm_params or {},
)
azure_client: AzureOpenAI | AsyncAzureOpenAI | OpenAI | AsyncOpenAI | None = self.get_azure_openai_client(
api_key=api_key,
api_base=api_base,
api_version=api_version,
client=client,
_is_async=_is_async,
litellm_params=litellm_params or {},
)
if azure_client is None:
raise ValueError(
@ -172,13 +167,13 @@ class AzureBatchesAPI(BaseAzureLLM):
"Azure client is not an instance of AzureOpenAI or OpenAI. Make sure you passed a sync client."
)
response = azure_client.batches.cancel(**cancel_batch_data)
return LiteLLMBatch(**response.model_dump())
return LiteLLMBatch.model_validate(response.model_dump())
async def alist_batches(
self,
client: Union[AsyncAzureOpenAI, AsyncOpenAI],
after: Optional[str] = None,
limit: Optional[int] = None,
client: AsyncAzureOpenAI | AsyncOpenAI,
after: str | None = None,
limit: int | None = None,
):
response = await client.batches.list(after=after, limit=limit) # type: ignore
return response
@ -186,25 +181,23 @@ class AzureBatchesAPI(BaseAzureLLM):
def list_batches(
self,
_is_async: bool,
api_key: Optional[str],
api_base: Optional[str],
api_version: Optional[str],
timeout: Union[float, httpx.Timeout],
max_retries: Optional[int],
after: Optional[str] = None,
limit: Optional[int] = None,
client: Optional[Union[AzureOpenAI, AsyncAzureOpenAI, OpenAI, AsyncOpenAI]] = None,
litellm_params: Optional[dict] = None,
api_key: str | None,
api_base: str | None,
api_version: str | None,
timeout: float | httpx.Timeout,
max_retries: int | None,
after: str | None = None,
limit: int | None = None,
client: AzureOpenAI | AsyncAzureOpenAI | OpenAI | AsyncOpenAI | None = None,
litellm_params: dict | None = None,
):
azure_client: Optional[Union[AzureOpenAI, AsyncAzureOpenAI, OpenAI, AsyncOpenAI]] = (
self.get_azure_openai_client(
api_key=api_key,
api_base=api_base,
api_version=api_version,
client=client,
_is_async=_is_async,
litellm_params=litellm_params or {},
)
azure_client: AzureOpenAI | AsyncAzureOpenAI | OpenAI | AsyncOpenAI | None = self.get_azure_openai_client(
api_key=api_key,
api_base=api_base,
api_version=api_version,
client=client,
_is_async=_is_async,
litellm_params=litellm_params or {},
)
if azure_client is None:
raise ValueError(

8
litellm/llms/azure_ai/anthropic/count_tokens/transformation.py
View file

@ -4,8 +4,6 @@ Azure AI Anthropic CountTokens API transformation logic.
Extends the base Anthropic CountTokens transformation with Azure authentication.
"""
from typing import Any, Dict, Optional
from litellm.constants import ANTHROPIC_TOKEN_COUNTING_BETA_VERSION
from litellm.llms.anthropic.count_tokens.transformation import (
AnthropicCountTokensConfig,
@ -25,8 +23,8 @@ class AzureAIAnthropicCountTokensConfig(AnthropicCountTokensConfig):
def get_required_headers(
self,
api_key: str,
litellm_params: Optional[Dict[str, Any]] = None,
) -> Dict[str, str]:
litellm_params: dict[str, object] | None = None,
) -> dict[str, str]:
"""
Get the required headers for the Azure AI Anthropic CountTokens API.
@ -53,7 +51,7 @@ class AzureAIAnthropicCountTokensConfig(AnthropicCountTokensConfig):
if "api_key" not in litellm_params:
litellm_params["api_key"] = api_key
litellm_params_obj = GenericLiteLLMParams(**litellm_params)
litellm_params_obj = GenericLiteLLMParams.model_validate(litellm_params)
# Get Azure auth headers (api-key or Authorization)
azure_headers = BaseAzureLLM._base_validate_azure_environment(headers={}, litellm_params=litellm_params_obj)

36
litellm/llms/bedrock/base_aws_llm.py
View file

@ -78,8 +78,10 @@ class BaseAWSLLM:
# Storage is in-process memory only: default ``DualCache()`` has no Redis backend unless attached
# elsewhere. Entry TTL: static access-key + secret + region use ``_get_default_ttl_for_boto3_credentials``
# (~59 minutes); ambient env (``_auth_with_env_vars`` returns ``ttl=None``) uses ``InMemoryCache``'s
# ``default_ttl`` (600 seconds / 10 minutes). AssumeRole, web identity, profiles, and explicit
# session-token tuples are not cached — see ``get_credentials`` and ``_get_or_set_cached_credentials``.
# ``default_ttl`` (600 seconds / 10 minutes); web identity STS credentials use
# ``_get_default_ttl_for_boto3_credentials`` (~59 minutes), keyed on all aws_* credential args
# plus ssl_verify. AssumeRole, profiles, and explicit session-token tuples are not cached — see
# ``get_credentials`` and ``_get_or_set_cached_credentials``.
_shared_iam_cache: ClassVar[DualCache] = DualCache()
def __init__(self) -> None:
@ -136,11 +138,12 @@ class BaseAWSLLM:
which ``InMemoryCache.set_cache`` resolves to ``default_ttl`` (600 seconds / 10 minutes by
default).
Used only for static access-key credentials and ambient credentials from
Used for static access-key credentials, ambient credentials from
``_auth_with_env_vars`` (including when skipping AssumeRole because the runtime identity
already matches ``aws_role_name``).
already matches ``aws_role_name``), and web identity STS credentials (plain
non-refreshable ``Credentials`` cached ~59 min, inside the 3600s STS session).
AssumeRole, web identity exchange, profiles, and explicit session-token tuples are not
AssumeRole, profiles, and explicit session-token tuples are not
cached here — shared ``Credentials`` / refresh state must not span logical sessions.
"""
cache_key = self.get_cache_key(credential_args)
@ -266,23 +269,26 @@ class BaseAWSLLM:
# Credentials - boto3.Credentials
# cache ttl - Optional[int]. If None, the credentials are not cached. Some auth flows have no expiry time.
#
# iam_cache: static keys and ambient env only (including skip-AssumeRole path).
# Do not cache AssumeRole / web identity / profile / explicit session-token paths here.
# iam_cache: static keys, ambient env (including skip-AssumeRole path), and web identity.
# Do not cache AssumeRole / profile / explicit session-token paths here.
#########################################################
if self._is_auth_with_web_identity_token(
aws_web_identity_token,
aws_role_name,
aws_session_name,
):
credentials, _cache_ttl = self._auth_with_web_identity_token(
aws_web_identity_token=cast(str, aws_web_identity_token),
aws_role_name=cast(str, aws_role_name),
aws_session_name=cast(str, aws_session_name),
aws_region_name=aws_region_name,
aws_sts_endpoint=aws_sts_endpoint,
aws_external_id=aws_external_id,
return self._get_or_set_cached_credentials(
args,
lambda: self._auth_with_web_identity_token(
aws_web_identity_token=cast(str, aws_web_identity_token),
aws_role_name=cast(str, aws_role_name),
aws_session_name=cast(str, aws_session_name),
aws_region_name=aws_region_name,
aws_sts_endpoint=aws_sts_endpoint,
aws_external_id=aws_external_id,
ssl_verify=ssl_verify,
),
)
return credentials
elif self._is_auth_with_aws_role(aws_role_name):
# Same role (IRSA/ECS/EC2): ambient creds via _get_or_set_cached_credentials like the
# default env branch; never pre-read cache (must run _is_already_running_as_role first).

23
litellm/llms/bedrock/messages/mantle_transformation.py
View file

@ -17,6 +17,10 @@ from litellm.llms.bedrock.common_utils import build_mantle_messages_url
from litellm.llms.bedrock.messages.invoke_transformations.anthropic_claude3_transformation import (
AmazonAnthropicClaudeMessagesConfig,
)
from litellm.types.llms.anthropic_messages.anthropic_response import (
AnthropicMessagesResponse,
AnthropicUsage,
)
from litellm.types.router import GenericLiteLLMParams
if TYPE_CHECKING:
@ -103,6 +107,25 @@ class AmazonMantleMessagesConfig(AmazonAnthropicClaudeMessagesConfig):
)
return {**request, "model": model_id, **stream_fields}
def transform_anthropic_messages_response(
self,
model: str,
raw_response: httpx.Response,
logging_obj: LiteLLMLoggingObj,
) -> AnthropicMessagesResponse:
response = super().transform_anthropic_messages_response(
model=model,
raw_response=raw_response,
logging_obj=logging_obj,
)
existing_usage: AnthropicUsage = response.get("usage") or AnthropicUsage()
normalized_usage: AnthropicUsage = {
"input_tokens": 0,
"output_tokens": 0,
**existing_usage,
}
return {**response, "usage": normalized_usage}
def get_async_streaming_response_iterator(
self,
model: str,

88
litellm/llms/openai/evals/transformation.py
View file

@ -2,7 +2,7 @@
OpenAI Evals API configuration and transformations
"""
from typing import Any, Dict, Optional, Tuple
from collections.abc import Mapping
import httpx
@ -31,6 +31,10 @@ from litellm.types.router import GenericLiteLLMParams
from litellm.types.utils import LlmProviders
def _parsed_response_json(raw_response: httpx.Response) -> Mapping[str, object]:
return raw_response.json()
class OpenAIEvalsConfig(BaseEvalsAPIConfig):
"""OpenAI-specific Evals API configuration"""
@ -38,7 +42,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
def custom_llm_provider(self) -> LlmProviders:
return LlmProviders.OPENAI
def validate_environment(self, headers: dict, litellm_params: Optional[GenericLiteLLMParams]) -> dict:
def validate_environment(self, headers: dict, litellm_params: GenericLiteLLMParams | None) -> dict:
"""Add OpenAI-specific headers"""
import litellm
from litellm.secret_managers.main import get_secret_str
@ -61,9 +65,9 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
def get_complete_url(
self,
api_base: Optional[str],
api_base: str | None,
endpoint: str,
eval_id: Optional[str] = None,
eval_id: str | None = None,
) -> str:
"""Get complete URL for OpenAI Evals API"""
if api_base is None:
@ -79,7 +83,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
create_request: CreateEvalRequest,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Dict:
) -> dict:
"""Transform create eval request for OpenAI"""
verbose_logger.debug("Transforming create eval request: %s", create_request)
@ -94,17 +98,17 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> Eval:
"""Transform OpenAI response to Eval object"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming create eval response: %s", response_json)
return Eval(**response_json)
return Eval.model_validate(response_json)
def transform_list_evals_request(
self,
list_params: ListEvalsParams,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict]:
) -> tuple[str, dict]:
"""Transform list evals request for OpenAI"""
api_base = "https://api.openai.com"
if litellm_params and litellm_params.api_base:
@ -113,7 +117,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
url = self.get_complete_url(api_base=api_base, endpoint="evals")
# Build query parameters
query_params: Dict[str, Any] = {}
query_params: dict[str, object] = {}
if "limit" in list_params and list_params["limit"]:
query_params["limit"] = list_params["limit"]
if "after" in list_params and list_params["after"]:
@ -138,10 +142,10 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> ListEvalsResponse:
"""Transform OpenAI response to ListEvalsResponse"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming list evals response: %s", response_json)
return ListEvalsResponse(**response_json)
return ListEvalsResponse.model_validate(response_json)
def transform_get_eval_request(
self,
@ -149,7 +153,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict]:
) -> tuple[str, dict]:
"""Transform get eval request for OpenAI"""
url = self.get_complete_url(api_base=api_base, endpoint="evals", eval_id=eval_id)
@ -163,10 +167,10 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> Eval:
"""Transform OpenAI response to Eval object"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming get eval response: %s", response_json)
return Eval(**response_json)
return Eval.model_validate(response_json)
def transform_update_eval_request(
self,
@ -175,7 +179,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict, Dict]:
) -> tuple[str, dict, dict]:
"""Transform update eval request for OpenAI"""
url = self.get_complete_url(api_base=api_base, endpoint="evals", eval_id=eval_id)
@ -192,10 +196,10 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> Eval:
"""Transform OpenAI response to Eval object"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming update eval response: %s", response_json)
return Eval(**response_json)
return Eval.model_validate(response_json)
def transform_delete_eval_request(
self,
@ -203,7 +207,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict]:
) -> tuple[str, dict]:
"""Transform delete eval request for OpenAI"""
url = self.get_complete_url(api_base=api_base, endpoint="evals", eval_id=eval_id)
@ -217,10 +221,10 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> DeleteEvalResponse:
"""Transform OpenAI response to DeleteEvalResponse"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming delete eval response: %s", response_json)
return DeleteEvalResponse(**response_json)
return DeleteEvalResponse.model_validate(response_json)
def transform_cancel_eval_request(
self,
@ -228,12 +232,12 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict, Dict]:
) -> tuple[str, dict, dict]:
"""Transform cancel eval request for OpenAI"""
url = f"{self.get_complete_url(api_base=api_base, endpoint='evals', eval_id=eval_id)}/cancel"
# Empty body for cancel request
request_body: Dict[str, Any] = {}
request_body: dict[str, object] = {}
verbose_logger.debug("Cancel eval request - URL: %s", url)
@ -245,10 +249,10 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> CancelEvalResponse:
"""Transform OpenAI response to CancelEvalResponse"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming cancel eval response: %s", response_json)
return CancelEvalResponse(**response_json)
return CancelEvalResponse.model_validate(response_json)
# Run API Transformations
def transform_create_run_request(
@ -257,7 +261,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
create_request: CreateRunRequest,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict]:
) -> tuple[str, dict]:
"""Transform create run request for OpenAI"""
api_base = "https://api.openai.com"
if litellm_params and litellm_params.api_base:
@ -279,10 +283,10 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> Run:
"""Transform OpenAI response to Run object"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming create run response: %s", response_json)
return Run(**response_json)
return Run.model_validate(response_json)
def transform_list_runs_request(
self,
@ -290,7 +294,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
list_params: ListRunsParams,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict]:
) -> tuple[str, dict]:
"""Transform list runs request for OpenAI"""
api_base = "https://api.openai.com"
if litellm_params and litellm_params.api_base:
@ -300,7 +304,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
url = f"{api_base}/v1/evals/{encoded_eval_id}/runs"
# Build query parameters
query_params: Dict[str, Any] = {}
query_params: dict[str, object] = {}
if "limit" in list_params and list_params["limit"]:
query_params["limit"] = list_params["limit"]
if "after" in list_params and list_params["after"]:
@ -323,10 +327,10 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> ListRunsResponse:
"""Transform OpenAI response to ListRunsResponse"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming list runs response: %s", response_json)
return ListRunsResponse(**response_json)
return ListRunsResponse.model_validate(response_json)
def transform_get_run_request(
self,
@ -335,7 +339,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict]:
) -> tuple[str, dict]:
"""Transform get run request for OpenAI"""
encoded_eval_id = encode_url_path_segment(eval_id, field_name="eval_id")
encoded_run_id = encode_url_path_segment(run_id, field_name="run_id")
@ -351,10 +355,10 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> Run:
"""Transform OpenAI response to Run object"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming get run response: %s", response_json)
return Run(**response_json)
return Run.model_validate(response_json)
def transform_cancel_run_request(
self,
@ -363,14 +367,14 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict, Dict]:
) -> tuple[str, dict, dict]:
"""Transform cancel run request for OpenAI"""
encoded_eval_id = encode_url_path_segment(eval_id, field_name="eval_id")
encoded_run_id = encode_url_path_segment(run_id, field_name="run_id")
url = f"{api_base}/v1/evals/{encoded_eval_id}/runs/{encoded_run_id}/cancel"
# Empty body for cancel request
request_body: Dict[str, Any] = {}
request_body: dict[str, object] = {}
verbose_logger.debug("Cancel run request - URL: %s", url)
@ -382,10 +386,10 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> CancelRunResponse:
"""Transform OpenAI response to CancelRunResponse"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming cancel run response: %s", response_json)
return CancelRunResponse(**response_json)
return CancelRunResponse.model_validate(response_json)
def transform_delete_run_request(
self,
@ -394,14 +398,14 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict, Dict]:
) -> tuple[str, dict, dict]:
"""Transform delete run request for OpenAI"""
encoded_eval_id = encode_url_path_segment(eval_id, field_name="eval_id")
encoded_run_id = encode_url_path_segment(run_id, field_name="run_id")
url = f"{api_base}/v1/evals/{encoded_eval_id}/runs/{encoded_run_id}"
# Empty body for delete request
request_body: Dict[str, Any] = {}
request_body: dict[str, object] = {}
verbose_logger.debug("Delete run request - URL: %s", url)
@ -413,7 +417,7 @@ class OpenAIEvalsConfig(BaseEvalsAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> RunDeleteResponse:
"""Transform OpenAI response to RunDeleteResponse"""
response_json = raw_response.json()
response_json = _parsed_response_json(raw_response)
verbose_logger.debug("Transforming delete run response: %s", response_json)
return RunDeleteResponse(**response_json)
return RunDeleteResponse.model_validate(response_json)

185
litellm/llms/volcengine/responses/transformation.py
View file

@ -1,11 +1,9 @@
from collections.abc import Callable, Mapping, Sequence
from typing import (
TYPE_CHECKING,
Any,
Dict,
List,
Literal,
Optional,
Tuple,
Protocol,
Union,
get_args,
get_origin,
@ -17,10 +15,10 @@ from pydantic import fields as pyd_fields
import litellm
from litellm._logging import verbose_logger
from litellm.litellm_core_utils.core_helpers import process_response_headers
from litellm.litellm_core_utils.url_utils import encode_url_path_segment
from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response import (
_safe_convert_created_field,
)
from litellm.litellm_core_utils.url_utils import encode_url_path_segment
from litellm.llms.openai.responses.transformation import OpenAIResponsesAPIConfig
from litellm.secret_managers.main import get_secret_str
from litellm.types.llms.openai import (
@ -47,8 +45,15 @@ else:
LiteLLMLoggingObj = Any
class _EventModelClass(Protocol):
@property
def model_fields(self) -> Mapping[str, pyd_fields.FieldInfo]: ...
def model_validate(self, obj: Mapping[str, object]) -> ResponsesAPIStreamingResponse: ...
class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
_SUPPORTED_OPTIONAL_PARAMS: List[str] = [
_SUPPORTED_OPTIONAL_PARAMS: list[str] = [
# Doc-listed knobs
"instructions",
"max_output_tokens",
@ -89,9 +94,7 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
supported.remove("metadata")
return supported
def get_error_class(
self, error_message: str, status_code: int, headers: Union[dict, httpx.Headers]
) -> VolcEngineError:
def get_error_class(self, error_message: str, status_code: int, headers: dict | httpx.Headers) -> VolcEngineError:
typed_headers: httpx.Headers = headers if isinstance(headers, httpx.Headers) else httpx.Headers(headers or {})
return VolcEngineError(
status_code=status_code,
@ -99,14 +102,14 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
headers=typed_headers,
)
def validate_environment(self, headers: dict, model: str, litellm_params: Optional[GenericLiteLLMParams]) -> dict:
def validate_environment(self, headers: dict, model: str, litellm_params: GenericLiteLLMParams | None) -> dict:
"""
Build auth headers for Volcengine Responses API.
"""
if litellm_params is None:
litellm_params = GenericLiteLLMParams()
elif isinstance(litellm_params, dict):
litellm_params = GenericLiteLLMParams(**litellm_params)
litellm_params = GenericLiteLLMParams.model_validate(litellm_params)
api_key = (
litellm_params.api_key
@ -122,7 +125,7 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
def get_complete_url(
self,
api_base: Optional[str],
api_base: str | None,
litellm_params: dict,
) -> str:
"""
@ -149,7 +152,7 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
response_api_optional_params: ResponsesAPIOptionalRequestParams,
model: str,
drop_params: bool,
) -> Dict:
) -> dict:
"""
Volcengine Responses API aligns with OpenAI parameters.
Remove parameters not supported by the public docs.
@ -173,11 +176,11 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
def transform_responses_api_request(
self,
model: str,
input: Union[str, ResponseInputParam],
response_api_optional_request_params: Dict,
input: str | ResponseInputParam,
response_api_optional_request_params: dict,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Dict:
) -> dict:
"""
Volcengine rejects any undocumented fields (including extra_body). Fail fast
with clear errors and re-filter with the documented whitelist before delegating
@ -210,7 +213,7 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
def transform_streaming_response(
self,
model: str,
parsed_chunk: dict,
parsed_chunk: Mapping[str, object],
logging_obj: LiteLLMLoggingObj,
) -> ResponsesAPIStreamingResponse:
"""
@ -222,18 +225,19 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
if isinstance(chunk, dict):
resp = chunk.get("response")
if isinstance(resp, dict) and "output" not in resp:
resp_items: Mapping[str, object] = resp
patched_chunk = dict(chunk)
patched_resp = dict(resp)
patched_resp = dict(resp_items)
patched_resp["output"] = []
patched_chunk["response"] = patched_resp
chunk = patched_chunk
event_type = str(chunk.get("type")) if isinstance(chunk, dict) else None
event_pydantic_model = OpenAIResponsesAPIConfig.get_event_model_class(event_type=event_type)
event_pydantic_model: _EventModelClass = OpenAIResponsesAPIConfig.get_event_model_class(event_type=event_type)
patched_chunk = self._fill_missing_fields(chunk, event_pydantic_model)
return event_pydantic_model(**patched_chunk)
return event_pydantic_model.model_validate(patched_chunk)
def transform_response_api_response(
self,
@ -246,7 +250,7 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
original_response=raw_response.text,
additional_args={"complete_input_dict": {}},
)
raw_response_json = raw_response.json()
raw_response_json = self._parsed_response_body(raw_response)
if "created_at" in raw_response_json:
raw_response_json["created_at"] = _safe_convert_created_field(raw_response_json["created_at"])
except Exception:
@ -256,10 +260,11 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
processed_headers = process_response_headers(raw_response_headers)
try:
response = ResponsesAPIResponse(**raw_response_json)
response = ResponsesAPIResponse.model_validate(raw_response_json)
except Exception:
verbose_logger.debug("Volcengine Responses API: falling back to model_construct for response parsing.")
response = ResponsesAPIResponse.model_construct(**raw_response_json)
construct_response: Callable[..., ResponsesAPIResponse] = ResponsesAPIResponse.model_construct
response = construct_response(**raw_response_json)
response._hidden_params["additional_headers"] = processed_headers
response._hidden_params["headers"] = raw_response_headers
@ -274,10 +279,10 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict]:
) -> tuple[str, dict]:
encoded_response_id = encode_url_path_segment(response_id, field_name="response_id")
url = f"{api_base}/{encoded_response_id}"
data: Dict = {}
data: dict = {}
return url, data
def transform_delete_response_api_response(
@ -286,16 +291,17 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> DeleteResponseResult:
try:
raw_response_json = raw_response.json()
raw_response_json = self._parsed_response_body(raw_response)
except Exception:
raise VolcEngineError(message=raw_response.text, status_code=raw_response.status_code)
try:
return DeleteResponseResult(**raw_response_json)
return DeleteResponseResult.model_validate(raw_response_json)
except Exception:
verbose_logger.debug(
"Volcengine Responses API: falling back to model_construct for delete response parsing."
)
return DeleteResponseResult.model_construct(**raw_response_json)
construct_delete_result: Callable[..., DeleteResponseResult] = DeleteResponseResult.model_construct
return construct_delete_result(**raw_response_json)
#########################################################
########## GET RESPONSE API TRANSFORMATION ###############
@ -306,10 +312,10 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict]:
) -> tuple[str, dict]:
encoded_response_id = encode_url_path_segment(response_id, field_name="response_id")
url = f"{api_base}/{encoded_response_id}"
data: Dict = {}
data: dict = {}
return url, data
def transform_get_response_api_response(
@ -318,14 +324,14 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> ResponsesAPIResponse:
try:
raw_response_json = raw_response.json()
raw_response_json = self._parsed_response_body(raw_response)
except Exception:
raise VolcEngineError(message=raw_response.text, status_code=raw_response.status_code)
raw_response_headers = dict(raw_response.headers)
processed_headers = process_response_headers(raw_response_headers)
response = ResponsesAPIResponse(**raw_response_json)
response = ResponsesAPIResponse.model_validate(raw_response_json)
response._hidden_params["additional_headers"] = processed_headers
response._hidden_params["headers"] = raw_response_headers
return response
@ -339,15 +345,15 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
after: Optional[str] = None,
before: Optional[str] = None,
include: Optional[List[str]] = None,
after: str | None = None,
before: str | None = None,
include: list[str] | None = None,
limit: int = 20,
order: Literal["asc", "desc"] = "desc",
) -> Tuple[str, Dict]:
) -> tuple[str, dict]:
encoded_response_id = encode_url_path_segment(response_id, field_name="response_id")
url = f"{api_base}/{encoded_response_id}/input_items"
params: Dict[str, Any] = {}
params: dict[str, str | int] = {}
if after is not None:
params["after"] = after
if before is not None:
@ -364,9 +370,9 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
self,
raw_response: httpx.Response,
logging_obj: LiteLLMLoggingObj,
) -> Dict:
) -> dict:
try:
return raw_response.json()
return self._parsed_response_body(raw_response)
except Exception:
raise VolcEngineError(message=raw_response.text, status_code=raw_response.status_code)
@ -379,10 +385,10 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
api_base: str,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[str, Dict]:
) -> tuple[str, dict]:
encoded_response_id = encode_url_path_segment(response_id, field_name="response_id")
url = f"{api_base}/{encoded_response_id}/cancel"
data: Dict = {}
data: dict = {}
return url, data
def transform_cancel_response_api_response(
@ -391,23 +397,23 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
logging_obj: LiteLLMLoggingObj,
) -> ResponsesAPIResponse:
try:
raw_response_json = raw_response.json()
raw_response_json = self._parsed_response_body(raw_response)
except Exception:
raise VolcEngineError(message=raw_response.text, status_code=raw_response.status_code)
raw_response_headers = dict(raw_response.headers)
processed_headers = process_response_headers(raw_response_headers)
response = ResponsesAPIResponse(**raw_response_json)
response = ResponsesAPIResponse.model_validate(raw_response_json)
response._hidden_params["additional_headers"] = processed_headers
response._hidden_params["headers"] = raw_response_headers
return response
def should_fake_stream(
self,
model: Optional[str],
stream: Optional[bool],
custom_llm_provider: Optional[str] = None,
model: str | None,
stream: bool | None,
custom_llm_provider: str | None = None,
) -> bool:
"""
Volcengine Responses API supports native streaming; never fall back to fake stream.
@ -415,7 +421,24 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
return False
@staticmethod
def _fill_missing_fields(chunk: Any, event_model: Any) -> Dict[str, Any]:
def _parsed_response_body(raw_response: httpx.Response) -> dict[str, object]:
return raw_response.json()
@staticmethod
def _annotation_origin(annotation: object) -> object:
return get_origin(annotation)
@staticmethod
def _annotation_args(annotation: object) -> tuple[object, ...]:
return get_args(annotation)
@staticmethod
def _field_annotation(field: pyd_fields.FieldInfo) -> object:
annotation: object = field.annotation
return annotation
@staticmethod
def _fill_missing_fields(chunk: Mapping[str, object], event_model: object | None) -> Mapping[str, object]:
"""
Heuristically fill missing required fields with safe defaults based on the
event model's field annotations. This keeps parsing tolerant of providers that
@ -424,31 +447,37 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
if not isinstance(chunk, dict) or event_model is None:
return chunk
patched: Dict[str, Any] = dict(chunk)
fields_map = getattr(event_model, "model_fields", {}) or {}
patched = dict(chunk)
fields_map: Mapping[str, pyd_fields.FieldInfo] = getattr(event_model, "model_fields", {}) or {}
for name, field in fields_map.items():
if name in patched:
patched[name] = VolcEngineResponsesAPIConfig._maybe_fill_nested(patched[name], field.annotation)
patched[name] = VolcEngineResponsesAPIConfig._maybe_fill_nested(
patched[name], VolcEngineResponsesAPIConfig._field_annotation(field)
)
continue
# Explicit default or factory
if field.default is not pyd_fields.PydanticUndefined and field.default is not None:
patched[name] = field.default
field_default: object = field.default
if field_default is not pyd_fields.PydanticUndefined and field_default is not None:
patched[name] = field_default
continue
if field.default_factory is not None and field.default_factory is not pyd_fields.PydanticUndefined:
patched[name] = field.default_factory()
default_factory: Callable[..., object] | None = field.default_factory
if default_factory is not None and default_factory is not pyd_fields.PydanticUndefined:
patched[name] = default_factory()
continue
# Heuristic defaults for missing required fields
patched[name] = VolcEngineResponsesAPIConfig._default_for_annotation(field.annotation)
patched[name] = VolcEngineResponsesAPIConfig._default_for_annotation(
VolcEngineResponsesAPIConfig._field_annotation(field)
)
return patched
@staticmethod
def _default_for_annotation(annotation: Any) -> Any:
origin = get_origin(annotation)
args = get_args(annotation)
def _default_for_annotation(annotation: object) -> object:
origin = VolcEngineResponsesAPIConfig._annotation_origin(annotation)
args = VolcEngineResponsesAPIConfig._annotation_args(annotation)
if annotation is int:
return 0
@ -456,7 +485,7 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
return []
if origin is Union:
# Prefer empty list when any option is a list
if any((arg is list or get_origin(arg) is list) for arg in args):
if any((arg is list or VolcEngineResponsesAPIConfig._annotation_origin(arg) is list) for arg in args):
return []
if type(None) in args:
return None
@ -467,53 +496,51 @@ class VolcEngineResponsesAPIConfig(OpenAIResponsesAPIConfig):
return None
@staticmethod
def _maybe_fill_nested(value: Any, annotation: Any) -> Any:
def _maybe_fill_nested(value: object, annotation: object) -> object:
"""
Recursively fill nested dict/list structures based on the annotated model.
"""
model_cls = VolcEngineResponsesAPIConfig._pick_model_class(annotation, value)
args = get_args(annotation)
args = VolcEngineResponsesAPIConfig._annotation_args(annotation)
if isinstance(value, dict) and model_cls is not None:
return VolcEngineResponsesAPIConfig._fill_missing_fields(value, model_cls)
nested_items: Mapping[str, object] = value
return VolcEngineResponsesAPIConfig._fill_missing_fields(nested_items, model_cls)
if isinstance(value, list):
# Attempt to fill list elements if we know the element annotation
elem_ann: Any = args[0] if args else None
elem_ann: object = args[0] if args else None
if elem_ann is not None:
return [VolcEngineResponsesAPIConfig._maybe_fill_nested(v, elem_ann) for v in value]
nested_elements: Sequence[object] = value
return [VolcEngineResponsesAPIConfig._maybe_fill_nested(v, elem_ann) for v in nested_elements]
return value
@staticmethod
def _pick_model_class(annotation: Any, value: Any) -> Optional[Any]:
def _pick_model_class(annotation: object, value: object) -> object | None:
"""
Choose the best-matching Pydantic model class for a nested dict.
"""
candidates: List[Any] = []
origin = get_origin(annotation)
if hasattr(annotation, "model_fields"):
candidates.append(annotation)
if origin is Union:
for arg in get_args(annotation):
if hasattr(arg, "model_fields"):
candidates.append(arg)
origin = VolcEngineResponsesAPIConfig._annotation_origin(annotation)
union_args = VolcEngineResponsesAPIConfig._annotation_args(annotation) if origin is Union else ()
candidates = tuple(candidate for candidate in (annotation, *union_args) if hasattr(candidate, "model_fields"))
if not candidates:
return None
# Try to match by literal "type" field when available
if isinstance(value, dict):
v_type = value.get("type")
value_items: Mapping[str, object] = value
v_type = value_items.get("type")
for candidate in candidates:
try:
type_field = candidate.model_fields.get("type")
candidate_fields: Mapping[str, pyd_fields.FieldInfo] = getattr(candidate, "model_fields")
type_field = candidate_fields.get("type")
if type_field is None:
continue
literal_ann = type_field.annotation
if get_origin(literal_ann) is Literal:
literal_values = get_args(literal_ann)
literal_ann = VolcEngineResponsesAPIConfig._field_annotation(type_field)
if VolcEngineResponsesAPIConfig._annotation_origin(literal_ann) is Literal:
literal_values = VolcEngineResponsesAPIConfig._annotation_args(literal_ann)
if v_type in literal_values:
return candidate
except Exception:

10
litellm/model_prices_and_context_window_backup.json
View file

@ -2887,7 +2887,7 @@
"input_cost_per_token": 5e-06,
"output_cost_per_token": 2.5e-05,
"litellm_provider": "azure_ai",
"max_input_tokens": 200000,
"max_input_tokens": 1000000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"mode": "chat",
@ -2916,7 +2916,7 @@
"input_cost_per_token": 5e-06,
"output_cost_per_token": 2.5e-05,
"litellm_provider": "azure_ai",
"max_input_tokens": 200000,
"max_input_tokens": 1000000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"mode": "chat",
@ -3010,7 +3010,7 @@
"input_cost_per_token": 5e-06,
"output_cost_per_token": 2.5e-05,
"litellm_provider": "azure_ai",
"max_input_tokens": 200000,
"max_input_tokens": 1000000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"mode": "chat",
@ -45890,6 +45890,7 @@
"supports_native_streaming": true,
"supports_parallel_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_system_messages": true,
"supports_tool_choice": true,
@ -45915,6 +45916,7 @@
"supports_native_streaming": true,
"supports_parallel_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_system_messages": true,
"supports_tool_choice": true,
@ -45940,6 +45942,7 @@
"supports_native_streaming": true,
"supports_parallel_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_system_messages": true,
"supports_tool_choice": true,
@ -45999,6 +46002,7 @@
"supports_native_streaming": true,
"supports_parallel_function_calling": true,
"supports_prompt_caching": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_system_messages": true,
"supports_tool_choice": true,

35
litellm/ocr/main.py
View file

@ -7,9 +7,10 @@ import base64
import mimetypes
import os
import re
from collections.abc import Callable, Coroutine, Mapping
from dataclasses import dataclass
from io import IOBase
from typing import Any, Callable, Coroutine, Union, cast
from typing import Any, cast
import httpx
@ -42,7 +43,7 @@ class _PreparedOCRRequest:
provider_config: BaseOCRConfig
optional_params: dict[str, object]
litellm_params: dict[str, object]
effective_timeout: Union[float, httpx.Timeout]
effective_timeout: float | httpx.Timeout
litellm_logging_obj: LiteLLMLoggingObj
@ -63,13 +64,13 @@ _RUST_OCR_PROVIDERS = {
def _prepare_ocr_request(
model: str,
document: dict[str, Any],
document: Mapping[str, object],
api_key: str | None,
api_base: str | None,
timeout: Union[float, httpx.Timeout] | None,
timeout: float | httpx.Timeout | None,
custom_llm_provider: str | None,
extra_headers: dict[str, Any] | None,
kwargs: dict[str, Any],
extra_headers: dict[str, object] | None,
kwargs: dict[str, object],
) -> _PreparedOCRRequest:
litellm_logging_obj = cast(LiteLLMLoggingObj, kwargs.pop("litellm_logging_obj"))
litellm_call_id = cast(str | None, kwargs.get("litellm_call_id", None))
@ -120,7 +121,7 @@ def _prepare_ocr_request(
verbose_logger.debug(f"OCR call - model: {model}, provider: {custom_llm_provider}")
litellm_params = GenericLiteLLMParams(**kwargs)
litellm_params = GenericLiteLLMParams.model_validate(kwargs)
supported_params = ocr_provider_config.get_supported_ocr_params(model=model)
non_default_params = {}
@ -155,7 +156,7 @@ def _prepare_ocr_request(
api_key=api_key,
api_base=api_base,
custom_llm_provider=custom_llm_provider,
extra_headers=cast(dict[str, object] | None, extra_headers),
extra_headers=extra_headers,
provider_config=ocr_provider_config,
optional_params=cast(dict[str, object], optional_params),
litellm_params=dict(litellm_params),
@ -305,13 +306,13 @@ async def _run_rust_aocr(
@client
async def aocr(
model: str,
document: dict[str, Any],
document: Mapping[str, object],
api_key: str | None = None,
api_base: str | None = None,
timeout: Union[float, httpx.Timeout] | None = None,
timeout: float | httpx.Timeout | None = None,
custom_llm_provider: str | None = None,
extra_headers: dict[str, Any] | None = None,
**kwargs,
extra_headers: dict[str, object] | None = None,
**kwargs: object,
) -> OCRResponse:
"""
Async OCR function.
@ -567,14 +568,14 @@ def convert_file_document_to_url_document(document: dict[str, Any]) -> dict[str,
@client
def ocr(
model: str,
document: dict[str, Any],
document: Mapping[str, object],
api_key: str | None = None,
api_base: str | None = None,
timeout: Union[float, httpx.Timeout] | None = None,
timeout: float | httpx.Timeout | None = None,
custom_llm_provider: str | None = None,
extra_headers: dict[str, Any] | None = None,
**kwargs,
) -> Union[OCRResponse, Coroutine[Any, Any, OCRResponse]]:
extra_headers: dict[str, object] | None = None,
**kwargs: object,
) -> OCRResponse | Coroutine[object, object, OCRResponse]:
"""
Synchronous OCR function.

9
litellm/proxy/_experimental/mcp_server/auth/litellm_auth_handler.py
View file

@ -1,12 +1,9 @@
from typing import TYPE_CHECKING, Dict, List, Optional
from typing import Dict, List, Optional
from mcp.server.auth.middleware.bearer_auth import AuthenticatedUser
from litellm.proxy._types import UserAPIKeyAuth
if TYPE_CHECKING:
from opentelemetry.trace import SpanContext
class MCPAuthenticatedUser(AuthenticatedUser):
"""
@ -19,8 +16,6 @@ class MCPAuthenticatedUser(AuthenticatedUser):
4. Server-specific authentication headers
5. OAuth2 headers
6. Raw headers - allows forwarding specific headers to the MCP server, specified by the admin.
7. Transport span context - the tracing span of the HTTP request carrying the current
message, which a stateful session's message handler cannot read from its own task.
"""
def __init__(
@ -33,7 +28,6 @@ class MCPAuthenticatedUser(AuthenticatedUser):
mcp_protocol_version: Optional[str] = None,
raw_headers: Optional[Dict[str, str]] = None,
client_ip: Optional[str] = None,
transport_span_context: Optional["SpanContext"] = None,
):
self.user_api_key_auth = user_api_key_auth
self.mcp_auth_header = mcp_auth_header
@ -43,4 +37,3 @@ class MCPAuthenticatedUser(AuthenticatedUser):
self.oauth2_headers = oauth2_headers
self.raw_headers = raw_headers
self.client_ip = client_ip
self.transport_span_context = transport_span_context

493
litellm/proxy/_experimental/mcp_server/db.py
View file

File diff suppressed because it is too large Load diff

19
litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py
View file

@ -21,7 +21,6 @@ from litellm.llms.custom_httpx.http_handler import (
)
from litellm.proxy._experimental.mcp_server.auth.token_endpoint_auth import (
TokenEndpointAuthConfigError,
build_token_endpoint_client_auth,
normalize_token_endpoint_auth_method,
)
from litellm.types.mcp_server.mcp_server_manager import MCPTokenEndpointAuthMethod
@ -54,7 +53,9 @@ from litellm.proxy._experimental.mcp_server.gateway_dcr_flow import (
)
from litellm.proxy._experimental.mcp_server.oauth_utils import (
TOKEN_NO_CACHE_HEADERS,
build_upstream_oauth2_token_request,
get_request_base_url,
resolve_upstream_resource,
validate_trusted_redirect_uri,
well_known_root_suffix,
)
@ -726,6 +727,7 @@ def _redirect_to_upstream_authorize(
to the upstream authorize endpoint verbatim, no relay state cookie is set, and the upstream
enforces its own registered redirect binding for the client."""
scope_value = scope or (" ".join(mcp_server.scopes) if mcp_server.scopes else None)
upstream_resource = resolve_upstream_resource(mcp_server)
passthrough_params = {
"client_id": client_id,
"redirect_uri": redirect_uri,
@ -734,6 +736,7 @@ def _redirect_to_upstream_authorize(
"code_challenge": code_challenge,
"code_challenge_method": code_challenge_method,
**({"scope": scope_value} if scope_value else {}),
**({"resource": upstream_resource} if upstream_resource else {}),
}
parsed_auth_url = urlparse(mcp_server.authorization_url or "")
merged_params = {**dict(parse_qsl(parsed_auth_url.query)), **passthrough_params}
@ -842,6 +845,10 @@ async def authorize_with_server(
if code_challenge_method:
params["code_challenge_method"] = code_challenge_method
upstream_resource = resolve_upstream_resource(mcp_server)
if upstream_resource:
params["resource"] = upstream_resource
parsed_auth_url = urlparse(mcp_server.authorization_url)
existing_params = dict(parse_qsl(parsed_auth_url.query))
existing_params.update(params)
@ -902,7 +909,8 @@ async def exchange_token_with_server(
else (client_token_endpoint_auth_method or mcp_server.token_endpoint_auth_method)
)
try:
client_auth = build_token_endpoint_client_auth(
token_request = build_upstream_oauth2_token_request(
mcp_server,
auth_method=resolved_auth_method,
client_id=resolved_client_id,
client_secret=resolved_client_secret,
@ -941,7 +949,7 @@ async def exchange_token_with_server(
token_data: dict = {
"grant_type": "refresh_token",
"refresh_token": upstream_refresh_token,
**client_auth.body,
**token_request.body,
}
refresh_request_scope = scope or bridge_upstream_scope
if refresh_request_scope:
@ -980,7 +988,7 @@ async def exchange_token_with_server(
"grant_type": "authorization_code",
"code": code,
"redirect_uri": resolved_redirect_uri,
**client_auth.body,
**token_request.body,
}
if code_verifier:
token_data["code_verifier"] = code_verifier
@ -991,11 +999,12 @@ async def exchange_token_with_server(
if not isinstance(prepared, _BridgeMintReady):
return _bridge_mint_error_response(prepared)
bridge_mint_ready = prepared
async_client = get_async_httpx_client(llm_provider=httpxSpecialProvider.Oauth2Check)
try:
response = await async_client.post(
mcp_server.token_url,
headers={"Accept": "application/json", **client_auth.headers},
headers={"Accept": "application/json", **token_request.headers},
data=token_data,
)
if response is not None:

17
litellm/proxy/_experimental/mcp_server/faults/classify.py
View file

@ -63,18 +63,21 @@ def _classify_oauth_error_code(
) -> UpstreamOAuthFault:
"""Blame assignment for a contract-conformant OAuth error code, shared by the token and DCR
classifiers. Codes by which the upstream blames itself keep that blame; ``invalid_target`` is a
gateway capability gap (RFC 8707 resource indicators, LIT-4339) no matter whose credentials were
presented; credential-indicting codes follow the credential source; everything else, including
codes we do not recognize, is the caller's to act on. The upstream's HTTP status is deliberately
never consulted: status derives from this classification at render time, which is what keeps
status and code from contradicting each other."""
gateway configuration gap (the RFC 8707 resource indicator this server sends, or fails to send)
no matter whose credentials were presented; credential-indicting codes follow the credential
source; everything else, including codes we do not recognize, is the caller's to act on. The
upstream's HTTP status is deliberately never consulted: status derives from this classification
at render time, which is what keeps status and code from contradicting each other."""
if code == "server_error" or code == "temporarily_unavailable":
return UpstreamReportedFault(code=code)
if code in GATEWAY_CAPABILITY_CODES:
verbose_logger.warning(
"MCP server %s: the upstream authorization server rejected the request with "
"invalid_target; it may require RFC 8707 resource indicators, which the gateway "
"does not send yet (tracked as LIT-4339)",
"invalid_target, meaning it did not accept the RFC 8707 resource indicator for this "
"request. Set upstream_resource on this server to the exact resource identifier the "
"authorization server expects (or to 'auto' to send the server's own canonical url); "
"if it is already set and the authorization server does not support resource "
"indicators, unset it and express the target audience through scopes instead",
log_context,
)
return GatewayRejected(code=code)

6
litellm/proxy/_experimental/mcp_server/faults/render_oauth.py
View file

@ -16,8 +16,10 @@ from litellm.proxy._experimental.mcp_server.oauth_utils import TOKEN_NO_CACHE_HE
def _gateway_rejected_description(code: str) -> str:
if code == "invalid_target":
return (
"the upstream authorization server rejected the request (invalid_target); "
"it may require RFC 8707 resource indicators, which the gateway does not send yet"
"the upstream authorization server rejected the request (invalid_target); it did not "
"accept this server's RFC 8707 resource indicator. Set upstream_resource on the MCP "
"server to the resource identifier the authorization server expects, or unset it if "
"that authorization server does not support resource indicators"
)
return (
f"the upstream authorization server rejected the gateway's configured client credentials "

7
litellm/proxy/_experimental/mcp_server/faults/types.py
View file

@ -25,9 +25,10 @@ gateway presented its own stored credentials, these are gateway-side faults the
when the caller supplied the credentials, they are the caller's to fix."""
GATEWAY_CAPABILITY_CODES: frozenset[str] = frozenset({"invalid_target"})
"""Codes that indict a gateway capability regardless of whose credentials were presented:
``invalid_target`` means the upstream wants RFC 8707 resource indicators, which the gateway does not
send yet (LIT-4339). Never the caller's fault."""
"""Codes that indict gateway configuration regardless of whose credentials were presented:
``invalid_target`` means the upstream did not accept the RFC 8707 resource indicator the server
sent, or requires one it was not configured to send (``upstream_resource``). Never the caller's
fault."""
UPSTREAM_FAULT_CODES: frozenset[str] = frozenset({"server_error", "temporarily_unavailable"})
"""Codes by which the upstream blames itself. Relaying them as caller faults would invert blame, so

38
litellm/proxy/_experimental/mcp_server/mcp_server_manager.py
View file

@ -71,6 +71,7 @@ from litellm.proxy._experimental.mcp_server.oauth2_token_cache import (
)
from litellm.proxy._experimental.mcp_server.oauth_utils import (
_redact_mcp_resource_url,
canonicalize_url_identity,
)
from litellm.proxy._experimental.mcp_server.outbound_credentials import (
Error,
@ -121,7 +122,6 @@ from litellm.proxy._experimental.mcp_server.utils import (
normalize_server_name,
parse_admin_env_vars,
split_server_prefix_from_name,
strip_known_server_prefix,
validate_mcp_server_name,
)
from litellm.proxy._types import (
@ -262,19 +262,11 @@ def _endpoints_yield_to_issuer(
def _normalized_authorize_endpoint(url: str) -> str:
"""Compare authorize endpoints on scheme, host, and path only. The default port is elided and
the host is lowercased so ``https://IDP.example.com:443/authorize/`` and
``https://idp.example.com/authorize`` are the same identity; query and trailing slash are not."""
parsed = urlparse(url)
scheme = parsed.scheme.lower()
host = (parsed.hostname or "").lower()
default_port = {"https": 443, "http": 80}.get(scheme)
try:
port = parsed.port
except ValueError:
port = None
authority = host if port is None or port == default_port else f"{host}:{port}"
return f"{scheme}://{authority}{parsed.path.rstrip('/')}"
"""Compare authorize endpoints / issuers on scheme, host, and path only, through the shared URL
canonicalizer: the default port is elided and the host is lowercased so
``https://IDP.example.com:443/authorize/`` and ``https://idp.example.com/authorize`` are the same
identity, while query, fragment and a trailing slash are dropped."""
return canonicalize_url_identity(url)
def _issuer_matches(claimed_issuer: object, configured_issuer: str) -> bool:
@ -1518,6 +1510,7 @@ class MCPServerManager:
"subject_token_type",
DEFAULT_SUBJECT_TOKEN_TYPE,
),
upstream_resource=server_config.get("upstream_resource", None),
# ID-JAG fields
id_jag_resource_token_endpoint=server_config.get("id_jag_resource_token_endpoint", None),
id_jag_resource=server_config.get("id_jag_resource", None),
@ -2017,6 +2010,7 @@ class MCPServerManager:
subject_token_type=mcp_server.subject_token_type
or (credentials_dict.get("subject_token_type") if credentials_dict else None)
or DEFAULT_SUBJECT_TOKEN_TYPE,
upstream_resource=(credentials_dict.get("upstream_resource") if credentials_dict else None),
# ID-JAG fields — read from credentials JSON blob
id_jag_resource_token_endpoint=(
credentials_dict.get("id_jag_resource_token_endpoint") if credentials_dict else None
@ -2492,6 +2486,13 @@ class MCPServerManager:
the given toolsets. Results are cached via ``user_api_key_cache`` (a
Redis-backed ``DualCache`` in production) so that cache entries are
shared across workers and cold-cache DB hits are minimised.
A row names a tool on the server identified by ``server_id``, so the
stored name is the tool's own name and is used as written. It is never
reduced by the server's wire prefix: that prefix is added on the way out
and is not part of any tool's identity, so treating a leading segment as
one silently renames the tool when a native name happens to begin with
it (``greyhound_internal_events`` on a server prefixed ``greyhound``).
"""
from litellm.proxy._experimental.mcp_server.toolset_db import list_mcp_toolsets
from litellm.proxy.proxy_server import prisma_client, user_api_key_cache
@ -2509,12 +2510,9 @@ class MCPServerManager:
tool_permissions: dict[str, list[str]] = {}
for toolset in toolsets:
for tool in toolset.tools:
raw_name = tool["tool_name"]
server = self.get_mcp_server_by_id(tool["server_id"])
unprefixed = strip_known_server_prefix(raw_name, server)
tool_permissions.setdefault(tool["server_id"], [])
if unprefixed not in tool_permissions[tool["server_id"]]:
tool_permissions[tool["server_id"]].append(unprefixed)
allowed_names = tool_permissions.setdefault(tool["server_id"], [])
if tool["tool_name"] not in allowed_names:
allowed_names.append(tool["tool_name"])
await user_api_key_cache.async_set_cache(
key=cache_key,
value=tool_permissions,

68
litellm/proxy/_experimental/mcp_server/oauth2_token_cache.py
View file

@ -6,6 +6,7 @@ with ``client_id``, ``client_secret``, and ``token_url``.
"""
import asyncio
import hashlib
from typing import TYPE_CHECKING, Dict, Optional, Tuple, Union
import httpx
@ -26,8 +27,9 @@ from litellm.proxy.common_utils.encrypt_decrypt_utils import (
decrypt_value_helper,
encrypt_value_helper,
)
from litellm.proxy._experimental.mcp_server.auth.token_endpoint_auth import (
build_token_endpoint_client_auth,
from litellm.proxy._experimental.mcp_server.oauth_utils import (
build_upstream_oauth2_token_request,
resolve_upstream_resource,
)
from litellm.types.llms.custom_http import httpxSpecialProvider
@ -37,10 +39,18 @@ if TYPE_CHECKING:
class MCPOAuth2TokenCache(InMemoryCache):
"""
In-memory cache for OAuth2 client_credentials tokens, keyed by server_id.
In-memory cache for OAuth2 client_credentials tokens, keyed by the identity of the token
request rather than by server_id alone.
A minted token is only reusable for the exact request that produced it. Keying on server_id
alone served a token minted under the previous configuration whenever any of those inputs
changed, so editing scopes, rotating the client secret, or setting ``upstream_resource``
silently kept handing out a token carrying the old scopes or audience until it expired. The
identity below covers every input ``_fetch_token`` puts on the wire, so a change to any of
them misses the cache and mints afresh.
Inherits from ``InMemoryCache`` for TTL-based storage and eviction.
Adds per-server ``asyncio.Lock`` to prevent duplicate concurrent fetches.
Adds a per-identity ``asyncio.Lock`` to prevent duplicate concurrent fetches.
"""
def __init__(self) -> None:
@ -50,8 +60,25 @@ class MCPOAuth2TokenCache(InMemoryCache):
)
self._locks: Dict[str, asyncio.Lock] = {}
def _get_lock(self, server_id: str) -> asyncio.Lock:
return self._locks.setdefault(server_id, asyncio.Lock())
@staticmethod
def _token_identity(server: "MCPServer") -> str:
"""Cache key for the token this server's config would mint, prefixed by server_id so a
single server's entries stay greppable and invalidatable. The secret is hashed with the
rest of the identity rather than stored in a key."""
material = "\x00".join(
(
server.token_url or "",
server.client_id or "",
server.client_secret or "",
" ".join(server.scopes or ()),
resolve_upstream_resource(server) or "",
server.token_endpoint_auth_method or "",
)
)
return f"{server.server_id}:{hashlib.sha256(material.encode()).hexdigest()}"
def _get_lock(self, identity: str) -> asyncio.Lock:
return self._locks.setdefault(identity, asyncio.Lock())
@staticmethod
def _has_client_credentials_config(server: "MCPServer") -> bool:
@ -67,21 +94,21 @@ class MCPOAuth2TokenCache(InMemoryCache):
if not self._has_client_credentials_config(server):
return None
server_id = server.server_id
identity = self._token_identity(server)
# Fast path — cached token is still valid
cached = self.get_cache(server_id)
cached = self.get_cache(identity)
if cached is not None:
return cached
# Slow path — acquire per-server lock then double-check
async with self._get_lock(server_id):
cached = self.get_cache(server_id)
# Slow path — acquire per-identity lock then double-check
async with self._get_lock(identity):
cached = self.get_cache(identity)
if cached is not None:
return cached
token, ttl = await self._fetch_token(server)
self.set_cache(server_id, token, ttl=ttl)
self.set_cache(identity, token, ttl=ttl)
return token
async def _fetch_token(self, server: "MCPServer") -> Tuple[str, int]:
@ -100,14 +127,15 @@ class MCPOAuth2TokenCache(InMemoryCache):
f"token_url={bool(server.token_url)}"
)
client_auth = build_token_endpoint_client_auth(
token_request = build_upstream_oauth2_token_request(
server,
auth_method=server.token_endpoint_auth_method,
client_id=server.client_id,
client_secret=server.client_secret,
)
data: Dict[str, str] = {
"grant_type": "client_credentials",
**client_auth.body,
**token_request.body,
}
if server.scopes:
data["scope"] = " ".join(server.scopes)
@ -117,7 +145,7 @@ class MCPOAuth2TokenCache(InMemoryCache):
server.server_id,
)
post_kwargs = {"data": data, **({"headers": client_auth.headers} if client_auth.headers else {})}
post_kwargs = {"data": data, **({"headers": token_request.headers} if token_request.headers else {})}
try:
response = await client.post(server.token_url, **post_kwargs)
response.raise_for_status()
@ -159,8 +187,14 @@ class MCPOAuth2TokenCache(InMemoryCache):
return access_token, ttl
def invalidate(self, server_id: str) -> None:
"""Remove a cached token (e.g. after a 401)."""
self.delete_cache(server_id)
"""Remove every cached token for a server (e.g. after a 401).
Entries are keyed by token identity, so one server can hold more than one entry across a
config change; a 401 invalidates all of them rather than only the current configuration's.
"""
prefix = f"{server_id}:"
for key in [k for k in self.cache_dict if isinstance(k, str) and k.startswith(prefix)]:
self.delete_cache(key)
mcp_oauth2_token_cache = MCPOAuth2TokenCache()

123
litellm/proxy/_experimental/mcp_server/oauth_utils.py
View file

@ -3,14 +3,22 @@
import os
from ipaddress import ip_address
from typing import Any, Dict, List, NoReturn, Optional
from typing import TYPE_CHECKING, Any, Dict, List, NoReturn, Optional
from urllib.parse import ParseResult, urlparse, urlsplit, urlunparse, urlunsplit
from fastapi import HTTPException, Request
from litellm._logging import verbose_logger
from litellm.proxy._experimental.mcp_server.auth.token_endpoint_auth import (
TokenEndpointClientAuth,
build_token_endpoint_client_auth,
normalize_token_endpoint_auth_method,
)
from litellm.proxy.auth.ip_address_utils import IPAddressUtils
if TYPE_CHECKING:
from litellm.types.mcp_server.mcp_server_manager import MCPServer
# RFC 6749 §5.1 / OAuth 2.1 draft-15 §4.1.3: token-endpoint responses
# must not be cached — both success and error bodies may reveal secrets.
TOKEN_NO_CACHE_HEADERS = {"Cache-Control": "no-store", "Pragma": "no-cache"}
@ -21,6 +29,10 @@ TOKEN_NO_CACHE_HEADERS = {"Cache-Control": "no-store", "Pragma": "no-cache"}
# explicit port, which would otherwise break a literal netloc compare).
_DEFAULT_PORTS = {"http": 80, "https": 443}
# Sentinel ``upstream_resource`` value meaning "derive the RFC 8707 resource identifier from the
# server's own url". RFC 8707 requires an absolute URI, so this can never be a real resource value.
UPSTREAM_RESOURCE_AUTO = "auto"
# Env var for ops to allowlist additional redirect_uri origins beyond
# same-origin + loopback — needed for first-party OAuth clients hosted
# on sister domains (e.g. a web app on app.example.com registering as
@ -574,3 +586,112 @@ def validate_trusted_redirect_uri(request: Request, redirect_uri: str) -> None:
if _trusted_redirect_uri_is_allowed(parsed, redirect_netloc, proxy_base):
return
_raise_trusted_redirect_uri_rejected(request, redirect_uri, parsed, redirect_netloc, proxy_base)
def canonicalize_url_identity(url: str) -> str:
"""Normalize a URL to a comparable identity: lowercase scheme and host, drop the scheme's default
port, and strip userinfo, params, query, fragment and a trailing slash while keeping IPv6
brackets. The one URL-canonicalization primitive shared by the RFC 8707 resource emitter and the
RFC 8414 issuer/authorize-endpoint comparison, so the default-port and IPv6 rules cannot be
present in one and missing in the other. The netloc (not ``parsed.hostname``) carries the
authority so ``[::1]:8080`` survives with its brackets intact."""
parsed = urlparse(url)
scheme = parsed.scheme.lower()
netloc = _strip_default_port(scheme, parsed.netloc.rpartition("@")[2])
return urlunparse((scheme, netloc, parsed.path.rstrip("/"), "", "", ""))
def _canonical_resource_uri(url: str) -> str | None:
"""Canonicalize an upstream MCP server URL into an RFC 8707 resource identifier.
Keeps only the scheme, host, port and path, which is the shape the MCP authorization spec's
"Canonical Server URI" section describes and every one of its examples takes; the reference
implementation is ``mcp.shared.auth_utils.resource_url_from_server_url``, and this is the stricter
variant. The scheme and host are lowercased, the scheme's default port is dropped so
``https://host:443/mcp`` and ``https://host/mcp`` never present as two resources, and a trailing
slash is dropped so ``https://host/mcp/`` and ``https://host/mcp`` do not either.
Userinfo, query and fragment are dropped rather than carried. A transport URL routinely holds
credentials in exactly those components (``user:password@``, ``?api_key=``), while a resource
indicator names the resource and nothing else; this value is published somewhere the transport
URL never goes, into the authorization redirect the browser follows and into token request
bodies, so carrying them would disclose them to the authorization server, its logs, and browser
history. RFC 8707 forbids a fragment outright and says a resource SHOULD NOT carry a query. An
upstream whose identifier genuinely needs more than this is served by setting
``upstream_resource`` explicitly, which is passed through untouched.
Returns ``None`` when the URL is not absolute, which cannot yield a valid resource identifier.
"""
parsed = urlparse(url)
if not parsed.scheme or not parsed.netloc:
return None
return canonicalize_url_identity(url)
def resolve_upstream_resource(mcp_server: "MCPServer") -> str | None:
"""Resolve the RFC 8707 ``resource`` value this server's upstream OAuth legs must carry.
The MCP authorization spec requires an MCP client to send ``resource`` on both the
authorization request and every token request, naming the canonical URI of the MCP server the
token is for. Authorization server temperaments are irreconcilable and undetectable, so this
stays an explicit per-server opt-in: most SaaS providers ignore the parameter, some hard-reject
it and express audience through scopes instead, and strict or MCP-native ones refuse to mint a
correctly scoped token without it (``invalid_target``).
``None`` or blank omits the parameter, which is the default and preserves the behavior of every
server working today. ``"auto"`` derives the canonical URI from the server's own URL; it is not
an absolute URI, so RFC 8707 guarantees it can never collide with a real resource value. Any
other value is sent verbatim, because the identifier has to match what the authorization server
expects exactly and normalizing it could break that match.
Every upstream leg for a server resolves through this one function, so the authorize request
and the token requests cannot disagree; a token request naming a resource the authorization
request never asked for is itself an ``invalid_target`` under RFC 8707.
"""
configured = (mcp_server.upstream_resource or "").strip()
if not configured:
return None
if configured.lower() != UPSTREAM_RESOURCE_AUTO:
return configured
if not mcp_server.url:
verbose_logger.warning(
"MCP server %s sets upstream_resource=auto but has no url to derive a resource "
"identifier from; omitting the RFC 8707 resource parameter. Set upstream_resource to "
"the exact resource identifier the authorization server expects instead.",
mcp_server.server_id,
)
return None
canonical = _canonical_resource_uri(mcp_server.url)
if canonical is None:
verbose_logger.warning(
"MCP server %s sets upstream_resource=auto but its url is not an absolute URI, so no "
"RFC 8707 resource identifier could be derived; omitting the resource parameter",
mcp_server.server_id,
)
return canonical
def build_upstream_oauth2_token_request(
mcp_server: "MCPServer",
*,
auth_method: object,
client_id: str | None,
client_secret: str | None,
) -> TokenEndpointClientAuth:
"""Client auth plus the RFC 8707 ``resource`` for one upstream plain-OAuth2 token request.
Resolving both in one call is what stops a leg authenticating without naming the resource its
sibling legs named; the RFC 8693 legs (OBO, id_jag) carry ``audience`` and stay on
``build_token_endpoint_client_auth``. The client-auth inputs are passed in because a leg may
authenticate as the caller's own client rather than the server's; ``resource`` always comes from
the server, so no leg can choose or forget it.
"""
client_auth = build_token_endpoint_client_auth(
auth_method=normalize_token_endpoint_auth_method(auth_method),
client_id=client_id,
client_secret=client_secret,
)
resource = resolve_upstream_resource(mcp_server)
if not resource:
return client_auth
return TokenEndpointClientAuth(headers=client_auth.headers, body={**client_auth.body, "resource": resource})

2
litellm/proxy/_experimental/mcp_server/outbound_credentials/adapter.py
View file

@ -18,6 +18,7 @@ from fastapi import HTTPException
from pydantic import SecretStr
from typing_extensions import assert_never
from litellm.proxy._experimental.mcp_server.oauth_utils import resolve_upstream_resource
from litellm.proxy._experimental.mcp_server.outbound_credentials.types import (
ApiKeyConfig,
AuthorizationCodeConfig,
@ -144,6 +145,7 @@ def _client_credentials_spec(server: MCPServer, resource: str) -> ServerSpec:
token_url=server.token_url,
scopes=tuple(server.scopes or ()),
audience=server.audience,
upstream_resource=resolve_upstream_resource(server),
token_endpoint_auth_method=server.token_endpoint_auth_method,
),
)

9
litellm/proxy/_experimental/mcp_server/outbound_credentials/authz_code_refresher.py
View file

@ -17,8 +17,8 @@ from typing import TYPE_CHECKING, Protocol
from litellm._logging import verbose_logger
from litellm.proxy._experimental.mcp_server.auth.token_endpoint_auth import (
TokenEndpointAuthConfigError,
build_token_endpoint_client_auth,
)
from litellm.proxy._experimental.mcp_server.oauth_utils import build_upstream_oauth2_token_request
from litellm.proxy._experimental.mcp_server.outbound_credentials.oauth_token_store import (
OAuthToken,
)
@ -92,7 +92,8 @@ class AuthorizationCodeRefresher:
return None
try:
client_auth = build_token_endpoint_client_auth(
token_request = build_upstream_oauth2_token_request(
server,
auth_method=server.token_endpoint_auth_method,
client_id=server.client_id,
client_secret=server.client_secret,
@ -103,9 +104,9 @@ class AuthorizationCodeRefresher:
form = {
"grant_type": "refresh_token",
"refresh_token": token.refresh_token,
**client_auth.body,
**token_request.body,
}
body = await self._token_endpoint(server.token_url, form, client_auth.headers)
body = await self._token_endpoint(server.token_url, form, token_request.headers)
if body is None:
return None
access_token = body.get("access_token")

2
litellm/proxy/_experimental/mcp_server/outbound_credentials/client_credentials.py
View file

@ -292,6 +292,7 @@ def _prepare_grant(config: ClientCredentialsConfig) -> Result[_PreparedGrant, Cr
**client_auth.body,
**({"scope": " ".join(config.scopes)} if config.scopes else {}),
**({"audience": config.audience} if config.audience else {}),
**({"resource": config.upstream_resource} if config.upstream_resource else {}),
}
return Ok(
_PreparedGrant(
@ -313,6 +314,7 @@ def _identity_key(config: ClientCredentialsConfig) -> str:
config.token_endpoint_auth_method or "",
" ".join(config.scopes),
config.audience or "",
config.upstream_resource or "",
)
)
return hashlib.sha256(material.encode("utf-8")).hexdigest()

4
litellm/proxy/_experimental/mcp_server/outbound_credentials/per_user_oauth_store.py
View file

@ -11,7 +11,7 @@ collaborators acquire their globals per call, mirroring v1's lazy-import pattern
from __future__ import annotations
import asyncio
from collections.abc import Callable
from collections.abc import Callable, Mapping
from typing import TYPE_CHECKING
from litellm._logging import verbose_logger
@ -54,7 +54,7 @@ ServerLookup = Callable[[str], "MCPServer | None"]
StoreBuilder = Callable[[ServerLookup], tuple[InvalidatableOAuthTokenStore, bool]]
async def _read_credential(user_id: str, server_id: str) -> dict[str, object] | None:
async def _read_credential(user_id: str, server_id: str) -> Mapping[str, object] | None:
from litellm.proxy._experimental.mcp_server.db import ( # noqa: PLC0415
get_user_oauth_credential,
)

1
litellm/proxy/_experimental/mcp_server/outbound_credentials/types.py
View file

@ -199,6 +199,7 @@ class ClientCredentialsConfig(BaseModel):
token_url: str | None = None
scopes: tuple[str, ...] = ()
audience: str | None = None
upstream_resource: str | None = None
token_endpoint_auth_method: Literal["client_secret_post", "client_secret_basic"] | None = None

6
litellm/proxy/_experimental/mcp_server/outbound_credentials/v2_token_store.py
View file

@ -10,14 +10,14 @@ injected, so the DB/decoding plumbing stays testable and out of this seam.
from __future__ import annotations
from collections.abc import Awaitable, Callable
from collections.abc import Awaitable, Callable, Mapping
from datetime import datetime, timezone
from litellm.proxy._experimental.mcp_server.outbound_credentials.oauth_token_store import (
OAuthToken,
)
CredentialReader = Callable[[str, str], Awaitable["dict[str, object] | None"]]
CredentialReader = Callable[[str, str], Awaitable["Mapping[str, object] | None"]]
def _iso_to_epoch(expires_at: str) -> float | None:
@ -39,7 +39,7 @@ def _to_scopes(raw: object) -> tuple[str, ...]:
return ()
def _to_oauth_token(payload: dict[str, object]) -> OAuthToken | None:
def _to_oauth_token(payload: Mapping[str, object]) -> OAuthToken | None:
access_token = payload.get("access_token")
if not isinstance(access_token, str):
return None

102
litellm/proxy/_experimental/mcp_server/rest_endpoints.py
View file

@ -1,18 +1,11 @@
import asyncio
import importlib
from collections.abc import Awaitable, Callable, Mapping
from datetime import datetime
from typing import (
TYPE_CHECKING,
Any,
Awaitable,
Callable,
Dict,
List,
Literal,
Mapping,
Optional,
Set,
Tuple,
Union,
)
import httpx
@ -38,6 +31,9 @@ from litellm.proxy._experimental.mcp_server.utils import (
from litellm.proxy._types import LitellmUserRoles, UserAPIKeyAuth
from litellm.proxy.auth.ip_address_utils import IPAddressUtils
from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
if TYPE_CHECKING:
from litellm.proxy._experimental.mcp_server.db import OAuthCredentialPayload
from litellm.proxy.common_utils.http_parsing_utils import _safe_get_request_headers
from litellm.types.mcp import MCPAuth
from litellm.types.utils import CallTypes
@ -97,12 +93,12 @@ if MCP_AVAILABLE:
########################################################
############ MCP Server REST API Routes #################
async def _safe_fire_mcp_tool_call_logging(
logging_obj: Optional[Any],
logging_obj: Any | None,
result: Any,
start_time: datetime,
end_time: datetime,
user_api_key_auth: Optional[UserAPIKeyAuth] = None,
request_data: Optional[Mapping[str, object]] = None,
user_api_key_auth: UserAPIKeyAuth | None = None,
request_data: Mapping[str, object] | None = None,
) -> None:
if logging_obj is None:
return
@ -134,7 +130,7 @@ if MCP_AVAILABLE:
async def _handle_virtual_mcp_tool(
request: Request,
data: Dict[str, Any],
data: dict[str, Any],
tool_name: str,
user_api_key_dict: UserAPIKeyAuth,
) -> Any:
@ -212,9 +208,9 @@ if MCP_AVAILABLE:
def _get_server_auth_header(
server,
mcp_server_auth_headers: Optional[Dict[str, Dict[str, str]]],
mcp_auth_header: Optional[str],
) -> Optional[Union[Dict[str, str], str]]:
mcp_server_auth_headers: dict[str, dict[str, str]] | None,
mcp_auth_header: str | None,
) -> dict[str, str] | str | None:
"""Helper function to get server-specific auth header with case-insensitive matching."""
from litellm.proxy._experimental.mcp_server.utils import (
lookup_mcp_server_auth_in_headers,
@ -230,7 +226,7 @@ if MCP_AVAILABLE:
return server_auth
return mcp_auth_header
def _is_v1_resolved_oauth2_server(server: Optional[MCPServer]) -> bool:
def _is_v1_resolved_oauth2_server(server: MCPServer | None) -> bool:
"""Whether this server's per-user OAuth2 token is still resolved by v1.
A server the v2 resolver owns reads its stored token from the resolver at connect
@ -246,7 +242,7 @@ if MCP_AVAILABLE:
return False
return to_server_spec(server) is None
def _v1_resolved_oauth2_server_ids(allowed_server_ids: List[str]) -> Set[str]:
def _v1_resolved_oauth2_server_ids(allowed_server_ids: list[str]) -> set[str]:
"""Return the subset of *allowed_server_ids* whose per-user OAuth2 token is still
resolved by v1.
@ -260,10 +256,10 @@ if MCP_AVAILABLE:
}
async def _get_user_oauth_extra_headers(
server,
server: MCPServer,
user_api_key_dict: UserAPIKeyAuth,
prefetched_creds: Optional[Dict[str, Dict[str, Any]]] = None,
) -> Optional[Dict[str, str]]:
prefetched_creds: dict[str, "OAuthCredentialPayload"] | None = None,
) -> dict[str, str] | None:
"""
For OAuth2 servers, look up the user's stored access token and return it
as extra_headers {"Authorization": "Bearer <token>"} so that it reaches
@ -315,7 +311,7 @@ if MCP_AVAILABLE:
async def _prefetch_user_oauth_creds(
user_api_key_dict: UserAPIKeyAuth,
) -> Dict[str, Dict[str, Any]]:
) -> dict[str, "OAuthCredentialPayload"]:
"""Fetch all OAuth2 credentials for the user in a single DB query.
Returns a dict keyed by server_id. Used to avoid N+1 DB queries when
@ -379,8 +375,8 @@ if MCP_AVAILABLE:
def _resolve_mcp_server_id_for_rest(
server_id: str,
allowed_server_ids: Union[Set[str], List[str]],
client_ip: Optional[str] = None,
allowed_server_ids: set[str] | list[str],
client_ip: str | None = None,
) -> str:
"""
Map REST ``server_id`` (UUID, server_name, or alias) to canonical server_id.
@ -400,7 +396,7 @@ if MCP_AVAILABLE:
request: Request,
user_api_key_dict: UserAPIKeyAuth,
server_id: str,
) -> Tuple[List[MCPServer], str]:
) -> tuple[list[MCPServer], str]:
"""
Resolve allowed MCP servers for a tool call with IP filtering.
@ -471,7 +467,7 @@ if MCP_AVAILABLE:
)
# Build allowed_mcp_servers list (only include allowed servers)
allowed_mcp_servers: List[MCPServer] = []
allowed_mcp_servers: list[MCPServer] = []
for allowed_server_id in allowed_server_ids_set:
server = global_mcp_server_manager.get_mcp_server_by_id(allowed_server_id)
if server is not None:
@ -482,9 +478,9 @@ if MCP_AVAILABLE:
async def _get_tools_for_single_server(
server,
server_auth_header,
raw_headers: Optional[Dict[str, str]] = None,
user_api_key_auth: Optional[UserAPIKeyAuth] = None,
extra_headers: Optional[Dict[str, str]] = None,
raw_headers: dict[str, str] | None = None,
user_api_key_auth: UserAPIKeyAuth | None = None,
extra_headers: dict[str, str] | None = None,
apply_tool_filters: bool = True,
):
"""Helper function to get tools for a single server.
@ -530,7 +526,7 @@ if MCP_AVAILABLE:
async def _resolve_allowed_mcp_servers_for_tool_call(
user_api_key_dict: UserAPIKeyAuth,
server_id: str,
) -> List[MCPServer]:
) -> list[MCPServer]:
"""Resolve allowed MCP servers for the given user and validate server_id access."""
auth_contexts = await build_effective_auth_contexts(user_api_key_dict)
allowed_server_ids_set = set()
@ -545,7 +541,7 @@ if MCP_AVAILABLE:
"message": f"The key is not allowed to access server {server_id}",
},
)
allowed_mcp_servers: List[MCPServer] = []
allowed_mcp_servers: list[MCPServer] = []
for allowed_server_id in allowed_server_ids_set:
server = global_mcp_server_manager.get_mcp_server_by_id(allowed_server_id)
if server is not None:
@ -554,10 +550,10 @@ if MCP_AVAILABLE:
async def _list_tools_for_single_server(
server_id: str,
allowed_server_ids: List[str],
rest_client_ip: Optional[str],
allowed_server_ids: list[str],
rest_client_ip: str | None,
mcp_server_auth_headers: dict,
mcp_auth_header: Optional[str],
mcp_auth_header: str | None,
raw_headers_from_request: dict,
user_api_key_dict: UserAPIKeyAuth,
apply_tool_filters: bool = True,
@ -644,12 +640,12 @@ if MCP_AVAILABLE:
"message": "Successfully retrieved tools",
}
def _as_query_str(value: Any) -> Optional[str]:
def _as_query_str(value: Any) -> str | None:
"""Coerce an Optional[str] Query param to str|None, dropping unresolved FastAPI defaults."""
return value if isinstance(value, str) else None
async def _resolve_toolset_scope(
toolset_name: Optional[str],
toolset_name: str | None,
user_api_key_dict: UserAPIKeyAuth,
) -> UserAPIKeyAuth:
"""Resolve ``toolset_name`` to its scoped ``UserAPIKeyAuth``, or return unchanged."""
@ -670,11 +666,9 @@ if MCP_AVAILABLE:
@router.get("/tools/list", dependencies=[Depends(user_api_key_auth)])
async def list_tool_rest_api(
request: Request,
server_id: Optional[str] = Query(None, description="The server id to list tools for"),
mcp_server_name: Optional[str] = Query(
None, description="Filter tools to a single MCP server by name or alias"
),
toolset_name: Optional[str] = Query(None, description="Filter tools to a single toolset by name"),
server_id: str | None = Query(None, description="The server id to list tools for"),
mcp_server_name: str | None = Query(None, description="Filter tools to a single MCP server by name or alias"),
toolset_name: str | None = Query(None, description="Filter tools to a single toolset by name"),
include_disabled_tools: bool = Query(
False,
description=(
@ -981,7 +975,7 @@ if MCP_AVAILABLE:
) = await _resolve_allowed_mcp_servers_with_ip_filter(request, user_api_key_dict, server_id)
# Look up per-user OAuth headers for this server (mirrors list_tool_rest_api).
user_oauth_extra_headers: Optional[Dict[str, str]] = None
user_oauth_extra_headers: dict[str, str] | None = None
target_server = next(
(s for s in allowed_mcp_servers if s.server_id == canonical_server_id),
None,
@ -1094,18 +1088,18 @@ if MCP_AVAILABLE:
(client_id, client_secret, scopes) — any value may be ``None``.
"""
creds = request.credentials if isinstance(request.credentials, dict) else {}
client_id: Optional[str] = creds.get("client_id")
client_secret: Optional[str] = creds.get("client_secret")
client_id: str | None = creds.get("client_id")
client_secret: str | None = creds.get("client_secret")
scopes_raw = creds.get("scopes")
scopes: Optional[List[str]] = scopes_raw if isinstance(scopes_raw, list) else None
scopes: list[str] | None = scopes_raw if isinstance(scopes_raw, list) else None
return client_id, client_secret, scopes
async def _execute_with_mcp_client(
request: NewMCPServerRequest,
operation: Callable[..., Awaitable[Any]],
mcp_auth_header: Optional[Union[str, Dict[str, str]]] = None,
oauth2_headers: Optional[Dict[str, str]] = None,
raw_headers: Optional[Dict[str, str]] = None,
mcp_auth_header: str | dict[str, str] | None = None,
oauth2_headers: dict[str, str] | None = None,
raw_headers: dict[str, str] | None = None,
) -> dict:
"""
Create a temporary MCP client from *request*, run *operation*, and return the result.
@ -1128,7 +1122,7 @@ if MCP_AVAILABLE:
try:
client_id, client_secret, scopes = _extract_credentials(request)
_oauth2_flow: Optional[Literal["client_credentials", "authorization_code"]] = request.oauth2_flow or (
_oauth2_flow: Literal["client_credentials", "authorization_code"] | None = request.oauth2_flow or (
"client_credentials" if client_id and client_secret and request.token_url else None
)
# client_credentials requires token_url to fetch a token; without it the
@ -1244,7 +1238,7 @@ if MCP_AVAILABLE:
spec = await load_openapi_spec_async(spec_path)
paths = spec.get("paths", {})
components = spec.get("components", {})
tools: List[dict] = []
tools: list[dict] = []
used_names: set = set()
for path, path_item in paths.items():
for method in ("get", "post", "put", "delete", "patch"):
@ -1351,7 +1345,7 @@ if MCP_AVAILABLE:
headers = request.headers
mcp_auth_header: Optional[str] = None
mcp_auth_header: str | None = None
if new_mcp_server_request.auth_type in {
MCPAuth.api_key,
MCPAuth.bearer_token,
@ -1365,7 +1359,7 @@ if MCP_AVAILABLE:
# Authorization doubles as the admission fallback (LITELLM_API_KEY_HEADER_NAME_SECONDARY):
# when the primary x-litellm-api-key header is absent, the Authorization value is the
# caller's LiteLLM key, not an upstream token, and must never be forwarded upstream.
oauth2_headers: Optional[Dict[str, str]] = None
oauth2_headers: dict[str, str] | None = None
if new_mcp_server_request.auth_type in _UPSTREAM_OAUTH_DISCOVERY_AUTH_TYPES and headers.get(
MCPRequestHandler.LITELLM_API_KEY_HEADER_NAME_PRIMARY
):
@ -1376,8 +1370,8 @@ if MCP_AVAILABLE:
return await session.list_tools()
list_tools_response = await client.run_with_session(_list_tools_session_operation)
list_tools_result: List[MCPTool] = list_tools_response.tools
model_dumped_tools: List[dict] = [tool.model_dump() for tool in list_tools_result]
list_tools_result: list[MCPTool] = list_tools_response.tools
model_dumped_tools: list[dict] = [tool.model_dump() for tool in list_tools_result]
return {
"tools": model_dumped_tools,
"error": None,

684
litellm/proxy/_experimental/mcp_server/server.py
View file

File diff suppressed because it is too large Load diff

2
litellm/proxy/_experimental/out/404.html
View file

File diff suppressed because one or more lines are too long

2
litellm/proxy/_experimental/out/404/index.html
View file

File diff suppressed because one or more lines are too long

8
litellm/proxy/_experimental/out/__next.!KGRhc2hib2FyZCk.__PAGE__.txt
View file

@ -1,9 +1,9 @@
1:"$Sreact.fragment"
2:I[347257,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"ClientPageRoot"]
3:I[871135,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js","/litellm-asset-prefix/_next/static/chunks/0mqbd99.ej13v.js","/litellm-asset-prefix/_next/static/chunks/16ufy1iyybswo.js","/litellm-asset-prefix/_next/static/chunks/0ayum-x.hkww~.js","/litellm-asset-prefix/_next/static/chunks/15jl-1gcakfwa.js","/litellm-asset-prefix/_next/static/chunks/003_1s9xbht43.js","/litellm-asset-prefix/_next/static/chunks/0u.r3vzo30ofk.js","/litellm-asset-prefix/_next/static/chunks/0g4qcx-c9gsxn.js","/litellm-asset-prefix/_next/static/chunks/14a-un1blorp~.js","/litellm-asset-prefix/_next/static/chunks/02_q4881cz6h~.js","/litellm-asset-prefix/_next/static/chunks/0lg.6rbfsd-l9.js","/litellm-asset-prefix/_next/static/chunks/0wdlbe750tuzr.js","/litellm-asset-prefix/_next/static/chunks/043q3g5-5-aju.js","/litellm-asset-prefix/_next/static/chunks/0nht59ws0elww.js","/litellm-asset-prefix/_next/static/chunks/03zkt5iyjiqcz.js","/litellm-asset-prefix/_next/static/chunks/0l02mpo6za6ie.js","/litellm-asset-prefix/_next/static/chunks/1667t2pcy0iqm.js","/litellm-asset-prefix/_next/static/chunks/0-_m4km7b1~oe.js","/litellm-asset-prefix/_next/static/chunks/0sx3mu2_l9g_y.js","/litellm-asset-prefix/_next/static/chunks/0sqw622fcvsv4.js","/litellm-asset-prefix/_next/static/chunks/0t8el_ijoskx..js","/litellm-asset-prefix/_next/static/chunks/0el08tticy_20.js","/litellm-asset-prefix/_next/static/chunks/0a.ljputcx8g5.js","/litellm-asset-prefix/_next/static/chunks/02-u6qtmsnqn0.js","/litellm-asset-prefix/_next/static/chunks/076.vm.7w-x2..js","/litellm-asset-prefix/_next/static/chunks/069dx5~5osue0.js","/litellm-asset-prefix/_next/static/chunks/0onea0n77pqw1.js","/litellm-asset-prefix/_next/static/chunks/16jfov1k2wrj0.js","/litellm-asset-prefix/_next/static/chunks/0l7em-5kjv49e.js","/litellm-asset-prefix/_next/static/chunks/0mh1wnrvmv_y7.js","/litellm-asset-prefix/_next/static/chunks/01reddhq423_f.js","/litellm-asset-prefix/_next/static/chunks/0h80lrrstjswl.js","/litellm-asset-prefix/_next/static/chunks/0cmepm.jkel-i.js","/litellm-asset-prefix/_next/static/chunks/0q6~n4y84cejn.js","/litellm-asset-prefix/_next/static/chunks/04jv9e6~9vi.l.js","/litellm-asset-prefix/_next/static/chunks/0d_sm.._5mw-p.js","/litellm-asset-prefix/_next/static/chunks/0elr0ye86.44-.js"],"default"]
6:I[897367,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"OutletBoundary"]
2:I[347257,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"ClientPageRoot"]
3:I[871135,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js","/litellm-asset-prefix/_next/static/chunks/1ioy8obpggx93.js","/litellm-asset-prefix/_next/static/chunks/3_3dj4vdy-3xy.js","/litellm-asset-prefix/_next/static/chunks/1zr7rrk4wkmju.js","/litellm-asset-prefix/_next/static/chunks/2cngn5bal3278.js","/litellm-asset-prefix/_next/static/chunks/2up3bks93iqds.js","/litellm-asset-prefix/_next/static/chunks/0fk0i3e2aixp7.js","/litellm-asset-prefix/_next/static/chunks/0zduf1gntl_f8.js","/litellm-asset-prefix/_next/static/chunks/0dbvgsc7ha049.js","/litellm-asset-prefix/_next/static/chunks/0g_w4tf2inv3i.js","/litellm-asset-prefix/_next/static/chunks/1vquuz09jxl5_.js","/litellm-asset-prefix/_next/static/chunks/17-6zku8f68gf.js","/litellm-asset-prefix/_next/static/chunks/1iakmimqrlpn0.js","/litellm-asset-prefix/_next/static/chunks/1a0bgy7kzrj91.js","/litellm-asset-prefix/_next/static/chunks/0f5fel02jwglw.js","/litellm-asset-prefix/_next/static/chunks/0g8wwba6umbim.js","/litellm-asset-prefix/_next/static/chunks/1di-caw05k3tq.js","/litellm-asset-prefix/_next/static/chunks/00g6xfr4yow7h.js","/litellm-asset-prefix/_next/static/chunks/2kcxwg1mpncp6.js","/litellm-asset-prefix/_next/static/chunks/1fmx49l6q8v39.js","/litellm-asset-prefix/_next/static/chunks/0ww76lz_0cphv.js","/litellm-asset-prefix/_next/static/chunks/2uc2pi4ob086w.js","/litellm-asset-prefix/_next/static/chunks/2hu1vyy-5pv13.js","/litellm-asset-prefix/_next/static/chunks/2l25bmiiw9ixp.js","/litellm-asset-prefix/_next/static/chunks/1uz3jt-tj9lkf.js","/litellm-asset-prefix/_next/static/chunks/3drq2_k-jeio2.js","/litellm-asset-prefix/_next/static/chunks/0dsiq_ok1yngk.js","/litellm-asset-prefix/_next/static/chunks/3srzg1la93pwv.js","/litellm-asset-prefix/_next/static/chunks/17nqbxvhztf3k.js","/litellm-asset-prefix/_next/static/chunks/2c90xukbd3il6.js","/litellm-asset-prefix/_next/static/chunks/0kap_rdm2-lem.js","/litellm-asset-prefix/_next/static/chunks/09l_m9l1emin2.js","/litellm-asset-prefix/_next/static/chunks/1cea03gg5a_c7.js","/litellm-asset-prefix/_next/static/chunks/323l6h8s7ahat.js","/litellm-asset-prefix/_next/static/chunks/112n0hv3cc2rg.js","/litellm-asset-prefix/_next/static/chunks/199uwr871eene.js","/litellm-asset-prefix/_next/static/chunks/105643dvf00hu.js","/litellm-asset-prefix/_next/static/chunks/12wsfsljxg4xv.js","/litellm-asset-prefix/_next/static/chunks/22iools_e0k44.js","/litellm-asset-prefix/_next/static/chunks/0am68mi9t9cb6.js","/litellm-asset-prefix/_next/static/chunks/23-g73xaw3kap.js"],"default"]
6:I[897367,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"OutletBoundary"]
7:"$Sreact.suspense"
0:{"rsc":["$","$1","c",{"children":[["$","$L2",null,{"Component":"$3","serverProvidedParams":{"searchParams":{},"params":{},"promises":["$@4","$@5"]}}],[["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/1667t2pcy0iqm.js","async":true}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/0-_m4km7b1~oe.js","async":true}],["$","script","script-2",{"src":"/litellm-asset-prefix/_next/static/chunks/0sx3mu2_l9g_y.js","async":true}],["$","script","script-3",{"src":"/litellm-asset-prefix/_next/static/chunks/0sqw622fcvsv4.js","async":true}],["$","script","script-4",{"src":"/litellm-asset-prefix/_next/static/chunks/0t8el_ijoskx..js","async":true}],["$","script","script-5",{"src":"/litellm-asset-prefix/_next/static/chunks/0el08tticy_20.js","async":true}],["$","script","script-6",{"src":"/litellm-asset-prefix/_next/static/chunks/0a.ljputcx8g5.js","async":true}],["$","script","script-7",{"src":"/litellm-asset-prefix/_next/static/chunks/02-u6qtmsnqn0.js","async":true}],["$","script","script-8",{"src":"/litellm-asset-prefix/_next/static/chunks/076.vm.7w-x2..js","async":true}],["$","script","script-9",{"src":"/litellm-asset-prefix/_next/static/chunks/069dx5~5osue0.js","async":true}],["$","script","script-10",{"src":"/litellm-asset-prefix/_next/static/chunks/0onea0n77pqw1.js","async":true}],["$","script","script-11",{"src":"/litellm-asset-prefix/_next/static/chunks/16jfov1k2wrj0.js","async":true}],["$","script","script-12",{"src":"/litellm-asset-prefix/_next/static/chunks/0l7em-5kjv49e.js","async":true}],["$","script","script-13",{"src":"/litellm-asset-prefix/_next/static/chunks/0mh1wnrvmv_y7.js","async":true}],["$","script","script-14",{"src":"/litellm-asset-prefix/_next/static/chunks/01reddhq423_f.js","async":true}],["$","script","script-15",{"src":"/litellm-asset-prefix/_next/static/chunks/0h80lrrstjswl.js","async":true}],["$","script","script-16",{"src":"/litellm-asset-prefix/_next/static/chunks/0cmepm.jkel-i.js","async":true}],["$","script","script-17",{"src":"/litellm-asset-prefix/_next/static/chunks/0q6~n4y84cejn.js","async":true}],["$","script","script-18",{"src":"/litellm-asset-prefix/_next/static/chunks/04jv9e6~9vi.l.js","async":true}],["$","script","script-19",{"src":"/litellm-asset-prefix/_next/static/chunks/0d_sm.._5mw-p.js","async":true}],["$","script","script-20",{"src":"/litellm-asset-prefix/_next/static/chunks/0elr0ye86.44-.js","async":true}]],["$","$L6",null,{"children":["$","$7",null,{"name":"Next.MetadataOutlet","children":"$@8"}]}]]}],"isPartial":false,"staleTime":300,"varyParams":null,"buildId":"DSHomUr6Sq46Bm2WLdUas"}
0:{"rsc":["$","$1","c",{"children":[["$","$L2",null,{"Component":"$3","serverProvidedParams":{"searchParams":{},"params":{},"promises":["$@4","$@5"]}}],[["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/00g6xfr4yow7h.js","async":true}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/2kcxwg1mpncp6.js","async":true}],["$","script","script-2",{"src":"/litellm-asset-prefix/_next/static/chunks/1fmx49l6q8v39.js","async":true}],["$","script","script-3",{"src":"/litellm-asset-prefix/_next/static/chunks/0ww76lz_0cphv.js","async":true}],["$","script","script-4",{"src":"/litellm-asset-prefix/_next/static/chunks/2uc2pi4ob086w.js","async":true}],["$","script","script-5",{"src":"/litellm-asset-prefix/_next/static/chunks/2hu1vyy-5pv13.js","async":true}],["$","script","script-6",{"src":"/litellm-asset-prefix/_next/static/chunks/2l25bmiiw9ixp.js","async":true}],["$","script","script-7",{"src":"/litellm-asset-prefix/_next/static/chunks/1uz3jt-tj9lkf.js","async":true}],["$","script","script-8",{"src":"/litellm-asset-prefix/_next/static/chunks/3drq2_k-jeio2.js","async":true}],["$","script","script-9",{"src":"/litellm-asset-prefix/_next/static/chunks/0dsiq_ok1yngk.js","async":true}],["$","script","script-10",{"src":"/litellm-asset-prefix/_next/static/chunks/3srzg1la93pwv.js","async":true}],["$","script","script-11",{"src":"/litellm-asset-prefix/_next/static/chunks/17nqbxvhztf3k.js","async":true}],["$","script","script-12",{"src":"/litellm-asset-prefix/_next/static/chunks/2c90xukbd3il6.js","async":true}],["$","script","script-13",{"src":"/litellm-asset-prefix/_next/static/chunks/0kap_rdm2-lem.js","async":true}],["$","script","script-14",{"src":"/litellm-asset-prefix/_next/static/chunks/09l_m9l1emin2.js","async":true}],["$","script","script-15",{"src":"/litellm-asset-prefix/_next/static/chunks/1cea03gg5a_c7.js","async":true}],["$","script","script-16",{"src":"/litellm-asset-prefix/_next/static/chunks/323l6h8s7ahat.js","async":true}],["$","script","script-17",{"src":"/litellm-asset-prefix/_next/static/chunks/112n0hv3cc2rg.js","async":true}],["$","script","script-18",{"src":"/litellm-asset-prefix/_next/static/chunks/199uwr871eene.js","async":true}],["$","script","script-19",{"src":"/litellm-asset-prefix/_next/static/chunks/105643dvf00hu.js","async":true}],["$","script","script-20",{"src":"/litellm-asset-prefix/_next/static/chunks/12wsfsljxg4xv.js","async":true}],["$","script","script-21",{"src":"/litellm-asset-prefix/_next/static/chunks/22iools_e0k44.js","async":true}],["$","script","script-22",{"src":"/litellm-asset-prefix/_next/static/chunks/0am68mi9t9cb6.js","async":true}],["$","script","script-23",{"src":"/litellm-asset-prefix/_next/static/chunks/23-g73xaw3kap.js","async":true}]],["$","$L6",null,{"children":["$","$7",null,{"name":"Next.MetadataOutlet","children":"$@8"}]}]]}],"isPartial":false,"staleTime":300,"varyParams":null,"buildId":"qXutWsQW5C1Pf62WxTkEI"}
4:{}
5:"$0:rsc:props:children:0:props:serverProvidedParams:params"
8:null

10
litellm/proxy/_experimental/out/__next.!KGRhc2hib2FyZCk.txt
View file

@ -1,7 +1,7 @@
1:"$Sreact.fragment"
2:I[92825,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"ClientSegmentRoot"]
3:I[216370,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js","/litellm-asset-prefix/_next/static/chunks/0mqbd99.ej13v.js","/litellm-asset-prefix/_next/static/chunks/16ufy1iyybswo.js","/litellm-asset-prefix/_next/static/chunks/0ayum-x.hkww~.js","/litellm-asset-prefix/_next/static/chunks/15jl-1gcakfwa.js","/litellm-asset-prefix/_next/static/chunks/003_1s9xbht43.js","/litellm-asset-prefix/_next/static/chunks/0u.r3vzo30ofk.js","/litellm-asset-prefix/_next/static/chunks/0g4qcx-c9gsxn.js","/litellm-asset-prefix/_next/static/chunks/14a-un1blorp~.js","/litellm-asset-prefix/_next/static/chunks/02_q4881cz6h~.js","/litellm-asset-prefix/_next/static/chunks/0lg.6rbfsd-l9.js","/litellm-asset-prefix/_next/static/chunks/0wdlbe750tuzr.js","/litellm-asset-prefix/_next/static/chunks/043q3g5-5-aju.js","/litellm-asset-prefix/_next/static/chunks/0nht59ws0elww.js","/litellm-asset-prefix/_next/static/chunks/03zkt5iyjiqcz.js","/litellm-asset-prefix/_next/static/chunks/0l02mpo6za6ie.js"],"default"]
4:I[339756,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"default"]
5:I[837457,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"default"]
0:{"rsc":["$","$1","c",{"children":[[["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/0mqbd99.ej13v.js","async":true}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/16ufy1iyybswo.js","async":true}],["$","script","script-2",{"src":"/litellm-asset-prefix/_next/static/chunks/0ayum-x.hkww~.js","async":true}],["$","script","script-3",{"src":"/litellm-asset-prefix/_next/static/chunks/15jl-1gcakfwa.js","async":true}],["$","script","script-4",{"src":"/litellm-asset-prefix/_next/static/chunks/003_1s9xbht43.js","async":true}],["$","script","script-5",{"src":"/litellm-asset-prefix/_next/static/chunks/0u.r3vzo30ofk.js","async":true}],["$","script","script-6",{"src":"/litellm-asset-prefix/_next/static/chunks/0g4qcx-c9gsxn.js","async":true}],["$","script","script-7",{"src":"/litellm-asset-prefix/_next/static/chunks/14a-un1blorp~.js","async":true}],["$","script","script-8",{"src":"/litellm-asset-prefix/_next/static/chunks/02_q4881cz6h~.js","async":true}],["$","script","script-9",{"src":"/litellm-asset-prefix/_next/static/chunks/0lg.6rbfsd-l9.js","async":true}],["$","script","script-10",{"src":"/litellm-asset-prefix/_next/static/chunks/0wdlbe750tuzr.js","async":true}],["$","script","script-11",{"src":"/litellm-asset-prefix/_next/static/chunks/043q3g5-5-aju.js","async":true}],["$","script","script-12",{"src":"/litellm-asset-prefix/_next/static/chunks/0nht59ws0elww.js","async":true}],["$","script","script-13",{"src":"/litellm-asset-prefix/_next/static/chunks/03zkt5iyjiqcz.js","async":true}],["$","script","script-14",{"src":"/litellm-asset-prefix/_next/static/chunks/0l02mpo6za6ie.js","async":true}]],["$","$L2",null,{"Component":"$3","slots":{"children":["$","$L4",null,{"parallelRouterKey":"children","template":["$","$L5",null,{}],"notFound":[[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":404}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],[]]}]},"serverProvidedParams":{"params":{},"promises":["$@6"]}}]]}],"isPartial":false,"staleTime":300,"varyParams":null,"buildId":"DSHomUr6Sq46Bm2WLdUas"}
2:I[92825,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"ClientSegmentRoot"]
3:I[216370,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js","/litellm-asset-prefix/_next/static/chunks/1ioy8obpggx93.js","/litellm-asset-prefix/_next/static/chunks/3_3dj4vdy-3xy.js","/litellm-asset-prefix/_next/static/chunks/1zr7rrk4wkmju.js","/litellm-asset-prefix/_next/static/chunks/2cngn5bal3278.js","/litellm-asset-prefix/_next/static/chunks/2up3bks93iqds.js","/litellm-asset-prefix/_next/static/chunks/0fk0i3e2aixp7.js","/litellm-asset-prefix/_next/static/chunks/0zduf1gntl_f8.js","/litellm-asset-prefix/_next/static/chunks/0dbvgsc7ha049.js","/litellm-asset-prefix/_next/static/chunks/0g_w4tf2inv3i.js","/litellm-asset-prefix/_next/static/chunks/1vquuz09jxl5_.js","/litellm-asset-prefix/_next/static/chunks/17-6zku8f68gf.js","/litellm-asset-prefix/_next/static/chunks/1iakmimqrlpn0.js","/litellm-asset-prefix/_next/static/chunks/1a0bgy7kzrj91.js","/litellm-asset-prefix/_next/static/chunks/0f5fel02jwglw.js","/litellm-asset-prefix/_next/static/chunks/0g8wwba6umbim.js","/litellm-asset-prefix/_next/static/chunks/1di-caw05k3tq.js"],"default"]
4:I[339756,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"default"]
5:I[837457,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"default"]
0:{"rsc":["$","$1","c",{"children":[[["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/1ioy8obpggx93.js","async":true}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/3_3dj4vdy-3xy.js","async":true}],["$","script","script-2",{"src":"/litellm-asset-prefix/_next/static/chunks/1zr7rrk4wkmju.js","async":true}],["$","script","script-3",{"src":"/litellm-asset-prefix/_next/static/chunks/2cngn5bal3278.js","async":true}],["$","script","script-4",{"src":"/litellm-asset-prefix/_next/static/chunks/2up3bks93iqds.js","async":true}],["$","script","script-5",{"src":"/litellm-asset-prefix/_next/static/chunks/0fk0i3e2aixp7.js","async":true}],["$","script","script-6",{"src":"/litellm-asset-prefix/_next/static/chunks/0zduf1gntl_f8.js","async":true}],["$","script","script-7",{"src":"/litellm-asset-prefix/_next/static/chunks/0dbvgsc7ha049.js","async":true}],["$","script","script-8",{"src":"/litellm-asset-prefix/_next/static/chunks/0g_w4tf2inv3i.js","async":true}],["$","script","script-9",{"src":"/litellm-asset-prefix/_next/static/chunks/1vquuz09jxl5_.js","async":true}],["$","script","script-10",{"src":"/litellm-asset-prefix/_next/static/chunks/17-6zku8f68gf.js","async":true}],["$","script","script-11",{"src":"/litellm-asset-prefix/_next/static/chunks/1iakmimqrlpn0.js","async":true}],["$","script","script-12",{"src":"/litellm-asset-prefix/_next/static/chunks/1a0bgy7kzrj91.js","async":true}],["$","script","script-13",{"src":"/litellm-asset-prefix/_next/static/chunks/0f5fel02jwglw.js","async":true}],["$","script","script-14",{"src":"/litellm-asset-prefix/_next/static/chunks/0g8wwba6umbim.js","async":true}],["$","script","script-15",{"src":"/litellm-asset-prefix/_next/static/chunks/1di-caw05k3tq.js","async":true}]],["$","$L2",null,{"Component":"$3","slots":{"children":["$","$L4",null,{"parallelRouterKey":"children","template":["$","$L5",null,{}],"notFound":[[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":404}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],[]]}]},"serverProvidedParams":{"params":{},"promises":["$@6"]}}]]}],"isPartial":false,"staleTime":300,"varyParams":null,"buildId":"qXutWsQW5C1Pf62WxTkEI"}
6:"$0:rsc:props:children:1:props:serverProvidedParams:params"

61
litellm/proxy/_experimental/out/__next._full.txt
View file

File diff suppressed because one or more lines are too long

8
litellm/proxy/_experimental/out/__next._head.txt
View file

@ -1,6 +1,6 @@
1:"$Sreact.fragment"
2:I[897367,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"ViewportBoundary"]
3:I[897367,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"MetadataBoundary"]
2:I[897367,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"ViewportBoundary"]
3:I[897367,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"MetadataBoundary"]
4:"$Sreact.suspense"
5:I[27201,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"IconMark"]
0:{"rsc":["$","$1","h",{"children":[null,["$","$L2",null,{"children":[["$","meta","0",{"charSet":"utf-8"}],["$","meta","1",{"name":"viewport","content":"width=device-width, initial-scale=1"}]]}],["$","div",null,{"hidden":true,"children":["$","$L3",null,{"children":["$","$4",null,{"name":"Next.Metadata","children":[["$","title","0",{"children":"LiteLLM Dashboard"}],["$","meta","1",{"name":"description","content":"LiteLLM Proxy Admin UI"}],["$","link","2",{"rel":"icon","href":"/favicon.ico?favicon.0~dgapwhi~75y.ico","sizes":"48x48","type":"image/x-icon"}],["$","link","3",{"rel":"icon","href":"/get_favicon"}],["$","$L5","4",{}]]}]}]}],["$","meta",null,{"name":"next-size-adjust","content":""}]]}],"isPartial":false,"staleTime":300,"varyParams":null,"buildId":"DSHomUr6Sq46Bm2WLdUas"}
5:I[27201,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"IconMark"]
0:{"rsc":["$","$1","h",{"children":[null,["$","$L2",null,{"children":[["$","meta","0",{"charSet":"utf-8"}],["$","meta","1",{"name":"viewport","content":"width=device-width, initial-scale=1"}]]}],["$","div",null,{"hidden":true,"children":["$","$L3",null,{"children":["$","$4",null,{"name":"Next.Metadata","children":[["$","title","0",{"children":"LiteLLM Dashboard"}],["$","meta","1",{"name":"description","content":"LiteLLM Proxy Admin UI"}],["$","link","2",{"rel":"icon","href":"/favicon.ico?favicon.3arlap5n8tyzg.ico","sizes":"48x48","type":"image/x-icon"}],["$","link","3",{"rel":"icon","href":"/get_favicon"}],["$","$L5","4",{}]]}]}]}],["$","meta",null,{"name":"next-size-adjust","content":""}]]}],"isPartial":false,"staleTime":300,"varyParams":null,"buildId":"qXutWsQW5C1Pf62WxTkEI"}

16
litellm/proxy/_experimental/out/__next._index.txt
View file

@ -1,9 +1,9 @@
1:"$Sreact.fragment"
2:I[867271,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"default"]
3:I[71195,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"default"]
4:I[557951,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"AuthProvider"]
5:I[339756,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"default"]
6:I[837457,["/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js"],"default"]
:HL["/litellm-asset-prefix/_next/static/chunks/05qmwjqau64bz.css","style"]
:HL["/litellm-asset-prefix/_next/static/chunks/0aapv6n5bztwf.css","style"]
0:{"rsc":["$","$1","c",{"children":[[["$","link","0",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/05qmwjqau64bz.css","precedence":"next"}],["$","link","1",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/0aapv6n5bztwf.css","precedence":"next"}],["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/09qhs_ev_bxr8.js","async":true}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/0l6q6-77u4dy~.js","async":true}],["$","script","script-2",{"src":"/litellm-asset-prefix/_next/static/chunks/0hpxif-db_y5-.js","async":true}]],["$","html",null,{"lang":"en","children":["$","body",null,{"className":"inter_5972bc34-module__OU16Qa__className","children":["$","$L2",null,{"children":["$","$L3",null,{"children":["$","$L4",null,{"children":["$","$L5",null,{"parallelRouterKey":"children","template":["$","$L6",null,{}],"notFound":[[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":404}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],[]]}]}]}]}]}]}]]}],"isPartial":false,"staleTime":300,"varyParams":null,"buildId":"DSHomUr6Sq46Bm2WLdUas"}
2:I[867271,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"default"]
3:I[71195,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"default"]
4:I[557951,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"AuthProvider"]
5:I[339756,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"default"]
6:I[837457,["/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js"],"default"]
:HL["/litellm-asset-prefix/_next/static/chunks/1kid9zr1--h6y.css","style"]
:HL["/litellm-asset-prefix/_next/static/chunks/3254j4ut19q6_.css","style"]
0:{"rsc":["$","$1","c",{"children":[[["$","link","0",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/1kid9zr1--h6y.css","precedence":"next"}],["$","link","1",{"rel":"stylesheet","href":"/litellm-asset-prefix/_next/static/chunks/3254j4ut19q6_.css","precedence":"next"}],["$","script","script-0",{"src":"/litellm-asset-prefix/_next/static/chunks/0vggytdohwe7o.js","async":true}],["$","script","script-1",{"src":"/litellm-asset-prefix/_next/static/chunks/3c02m_kr-u94p.js","async":true}],["$","script","script-2",{"src":"/litellm-asset-prefix/_next/static/chunks/1jfookxfajkeo.js","async":true}]],["$","html",null,{"lang":"en","children":["$","body",null,{"className":"inter_5972bc34-module__OU16Qa__className","children":["$","$L2",null,{"children":["$","$L3",null,{"children":["$","$L4",null,{"children":["$","$L5",null,{"parallelRouterKey":"children","template":["$","$L6",null,{}],"notFound":[[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":404}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],[]]}]}]}]}]}]}]]}],"isPartial":false,"staleTime":300,"varyParams":null,"buildId":"qXutWsQW5C1Pf62WxTkEI"}

8
litellm/proxy/_experimental/out/__next._tree.txt
View file

@ -1,4 +1,4 @@
:HL["/litellm-asset-prefix/_next/static/chunks/05qmwjqau64bz.css","style"]
:HL["/litellm-asset-prefix/_next/static/chunks/0aapv6n5bztwf.css","style"]
:HL["/litellm-asset-prefix/_next/static/media/83afe278b6a6bb3c-s.p.0q-301v4kxxnr.woff2","font",{"crossOrigin":"","type":"font/woff2"}]
0:{"tree":{"name":"","param":null,"prefetchHints":16,"slots":{"children":{"name":"(dashboard)","param":null,"prefetchHints":0,"slots":{"children":{"name":"__PAGE__","param":null,"prefetchHints":0,"slots":null}}}}},"staleTime":300,"buildId":"DSHomUr6Sq46Bm2WLdUas"}
:HL["/litellm-asset-prefix/_next/static/chunks/1kid9zr1--h6y.css","style"]
:HL["/litellm-asset-prefix/_next/static/chunks/3254j4ut19q6_.css","style"]
:HL["/litellm-asset-prefix/_next/static/media/83afe278b6a6bb3c-s.p.2bn3s6zvc0dyp.woff2","font",{"crossOrigin":"","type":"font/woff2"}]
0:{"tree":{"name":"","param":null,"prefetchHints":16,"slots":{"children":{"name":"(dashboard)","param":null,"prefetchHints":0,"slots":{"children":{"name":"__PAGE__","param":null,"prefetchHints":0,"slots":null}}}}},"staleTime":300,"buildId":"qXutWsQW5C1Pf62WxTkEI"}

1
litellm/proxy/_experimental/out/_next/static/chunks/0-_m4km7b1~oe.js
View file

File diff suppressed because one or more lines are too long

8
litellm/proxy/_experimental/out/_next/static/chunks/0-ahu72ndvhwn.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/0.3q2b74j~ty5.js
View file

File diff suppressed because one or more lines are too long

2
litellm/proxy/_experimental/out/_next/static/chunks/0.mwuwep0859t.js
View file

File diff suppressed because one or more lines are too long

10
litellm/proxy/_experimental/out/_next/static/chunks/0.p~s6ih~c~xe.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/00-dyuivh_bf-.js Normal file
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/003_1s9xbht43.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/003w1n3_ylv_2.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/007c8g8hmd9qz.js Normal file
View file

File diff suppressed because one or more lines are too long

8
litellm/proxy/_experimental/out/_next/static/chunks/00ccjtnk99zr7.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/00g6xfr4yow7h.js Normal file
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/00qiry~y.broe.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/00tczcrtv5upo.js
View file

File diff suppressed because one or more lines are too long

8
litellm/proxy/_experimental/out/_next/static/chunks/00zxtugv201bq.js
View file

File diff suppressed because one or more lines are too long

16
litellm/proxy/_experimental/out/_next/static/chunks/0175usbyz91lt.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/017kxo-8o84bv.js Normal file
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/01hy_w_4bnb34.js
View file

File diff suppressed because one or more lines are too long

8
litellm/proxy/_experimental/out/_next/static/chunks/01ozl298h03bw.js
View file

File diff suppressed because one or more lines are too long

17
litellm/proxy/_experimental/out/_next/static/chunks/01reddhq423_f.js
View file

File diff suppressed because one or more lines are too long

4
litellm/proxy/_experimental/out/_next/static/chunks/01y._o853f7le.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/01yk5y7rumzgt.js
View file

File diff suppressed because one or more lines are too long

4
litellm/proxy/_experimental/out/_next/static/chunks/022.sz94ycw4x.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/023jsye4cz4a7.js
View file

File diff suppressed because one or more lines are too long

7
litellm/proxy/_experimental/out/_next/static/chunks/025ocjcb8e481.js Normal file
View file

File diff suppressed because one or more lines are too long

2
litellm/proxy/_experimental/out/_next/static/chunks/026n9mracjd5k.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/027d2u2cl335o.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/02_q4881cz6h~.js
View file

File diff suppressed because one or more lines are too long

2
litellm/proxy/_experimental/out/_next/static/chunks/02nioff5-e.ez.js
View file

File diff suppressed because one or more lines are too long

17
litellm/proxy/_experimental/out/_next/static/chunks/02ucg1k1-nq5m.js Normal file
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/02wxbd2ona7u_.js
View file

File diff suppressed because one or more lines are too long

68
litellm/proxy/_experimental/out/_next/static/chunks/030xj-a9q0ur8.js Normal file
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/0337vg5sc7rt~.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/0369tkoo6z4yx.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/036yal3~xlgjh.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/03oh9wvqpsr-g.js
View file

File diff suppressed because one or more lines are too long

31
litellm/proxy/_experimental/out/_next/static/chunks/03sdszpwi459j.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/03zkt5iyjiqcz.js
View file

File diff suppressed because one or more lines are too long

56
litellm/proxy/_experimental/out/_next/static/chunks/04.hopkzyt7jd.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/046-gw19n7owc.js
View file

File diff suppressed because one or more lines are too long

1
litellm/proxy/_experimental/out/_next/static/chunks/046q5lwe95zp6.js Normal file
View file

File diff suppressed because one or more lines are too long

8
litellm/proxy/_experimental/out/_next/static/chunks/04_xp3aju8b3x.js
View file

File diff suppressed because one or more lines are too long

Some files were not shown because too many files have changed in this diff Show more