scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-13 23:11:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 101

Merge branch 'BerriAI:main' into main

Browse source
This commit is contained in:
Esteban Zeller 2026-03-04 11:24:59 -03:00 • committed by Esteban
commit 9dcc1127b2
250 changed files with 20813 additions and 2624 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

114
.circleci/config.yml
View file

@ -3689,6 +3689,114 @@ jobs:
- store_test_results:
path: test-results
proxy_e2e_azure_batches_tests:
machine:
image: ubuntu-2204:2023.10.1
resource_class: xlarge
working_directory: ~/project
steps:
- checkout
- setup_google_dns
- run:
name: Install Docker CLI
command: |
curl -fsSL https://get.docker.com | sh
sudo usermod -aG docker $USER
docker version
- run:
name: Install Python 3.12
command: |
curl https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh --output miniconda.sh
bash miniconda.sh -b -p $HOME/miniconda
export PATH="$HOME/miniconda/bin:$PATH"
conda init bash
source ~/.bashrc
conda create -n myenv python=3.12 -y
conda activate myenv
python --version
- run:
name: Install Poetry
command: |
export PATH="$HOME/miniconda/bin:$PATH"
source $HOME/miniconda/etc/profile.d/conda.sh
conda activate myenv
pip install poetry
- run:
name: Install dockerize
command: |
wget https://github.com/jwilder/dockerize/releases/download/v0.6.1/dockerize-linux-amd64-v0.6.1.tar.gz
sudo tar -C /usr/local/bin -xzvf dockerize-linux-amd64-v0.6.1.tar.gz
rm dockerize-linux-amd64-v0.6.1.tar.gz
- run:
name: Start PostgreSQL Database
command: |
docker run -d \
--name postgres-db \
-e POSTGRES_USER=llmproxy \
-e POSTGRES_PASSWORD=dbpassword9090 \
-e POSTGRES_DB=litellm \
-p 5432:5432 \
postgres:15
- run:
name: Wait for PostgreSQL to be ready
command: dockerize -wait tcp://localhost:5432 -timeout 1m
- run:
name: Install system dependencies
command: |
sudo apt-get update -y
sudo apt-get install -y libpq-dev
- run:
name: Install Dependencies
command: |
export PATH="$HOME/miniconda/bin:$PATH"
source $HOME/miniconda/etc/profile.d/conda.sh
conda activate myenv
poetry config virtualenvs.in-project true
poetry install --with dev,proxy-dev --extras "proxy"
poetry run pip install psycopg2-binary uvicorn fastapi httpx tenacity
- run:
name: Setup litellm-enterprise
command: |
export PATH="$HOME/miniconda/bin:$PATH"
source $HOME/miniconda/etc/profile.d/conda.sh
conda activate myenv
poetry run pip install --force-reinstall --no-deps -e enterprise/
- run:
name: Generate Prisma client
command: |
export PATH="$HOME/miniconda/bin:$PATH"
source $HOME/miniconda/etc/profile.d/conda.sh
conda activate myenv
poetry run prisma generate --schema litellm/proxy/schema.prisma
- run:
name: Run Prisma migrations
command: |
export PATH="$HOME/miniconda/bin:$PATH"
source $HOME/miniconda/etc/profile.d/conda.sh
conda activate myenv
export DATABASE_URL=postgresql://llmproxy:dbpassword9090@localhost:5432/litellm
cd litellm/proxy
poetry run prisma migrate deploy --schema schema.prisma
cd ../..
- run:
name: Run Azure Batch E2E Tests
command: |
export PATH="$HOME/miniconda/bin:$PATH"
source $HOME/miniconda/etc/profile.d/conda.sh
conda activate myenv
export DATABASE_URL=postgresql://llmproxy:dbpassword9090@localhost:5432/litellm
export USE_LOCAL_LITELLM=true
export USE_MOCK_MODELS=true
export USE_STATE_TRACKER=true
export LITELLM_LOG=DEBUG
poetry run pytest tests/proxy_e2e_azure_batches_tests/test_proxy_e2e_azure_batches.py \
-vv -s -k "test_e2e_managed_batch" \
--tb=short \
--maxfail=3 \
--durations=10 \
--junitxml=test-results/junit.xml
no_output_timeout: 30m
upload-coverage:
docker:
- image: cimg/python:3.9
@ -4458,6 +4566,12 @@ workflows:
only:
- main
- /litellm_.*/
- proxy_e2e_azure_batches_tests:
filters:
branches:
only:
- main
- /litellm_.*/
- llm_translation_testing:
filters:
branches:

2
.github/workflows/test-litellm.yml vendored
View file

@ -38,7 +38,7 @@ jobs:
poetry run pip install "google-genai==1.22.0"
poetry run pip install "google-cloud-aiplatform>=1.38"
poetry run pip install "fastapi-offline==1.7.3"
poetry run pip install "python-multipart==0.0.22"
poetry run pip install "python-multipart>=0.0.20"
poetry run pip install "openapi-core"
- name: Setup litellm-enterprise as local package
run: |

90
.github/workflows/test-proxy-e2e-azure-batches.yml vendored Normal file
View file

@ -0,0 +1,90 @@
name: Proxy E2E Azure Batches Tests
on:
pull_request:
branches: [main]
workflow_dispatch:
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
cancel-in-progress: true
jobs:
proxy_e2e_azure_batches_tests:
runs-on: ubuntu-latest
timeout-minutes: 30
services:
postgres:
image: postgres:15
env:
POSTGRES_USER: llmproxy
POSTGRES_PASSWORD: dbpassword9090
POSTGRES_DB: litellm
ports:
- 5432:5432
options: >-
--health-cmd pg_isready
--health-interval 10s
--health-timeout 5s
--health-retries 5
steps:
- uses: actions/checkout@v4
- name: Set up Python
uses: actions/setup-python@v5
with:
python-version: "3.12"
- name: Install Poetry
uses: snok/install-poetry@v1
- name: Cache Poetry dependencies
uses: actions/cache@v4
with:
path: |
~/.cache/pypoetry
~/.cache/pip
.venv
key: ${{ runner.os }}-poetry-e2e-batches-${{ hashFiles('poetry.lock') }}
restore-keys: |
${{ runner.os }}-poetry-e2e-batches-
${{ runner.os }}-poetry-
- name: Install dependencies
run: |
poetry config virtualenvs.in-project true
poetry install --with dev,proxy-dev --extras "proxy"
poetry run pip install psycopg2-binary uvicorn fastapi httpx
- name: Setup litellm-enterprise
run: |
poetry run pip install --force-reinstall --no-deps -e enterprise/
- name: Generate Prisma client
run: |
poetry run prisma generate --schema litellm/proxy/schema.prisma
- name: Run Prisma migrations
env:
DATABASE_URL: postgresql://llmproxy:dbpassword9090@localhost:5432/litellm
run: |
cd litellm/proxy
poetry run prisma migrate deploy --schema schema.prisma
cd ../..
- name: Run Azure Batch E2E Tests
env:
DATABASE_URL: postgresql://llmproxy:dbpassword9090@localhost:5432/litellm
USE_LOCAL_LITELLM: "true"
USE_MOCK_MODELS: "true"
USE_STATE_TRACKER: "true"
LITELLM_LOG: DEBUG
run: |
poetry run pytest tests/proxy_e2e_azure_batches_tests/test_proxy_e2e_azure_batches.py \
-vv -s -k "test_e2e_managed_batch" \
--tb=short \
--maxfail=3 \
--durations=10

3
AGENTS.md
View file

@ -109,6 +109,8 @@ Key files:
- `litellm/proxy/auth/` - Authentication logic
- `litellm/proxy/management_endpoints/` - Admin API endpoints
**Database (proxy)**: Use Prisma model methods (`prisma_client.db.<model>.upsert`, `.find_many`, `.find_unique`, etc.), not raw SQL (`execute_raw`/`query_raw`). See COMMON PITFALLS for details.
## MCP (MODEL CONTEXT PROTOCOL) SUPPORT
LiteLLM supports MCP for agent workflows:
@ -176,6 +178,7 @@ When opening issues or pull requests, follow these templates:
5. **Dependencies**: Keep dependencies minimal and well-justified
6. **UI/Backend Contract Mismatch**: When adding a new entity type to the UI, always check whether the backend endpoint accepts a single value or an array. Match the UI control accordingly (single-select vs. multi-select) to avoid silently dropping user selections
7. **Missing Tests for New Entity Types**: When adding a new entity type (e.g., in `EntityUsage`, `UsageViewSelect`), always add corresponding tests in the existing test files and update any icon/component mocks
8. **Raw SQL in proxy DB code**: Do not use `execute_raw` or `query_raw` for proxy database access. Use Prisma model methods (e.g. `prisma_client.db.litellm_tooltable.upsert()`, `.find_many()`, `.find_unique()`) so behavior stays consistent with the schema, the client stays mockable in tests, and you avoid the pitfalls of hand-written SQL (parameter ordering, type casting, schema drift)
8. **Do not hardcode model-specific flags**: Put model-specific capability flags in `model_prices_and_context_window.json` and read them via `get_model_info` (or existing helpers like `supports_reasoning`). This prevents users from needing to upgrade LiteLLM each time a new model supports a feature.

4
CLAUDE.md
View file

@ -107,6 +107,10 @@ LiteLLM is a unified interface for 100+ LLM providers with two main components:
- Migration files auto-generated with `prisma migrate dev`
- Always test migrations against both PostgreSQL and SQLite
### Proxy database access
- **Do not write raw SQL** for proxy DB operations. Use Prisma model methods instead of `execute_raw` / `query_raw`.
- Use the generated client: `prisma_client.db.<model>` (e.g. `litellm_tooltable`, `litellm_usertable`) with `.upsert()`, `.find_many()`, `.find_unique()`, `.update()`, `.update_many()` as appropriate. This avoids schema/client drift, keeps code testable with simple mocks, and matches patterns used in spend logs and other proxy code.
### Enterprise Features
- Enterprise-specific code in `enterprise/` directory
- Optional features enabled via environment variables

13
dev_config.yaml Normal file
View file

@ -0,0 +1,13 @@
model_list:
- model_name: fake-openai-endpoint
litellm_params:
model: openai/fake-model
api_key: fake-key
api_base: https://exampleopenaiendpoint-production.up.railway.app/
general_settings:
master_key: sk-1234
litellm_settings:
drop_params: True
telemetry: False

175
docs/my-website/blog/gemini_3_1_flash_lite/index.md Normal file
View file

@ -0,0 +1,175 @@
---
slug: gemini_3_1_flash_lite_preview
title: "DAY 0 Support: Gemini 3.1 Flash Lite Preview on LiteLLM"
date: 2026-03-03T08:00:00
authors:
- name: Sameer Kankute
title: SWE @ LiteLLM (LLM Translation)
url: https://www.linkedin.com/in/sameer-kankute/
image_url: https://pbs.twimg.com/profile_images/2001352686994907136/ONgNuSk5_400x400.jpg
- name: Krrish Dholakia
title: "CEO, LiteLLM"
url: https://www.linkedin.com/in/krish-d/
image_url: https://pbs.twimg.com/profile_images/1298587542745358340/DZv3Oj-h_400x400.jpg
- name: Ishaan Jaff
title: "CTO, LiteLLM"
url: https://www.linkedin.com/in/reffajnaahsi/
image_url: https://pbs.twimg.com/profile_images/1613813310264340481/lz54oEiB_400x400.jpg
description: "Guide to using Gemini 3.1 Flash Lite Preview on LiteLLM Proxy and SDK with day 0 support."
tags: [gemini, day 0 support, llms, supernova]
hide_table_of_contents: false
---
import Tabs from '@theme/Tabs';
import TabItem from '@theme/TabItem';
# Gemini 3.1 Flash Lite Preview Day 0 Support
LiteLLM now supports `gemini-3.1-flash-lite-preview` with full day 0 support!
:::note
If you only want cost tracking, you need no change in your current Litellm version. But if you want the support for new features introduced along with it like thinking levels, you will need to use v1.80.8-stable.1 or above.
:::
## Deploy this version
<Tabs>
<TabItem value="docker" label="Docker">
``` showLineNumbers title="docker run litellm"
docker run \
-e STORE_MODEL_IN_DB=True \
-p 4000:4000 \
ghcr.io/berriai/litellm:main-v1.80.8-stable.1
```
</TabItem>
<TabItem value="pip" label="Pip">
``` showLineNumbers title="pip install litellm"
pip install litellm==v1.80.8-stable.1
```
</TabItem>
</Tabs>
## What's New
Supports all four thinking levels:
- **MINIMAL**: Ultra-fast responses with minimal reasoning
- **LOW**: Simple instruction following
- **MEDIUM**: Balanced reasoning for complex tasks
- **HIGH**: Maximum reasoning depth (dynamic)
---
## Quick Start
<Tabs>
<TabItem value="sdk" label="SDK">
**Basic Usage**
```python
from litellm import completion
response = completion(
model="gemini/gemini-3.1-flash-lite-preview",
messages=[{"role": "user", "content": "Extract key entities from this text: ..."}],
)
print(response.choices[0].message.content)
```
**With Thinking Levels**
```python
from litellm import completion
# Use MEDIUM thinking for complex reasoning tasks
response = completion(
model="gemini/gemini-3.1-flash-lite-preview",
messages=[{"role": "user", "content": "Analyze this dataset and identify patterns"}],
reasoning_effort="medium", # low, medium , high
)
print(response.choices[0].message.content)
```
</TabItem>
<TabItem value="proxy" label="PROXY">
**1. Setup config.yaml**
```yaml
model_list:
- model_name: gemini-3.1-flash-lite
litellm_params:
model: gemini/gemini-3.1-flash-lite-preview
api_key: os.environ/GEMINI_API_KEY
# Or use Vertex AI
- model_name: vertex-gemini-3.1-flash-lite
litellm_params:
model: vertex_ai/gemini-3.1-flash-lite-preview
vertex_project: your-project-id
vertex_location: us-central1
```
**2. Start proxy**
```bash
litellm --config /path/to/config.yaml
```
**3. Make requests**
```bash
curl -X POST http://localhost:4000/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer <YOUR-LITELLM-KEY>" \
-d '{
"model": "gemini-3.1-flash-lite",
"messages": [{"role": "user", "content": "Extract structured data from this text"}],
"reasoning_effort": "low"
}'
```
</TabItem>
</Tabs>
---
## Supported Endpoints
LiteLLM provides **full end-to-end support** for Gemini 3.1 Flash Lite Preview on:
- ✅ `/v1/chat/completions` - OpenAI-compatible chat completions endpoint
- ✅ `/v1/responses` - OpenAI Responses API endpoint (streaming and non-streaming)
- ✅ [`/v1/messages`](../../docs/anthropic_unified) - Anthropic-compatible messages endpoint
- ✅ `/v1/generateContent` – [Google Gemini API](../../docs/generateContent.md) compatible endpoint
All endpoints support:
- Streaming and non-streaming responses
- Function calling with thought signatures
- Multi-turn conversations
- All Gemini 3-specific features (thinking levels, thought signatures)
- Full multimodal support (text, image, audio, video)
---
## `reasoning_effort` Mapping for Gemini 3.1
LiteLLM automatically maps OpenAI's `reasoning_effort` parameter to Gemini's `thinkingLevel`:
| reasoning_effort | thinking_level | Use Case |
|------------------|----------------|----------|
| `minimal` | `minimal` | Ultra-fast responses, simple queries |
| `low` | `low` | Basic instruction following |
| `medium` | `medium` | Balanced reasoning for moderate complexity |
| `high` | `high` | Maximum reasoning depth, complex problems |
| `disable` | `minimal` | Disable extended reasoning |
| `none` | `minimal` | No extended reasoning |

321
docs/my-website/blog/responses_api_encrypted_content_incident/index.md Normal file
View file

@ -0,0 +1,321 @@
---
slug: responses-api-encrypted-content-incident
title: "Incident Report: Encrypted Content Failures in Multi-Region Responses API Load Balancing"
date: 2026-02-24T10:00:00
authors:
- name: Sameer Kankute
title: SWE @ LiteLLM (LLM Translation)
url: https://www.linkedin.com/in/sameer-kankute/
image_url: https://pbs.twimg.com/profile_images/2001352686994907136/ONgNuSk5_400x400.jpg
- name: Krrish Dholakia
title: "CEO, LiteLLM"
url: https://www.linkedin.com/in/krish-d/
image_url: https://pbs.twimg.com/profile_images/1298587542745358340/DZv3Oj-h_400x400.jpg
- name: Ishaan Jaff
title: "CTO, LiteLLM"
url: https://www.linkedin.com/in/reffajnaahsi/
image_url: https://pbs.twimg.com/profile_images/1613813310264340481/lz54oEiB_400x400.jpg
tags: [incident-report, proxy, responses-api, load-balancing]
hide_table_of_contents: false
---
**Date:** Feb 24, 2026
**Duration:** Ongoing (until fix deployed)
**Severity:** High (for users load balancing Responses API across different API keys)
**Status:** Resolved
## Summary
When load balancing OpenAI's Responses API across deployments with **different API keys** (e.g., different Azure regions or OpenAI organizations), follow-up requests containing encrypted content items (like `rs_...` reasoning items) would fail with:
```json
{
"error": {
"message": "The encrypted content for item rs_0d09d6e56879e76500699d6feee41c8197bd268aae76141f87 could not be verified. Reason: Encrypted content organization_id did not match the target organization.",
"type": "invalid_request_error",
"code": "invalid_encrypted_content"
}
}
```
Encrypted content items are cryptographically tied to the API key's organization that created them. When the router load balanced a follow-up request to a deployment with a different API key, decryption failed.
- **Responses API calls with encrypted content:** Complete failure when routed to wrong deployment
- **Initial requests:** Unaffected — only follow-up requests containing encrypted items failed
- **Other API endpoints:** No impact — chat completions, embeddings, etc. functioned normally
{/* truncate */}
---
## Background
OpenAI's Responses API can return encrypted "reasoning items" (with IDs like `rs_...`) that contain intermediate reasoning steps. These items are encrypted with the organization's key and can only be decrypted by the same organization's API key.
When load balancing across deployments with different API keys, the existing affinity mechanisms were insufficient:
- **`responses_api_deployment_check`**: Requires `previous_response_id` which some clients (like Codex) don't provide
- **`deployment_affinity`**: Too broad — pins *all* requests from a user to one deployment, reducing effective quota by the number of users
- **`session_affinity`**: Requires explicit session IDs and still reduces quota
```mermaid
flowchart TD
A["1. Initial request to Responses API
router.aresponses()"] --> B["2. Router load balances to Deployment A
(API Key 1, Azure East US)"]
B --> C["3. Response contains encrypted item
rs_abc123 (encrypted with Org 1 key)"]
C --> D["4. Follow-up request includes rs_abc123 in input"]
D --> E["5. Router load balances to Deployment B
(API Key 2, Azure West Europe)"]
E -->|"Different API key"| F["6. ❌ Deployment B cannot decrypt rs_abc123
Error: invalid_encrypted_content"]
D -.->|"With encrypted_content_affinity"| G["5b. Router detects rs_abc123 was created by Deployment A"]
G --> H["6b. ✅ Routes to Deployment A (bypasses rate limits)
Request succeeds"]
style F fill:#f8d7da,stroke:#dc3545
style H fill:#d4edda,stroke:#28a745
style E fill:#fff3cd,stroke:#ffc107
style G fill:#d4edda,stroke:#28a745
```
---
## Root Cause
LiteLLM's router had no mechanism to track which deployment created specific encrypted content items and route follow-up requests accordingly. The router treated all deployments as interchangeable, leading to decryption failures when encrypted content crossed organizational boundaries.
**The Problem Flow:**
1. User calls `router.aresponses()` with model `gpt-5.1-codex`
2. Router load balances to Deployment A (Azure East US, API Key 1)
3. Response contains encrypted reasoning item `rs_abc123` (encrypted with Org 1's key)
4. User makes follow-up request with `rs_abc123` in the input
5. Router load balances to Deployment B (Azure West Europe, API Key 2)
6. Deployment B tries to decrypt `rs_abc123` with Org 2's key → **fails**
**Why Existing Solutions Didn't Work:**
- **`previous_response_id`**: Not provided by all clients (e.g., Codex)
- **`deployment_affinity`**: Pins *all* user requests to one deployment → reduces quota to 1/N where N = number of deployments
- **`session_affinity`**: Requires explicit session management and still reduces quota
**Timeline:**
1. Users configured multi-region Responses API load balancing with different API keys
2. Initial requests succeeded, but follow-up requests with encrypted content failed intermittently
3. Error rate correlated with number of deployments (more deployments = higher chance of routing to wrong one)
4. Investigation revealed encrypted content was organization-bound
5. Existing affinity mechanisms deemed unsuitable (quota reduction, missing `previous_response_id`)
6. New solution designed and implemented: `encrypted_content_affinity`
---
## The Fix
Implemented a new `encrypted_content_affinity` pre-call check that intelligently tracks encrypted content and routes follow-up requests **only when necessary**.
### Implementation
**1. Encoding `model_id` into output items** ([`responses/utils.py`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm/responses/utils.py))
The same approach used for `previous_response_id` affinity — no cache needed. When a response contains output items with `encrypted_content`, LiteLLM encodes the originating deployment's `model_id` in **two places** for redundancy:
1. **Into the item ID** (if present): `rs_abc123` → `encitem_{base64("litellm:model_id:{model_id};item_id:rs_abc123")}`
2. **Into the encrypted_content itself**: Wraps the content with `litellm_enc:{base64("model_id:{model_id}")};{original_encrypted_content}`
```python
# Encoding item IDs (when present)
def _build_encrypted_item_id(model_id: str, item_id: str) -> str:
assembled = f"litellm:model_id:{model_id};item_id:{item_id}"
encoded = base64.b64encode(assembled.encode("utf-8")).decode("utf-8")
return f"encitem_{encoded}"
# Wrapping encrypted_content (always, for redundancy)
def _wrap_encrypted_content_with_model_id(encrypted_content: str, model_id: str) -> str:
metadata = f"model_id:{model_id}"
encoded_metadata = base64.b64encode(metadata.encode("utf-8")).decode("utf-8")
return f"litellm_enc:{encoded_metadata};{encrypted_content}"
```
**Why wrap encrypted_content directly?** Some clients (like Codex) don't consistently send item IDs in follow-up requests, but they always send the `encrypted_content` itself. By embedding `model_id` into the content, affinity works even when IDs are missing.
**Streaming responses:** The wrapping logic is applied to both:
- Final response objects (non-streaming)
- Individual streaming events (`response.output_item.added`, `response.output_item.done`)
This ensures clients receiving streaming responses get wrapped content they can send back.
Before forwarding to the upstream provider, LiteLLM restores the original item IDs and unwraps encrypted_content so the provider never sees the encoded form:
```python
# In responses/main.py — before calling the handler
input = ResponsesAPIRequestUtils._restore_encrypted_content_item_ids_in_input(input)
```
**2. `EncryptedContentAffinityCheck` — routing only** ([`encrypted_content_affinity_check.py`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm/router_utils/pre_call_checks/encrypted_content_affinity_check.py))
No `async_log_success_event` or cache lookups — the `model_id` is decoded directly from the item ID or encrypted_content:
```python
class EncryptedContentAffinityCheck(CustomLogger):
async def async_filter_deployments(self, model, healthy_deployments, ...):
"""Extract model_id from input items (ID or encrypted_content) and pin to that deployment."""
for item in request_kwargs.get("input", []):
# Try to extract model_id from two sources:
model_id = self._extract_model_id_from_input(item)
if model_id:
deployment = self._find_deployment_by_model_id(
healthy_deployments, model_id
)
if deployment:
request_kwargs["_encrypted_content_affinity_pinned"] = True
return [deployment]
return healthy_deployments
def _extract_model_id_from_input(self, item: dict) -> Optional[str]:
"""Extract model_id from either encoded ID or wrapped encrypted_content."""
# 1. Try decoding from item ID (if present)
item_id = item.get("id", "")
if item_id:
decoded = ResponsesAPIRequestUtils._decode_encrypted_item_id(item_id)
if decoded:
return decoded["model_id"]
# 2. Try unwrapping from encrypted_content (fallback for clients that omit IDs)
encrypted_content = item.get("encrypted_content", "")
if encrypted_content and encrypted_content.startswith("litellm_enc:"):
model_id, _ = ResponsesAPIRequestUtils._unwrap_encrypted_content_with_model_id(
encrypted_content
)
return model_id
return None
```
**3. Rate Limit Bypass** ([`router.py`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm/router.py))
When encrypted content requires a specific deployment, RPM/TPM limits are bypassed (the request would fail on any other deployment anyway):
```python
# In async_get_available_deployment, after filtering healthy deployments:
if (
request_kwargs.get("_encrypted_content_affinity_pinned")
and len(healthy_deployments) == 1
):
return healthy_deployments[0] # Bypass routing strategy (RPM/TPM checks)
```
**3. Configuration**
```yaml
router_settings:
routing_strategy: usage-based-routing-v2
enable_pre_call_checks: true
optional_pre_call_checks:
- encrypted_content_affinity
deployment_affinity_ttl_seconds: 86400 # 24 hours
```
### Key Benefits
✅ **No quota reduction**: Only pins requests containing encrypted items
✅ **Bypasses rate limits**: When encrypted content requires a specific deployment, RPM/TPM limits don't block it
✅ **No `previous_response_id` required**: Works by encoding `model_id` directly into the item ID
✅ **No cache required**: `model_id` is decoded on-the-fly from the item ID — no Redis, no TTL
✅ **Globally safe**: Can be enabled for all models; non-Responses-API calls are unaffected
✅ **Surgical precision**: Normal requests continue to load balance freely
---
## Remediation
| # | Action | Status | Code |
|---|---|---|---|
| 1 | Encode `model_id` into encrypted-content item IDs on response | ✅ Done | [`responses/utils.py`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm/responses/utils.py) |
| 2 | Restore original item IDs before forwarding to upstream provider | ✅ Done | [`responses/main.py`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm/responses/main.py) |
| 3 | `EncryptedContentAffinityCheck`: decode item IDs to route (no cache) | ✅ Done | [`encrypted_content_affinity_check.py`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm/router_utils/pre_call_checks/encrypted_content_affinity_check.py) |
| 4 | Add `encrypted_content_affinity` to `OptionalPreCallChecks` type | ✅ Done | [`types/router.py`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm/types/router.py) |
| 5 | Implement rate limit bypass for affinity-pinned requests | ✅ Done | [`router.py`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm/router.py) |
| 6 | Unit tests: encoding/decoding utilities, routing, RPM bypass | ✅ Done | [`test_encrypted_content_affinity_check.py`](https://github.com/BerriAI/litellm/blob/main/litellm/tests/test_litellm/router_utils/pre_call_checks/test_encrypted_content_affinity_check.py) |
| 7 | Documentation: Responses API guide, load balancing guide, config reference | ✅ Done | [Docs](https://docs.litellm.ai/docs/response_api#encrypted-content-affinity-multi-region-load-balancing) |
| 8 | **[Mar 3]** Fix streaming events to wrap encrypted_content | ✅ Done | [`responses/streaming_iterator.py`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm/responses/streaming_iterator.py) |
---
## Follow-up Fix: Streaming Responses (Mar 3, 2026)
### The Issue
After the initial fix was deployed, users reported that the `invalid_encrypted_content` error **still occurred** when using streaming responses with clients like Codex. Investigation revealed:
- ✅ Non-streaming responses: `encrypted_content` was correctly wrapped with `litellm_enc:` prefix
- ❌ Streaming responses: Individual `response.output_item.added` and `response.output_item.done` events contained **raw, unwrapped** `encrypted_content`
Since Codex and other clients consume responses as streams, they received unwrapped content in these events and sent it back in follow-up requests, causing the affinity check to fail.
### The Root Cause
The `_update_encrypted_content_item_ids_in_response` function only modified the **final** response object, which is used for non-streaming responses. For streaming responses, individual chunks are processed by `ResponsesAPIStreamingIterator._process_chunk`, which was **not** applying the wrapping logic to streaming events.
### The Fix
Modified `litellm/litellm/responses/streaming_iterator.py` to wrap `encrypted_content` in streaming events:
```python
# In ResponsesAPIStreamingIterator._process_chunk
if (
self.litellm_metadata
and self.litellm_metadata.get("encrypted_content_affinity_enabled")
):
event_type = getattr(openai_responses_api_chunk, "type", None)
if event_type in (
ResponsesAPIStreamEvents.OUTPUT_ITEM_ADDED,
ResponsesAPIStreamEvents.OUTPUT_ITEM_DONE,
):
item = getattr(openai_responses_api_chunk, "item", None)
if item:
encrypted_content = getattr(item, "encrypted_content", None)
if encrypted_content and isinstance(encrypted_content, str):
model_id = (
self.litellm_metadata.get("model_info", {}).get("id")
if self.litellm_metadata
else None
)
if model_id:
wrapped_content = ResponsesAPIRequestUtils._wrap_encrypted_content_with_model_id(
encrypted_content, model_id
)
setattr(item, "encrypted_content", wrapped_content)
```
This ensures that **all** `encrypted_content` sent to clients (streaming or non-streaming) is wrapped with `model_id` metadata, enabling consistent affinity routing.
---
## Migration Guide
### Before (Using `deployment_affinity`)
```yaml
router_settings:
optional_pre_call_checks:
- deployment_affinity # ❌ Reduces quota by number of users
```
**Problem:** All requests from a user pin to one deployment, reducing effective quota to 1/N.
### After (Using `encrypted_content_affinity`)
```yaml
router_settings:
optional_pre_call_checks:
- encrypted_content_affinity # ✅ Only pins requests with encrypted content
```
**Benefit:** Normal requests load balance freely, only encrypted content requests pin when necessary.
---

71
docs/my-website/docs/image_edits.md
View file

@ -16,7 +16,7 @@ LiteLLM provides image editing functionality that maps to OpenAI's `/images/edit
| Supported operations | Create image edits | Single and multiple images supported |
| Supported LiteLLM SDK Versions | 1.63.8+ | Gemini support requires 1.79.3+ |
| Supported LiteLLM Proxy Versions | 1.71.1+ | Gemini support requires 1.79.3+ |
| Supported LLM providers | **OpenAI**, **Gemini (Google AI Studio)**, **Vertex AI**, **Stability AI**, **AWS Bedrock (Stability)** | Gemini supports the new `gemini-2.5-flash-image` family. Vertex AI supports both Gemini and Imagen models. Stability AI and Bedrock Stability support various image editing operations. |
| Supported LLM providers | **OpenAI**, **Gemini (Google AI Studio)**, **Vertex AI**, **OpenRouter**, **Stability AI**, **AWS Bedrock (Stability)** | Gemini supports the new `gemini-2.5-flash-image` family. Vertex AI supports both Gemini and Imagen models. OpenRouter routes image edits through chat completions. Stability AI and Bedrock Stability support various image editing operations. |
#### ⚡️See all supported models and providers at [models.litellm.ai](https://models.litellm.ai/)
@ -244,6 +244,47 @@ response = litellm.image_edit(
print(response)
```
</TabItem>
<TabItem value="openrouter" label="OpenRouter">
#### Basic Image Edit
```python showLineNumbers title="OpenRouter Image Edit"
import os
from litellm import image_edit
os.environ["OPENROUTER_API_KEY"] = "your-api-key"
response = image_edit(
model="openrouter/google/gemini-2.5-flash-image",
image=open("original_image.png", "rb"),
prompt="Add aurora borealis to the night sky",
)
print(response)
```
#### Multiple Images Edit
```python showLineNumbers title="OpenRouter Multiple Images Edit"
import os
from litellm import image_edit
os.environ["OPENROUTER_API_KEY"] = "your-api-key"
response = image_edit(
model="openrouter/google/gemini-2.5-flash-image",
image=[
open("scene.png", "rb"),
open("style_reference.png", "rb"),
],
prompt="Blend the reference style into the scene",
size="1536x1024", # mapped to aspect_ratio 3:2
quality="high", # mapped to image_size 4K
)
print(response)
```
</TabItem>
</Tabs>
@ -398,6 +439,34 @@ curl -X POST "http://0.0.0.0:4000/v1/images/edits" \
-F "size=1024x1024"
```
</TabItem>
<TabItem value="openrouter" label="OpenRouter">
1. Add the OpenRouter image edit model to your `config.yaml`:
```yaml showLineNumbers title="OpenRouter Proxy Configuration"
model_list:
- model_name: openrouter-image-edit
litellm_params:
model: openrouter/google/gemini-2.5-flash-image
api_key: os.environ/OPENROUTER_API_KEY
```
2. Start the LiteLLM proxy server:
```bash showLineNumbers title="Start LiteLLM Proxy Server"
litellm --config /path/to/config.yaml
```
3. Make an image edit request:
```bash showLineNumbers title="OpenRouter Proxy Image Edit"
curl -X POST "http://0.0.0.0:4000/v1/images/edits" \
-H "Authorization: Bearer <YOUR-LITELLM-KEY>" \
-F "model=openrouter-image-edit" \
-F "image=@original_image.png" \
-F "prompt=Make the sky a vibrant purple sunset" \
-F "size=1024x1024"
```
</TabItem>
</Tabs>

1
docs/my-website/docs/providers/gemini.md
View file

@ -2041,6 +2041,7 @@ response = litellm.completion(
| gemini-2.0-flash-lite-preview-02-05 | `completion(model='gemini/gemini-2.0-flash-lite-preview-02-05', messages)` | `os.environ['GEMINI_API_KEY']` |
| gemini-2.5-flash-preview-09-2025 | `completion(model='gemini/gemini-2.5-flash-preview-09-2025', messages)` | `os.environ['GEMINI_API_KEY']` |
| gemini-2.5-flash-lite-preview-09-2025 | `completion(model='gemini/gemini-2.5-flash-lite-preview-09-2025', messages)` | `os.environ['GEMINI_API_KEY']` |
| gemini-3.1-flash-lite-preview | `completion(model='gemini/gemini-3.1-flash-lite-preview', messages)` | `os.environ['GEMINI_API_KEY']` |
| gemini-flash-latest | `completion(model='gemini/gemini-flash-latest', messages)` | `os.environ['GEMINI_API_KEY']` |
| gemini-flash-lite-latest | `completion(model='gemini/gemini-flash-lite-latest', messages)` | `os.environ['GEMINI_API_KEY']` |

1
docs/my-website/docs/providers/openai.md
View file

@ -191,6 +191,7 @@ os.environ["OPENAI_BASE_URL"] = "https://your_host/v1" # OPTIONAL
| gpt-5.2 | `response = completion(model="gpt-5.2", messages=messages)` |
| gpt-5.2-2025-12-11 | `response = completion(model="gpt-5.2-2025-12-11", messages=messages)` |
| gpt-5.2-chat-latest | `response = completion(model="gpt-5.2-chat-latest", messages=messages)` |
| gpt-5.3-chat-latest | `response = completion(model="gpt-5.3-chat-latest", messages=messages)` |
| gpt-5.2-pro | `response = completion(model="gpt-5.2-pro", messages=messages)` |
| gpt-5.2-pro-2025-12-11 | `response = completion(model="gpt-5.2-pro-2025-12-11", messages=messages)` |
| gpt-5.1 | `response = completion(model="gpt-5.1", messages=messages)` |

87
docs/my-website/docs/providers/openrouter.md
View file

@ -210,3 +210,90 @@ response = image_generation(
# Cost is available in the response metadata
print(f"Request cost: ${response._hidden_params['additional_headers']['llm_provider-x-litellm-response-cost']}")
```
## Image Edit
OpenRouter supports image editing through select models like Google Gemini image models. LiteLLM routes image edit requests to OpenRouter's chat completions endpoint with the source image sent as a base64 data URL and `modalities: ["image", "text"]`.
### Supported Models
| Model | Description |
|-------|-------------|
| `openrouter/google/gemini-2.5-flash-image` | Gemini 2.5 Flash with image editing |
See all available image models on [OpenRouter's model list](https://openrouter.ai/models?modality=image).
### Supported Parameters
| Parameter | OpenRouter Mapping | Notes |
|-----------|--------------------|-------|
| `size` | `image_config.aspect_ratio` | `1024x1024` → `1:1`, `1536x1024` → `3:2`, `1024x1536` → `2:3`, `1792x1024` → `16:9`, `1024x1792` → `9:16` |
| `quality` | `image_config.image_size` | `low`/`standard` → `1K`, `medium` → `2K`, `high`/`hd` → `4K` |
| `n` | `n` | Number of images |
:::note
`quality=high` (4K) is only supported by `google/gemini-3-pro-image-preview` and `google/gemini-3.1-flash-image-preview`. The `google/gemini-2.5-flash-image` model supports up to `medium` (2K).
:::
### Usage
```python
from litellm import image_edit
import os
os.environ["OPENROUTER_API_KEY"] = "your-api-key"
# Basic image edit
response = image_edit(
model="openrouter/google/gemini-2.5-flash-image",
image=open("original_image.png", "rb"),
prompt="Make the sky a vibrant purple sunset",
)
print(response)
```
### Advanced Usage with Parameters
```python
from litellm import image_edit
import os
os.environ["OPENROUTER_API_KEY"] = "your-api-key"
# Edit with size and quality parameters
response = image_edit(
model="openrouter/google/gemini-2.5-flash-image",
image=open("photo.png", "rb"),
prompt="Add northern lights to the sky",
size="1536x1024", # Maps to aspect_ratio 3:2
quality="high", # Maps to image_size 4K
)
# Access the edited image
image_data = response.data[0]
if image_data.b64_json:
import base64
with open("edited.png", "wb") as f:
f.write(base64.b64decode(image_data.b64_json))
```
### Multiple Images Edit
```python
from litellm import image_edit
import os
os.environ["OPENROUTER_API_KEY"] = "your-api-key"
response = image_edit(
model="openrouter/google/gemini-2.5-flash-image",
image=[
open("scene.png", "rb"),
open("style_reference.png", "rb"),
],
prompt="Blend the reference style into the scene",
)
print(response)
```

1
docs/my-website/docs/providers/vertex.md
View file

@ -1685,6 +1685,7 @@ litellm.vertex_location = "us-central1 # Your Location
| gemini-2.5-pro | `completion('gemini-2.5-pro', messages)`, `completion('vertex_ai/gemini-2.5-pro', messages)` |
| gemini-2.5-flash-preview-09-2025 | `completion('gemini-2.5-flash-preview-09-2025', messages)`, `completion('vertex_ai/gemini-2.5-flash-preview-09-2025', messages)` |
| gemini-2.5-flash-lite-preview-09-2025 | `completion('gemini-2.5-flash-lite-preview-09-2025', messages)`, `completion('vertex_ai/gemini-2.5-flash-lite-preview-09-2025', messages)` |
| gemini-3.1-flash-lite-preview | `completion('gemini-3.1-flash-lite-preview', messages)`, `completion('vertex_ai/gemini-3.1-flash-lite-preview', messages)` |
## Private Service Connect (PSC) Endpoints

2
docs/my-website/docs/proxy/config_settings.md
View file

@ -360,7 +360,7 @@ router_settings:
| redis_url | str | URL for Redis server. **Known performance issue with Redis URL.** |
| cache_responses | boolean | Flag to enable caching LLM Responses, if cache set under `router_settings`. If true, caches responses. Defaults to False. |
| router_general_settings | RouterGeneralSettings | [SDK-Only] Router general settings - contains optimizations like 'async_only_mode'. [Docs](../routing.md#router-general-settings) |
| optional_pre_call_checks | List[str] | List of pre-call checks to add to the router. Supported: `router_budget_limiting`, `prompt_caching`, `responses_api_deployment_check`, `deployment_affinity`, `forward_client_headers_by_model_group` |
| optional_pre_call_checks | List[str] | List of pre-call checks to add to the router. Supported: `router_budget_limiting`, `prompt_caching`, `responses_api_deployment_check`, `encrypted_content_affinity`, `deployment_affinity`, `session_affinity`, `forward_client_headers_by_model_group` |
| deployment_affinity_ttl_seconds | int | TTL (seconds) for user-key → deployment affinity mapping when `deployment_affinity` is enabled (configured at Router init / proxy startup). Defaults to `3600` (1 hour). |
| ignore_invalid_deployments | boolean | If true, ignores invalid deployments. Default for proxy is True - to prevent invalid models from blocking other models from being loaded. |
| search_tools | List[SearchToolTypedDict] | List of search tool configurations for Search API integration. Each tool specifies a search_tool_name and litellm_params with search_provider, api_key, api_base, etc. [Further Docs](../search.md) |

50
docs/my-website/docs/proxy/guardrails/quick_start.md
View file

@ -358,13 +358,13 @@ response = client.chat.completions.create(
}
],
extra_body={
"guardrails": [
"guardrails": {
"aporia-pre-guard": {
"extra_body": {
"success_threshold": 0.9
}
}
]
}
}
)
@ -387,13 +387,13 @@ curl --location 'http://0.0.0.0:4000/chat/completions' \
"content": "what llm are you"
}
],
"guardrails": [
"guardrails": {
"aporia-pre-guard": {
"extra_body": {
"success_threshold": 0.9
}
}
]
}
}'
```
</TabItem>
@ -451,7 +451,6 @@ curl -X POST 'http://0.0.0.0:4000/key/generate' \
-H 'Content-Type: application/json' \
-d '{
"guardrails": ["aporia-pre-guard", "aporia-post-guard"]
}
}'
```
@ -465,7 +464,6 @@ curl --location 'http://0.0.0.0:4000/key/update' \
--data '{
"key": "sk-jNm1Zar7XfNdZXp49Z1kSQ",
"guardrails": ["aporia-pre-guard", "aporia-post-guard"]
}
}'
```
@ -499,6 +497,11 @@ curl --location 'http://0.0.0.0:4000/chat/completions' \
Run guardrails based on the user-agent header. This is useful for running pre-call checks on OpenWebUI but only masking in logs for Claude CLI.
`default` can be a single mode string or a list of modes.
<Tabs>
<TabItem value="single" label="Single Default Mode">
```yaml
model_list:
- model_name: gpt-3.5-turbo
@ -519,6 +522,32 @@ guardrails:
default_on: true # run on every request
```
</TabItem>
<TabItem value="multi" label="Multiple Default Modes">
```yaml
model_list:
- model_name: gpt-3.5-turbo
litellm_params:
model: gpt-3.5-turbo
api_key: os.environ/OPENAI_API_KEY
guardrails:
- guardrail_name: "guardrails_ai-guard"
litellm_params:
guardrail: guardrails_ai
guard_name: "pii_detect"
mode:
tags:
"User-Agent: claude-cli": "logging_only"
default: ["pre_call", "post_call"] # Run on both pre and post call when no tags match
api_base: os.environ/GUARDRAILS_AI_API_BASE
default_on: true
```
</TabItem>
</Tabs>
### ✨ Model-level Guardrails
@ -640,13 +669,22 @@ guardrails:
Mode Specification
`default` accepts either a single string or a list of strings.
```python
from litellm.types.guardrails import Mode
# Single default mode
mode = Mode(
tags={"User-Agent: claude-cli": "logging_only"},
default="logging_only"
)
# Multiple default modes
mode = Mode(
tags={"User-Agent: claude-cli": "logging_only"},
default=["pre_call", "post_call"]
)
```
### `guardrails` Request Parameter

33
docs/my-website/docs/proxy/load_balancing.md
View file

@ -347,3 +347,36 @@ If `order=1` deployment is unavailable (e.g., rate-limited), the router falls ba
- **Higher throughput**: More requests handled simultaneously across deployments
- **Improved reliability**: If one deployment fails, traffic automatically routes to healthy ones
- **Better resource utilization**: Load spread evenly across all available deployments
## Special Considerations for Responses API
When load balancing OpenAI's Responses API across deployments with **different API keys** (e.g., different Azure regions or organizations), encrypted content items (like `rs_...` reasoning items) can only be decrypted by the originating API key.
**Solution:** Use the `encrypted_content_affinity` pre-call check to automatically route follow-up requests containing encrypted items to the correct deployment:
```yaml
model_list:
- model_name: gpt-5.1-codex
litellm_params:
model: azure/gpt-5.1-codex
api_base: https://eastus.openai.azure.com/
api_key: os.environ/AZURE_API_KEY_EASTUS
model_info:
id: "deployment-eastus"
- model_name: gpt-5.1-codex
litellm_params:
model: azure/gpt-5.1-codex
api_base: https://westeurope.openai.azure.com/
api_key: os.environ/AZURE_API_KEY_WESTEUROPE
model_info:
id: "deployment-westeurope"
router_settings:
optional_pre_call_checks:
- encrypted_content_affinity # 👈 Prevents invalid_encrypted_content errors
```
This ensures requests containing encrypted content are routed to the deployment that created them, while other requests continue to load balance normally.
**[Learn more about Encrypted Content Affinity →](../response_api.md#encrypted-content-affinity-multi-region-load-balancing)**

143
docs/my-website/docs/response_api.md
View file

@ -920,12 +920,17 @@ follow_up = await router.aresponses(
To enable session continuity for Responses API in your LiteLLM proxy, set `optional_pre_call_checks` in your proxy config.yaml.
- `responses_api_deployment_check`: high priority routing when `previous_response_id` is provided
- `encrypted_content_affinity`: **[Recommended]** content-aware routing for encrypted items (e.g., `rs_...` reasoning items)
- `session_affinity`: sticky sessions based on session id (takes priority over `deployment_affinity`)
- `deployment_affinity`: sticky sessions based on user key (applies even without `previous_response_id`)
:::tip Recommended: Use `encrypted_content_affinity`
For Responses API with load balancing across deployments with **different API keys**, use `encrypted_content_affinity` instead of `deployment_affinity`. It only pins requests that contain encrypted content, avoiding quota reduction while preventing `invalid_encrypted_content` errors.
:::
Notes:
- User-key affinity is keyed on `metadata.user_api_key_hash` (the API key hash). The OpenAI `user` request parameter is an end-user identifier and is intentionally not used for deployment affinity.
- Session-ID affinity is keyed on `metadata.session_id`. For proxy requests, this can be passed via the `x-litellm-session-id` HTTP header. For Python SDK requests, you can pass it via `litellm_metadata={"session_id": "value"}` in request args.
- Session-ID affinity is keyed on `metadata.session_id`. For proxy requests, this can be passed via the `x-litellm-session-id` or `x-litellm-trace-id` HTTP header (they are interchangeable for call chaining). For Python SDK requests, you can pass it via `litellm_metadata={"session_id": "value"}` in request args.
- `user_api_key_hash` is already SHA-256, and is used as-is (no double hashing).
- Affinity is scoped by a stable model identifier (the model-map key, e.g. `model_map_information.model_map_key`) so model aliases map to the same stickiness bucket.
- The mapping TTL is controlled by `deployment_affinity_ttl_seconds` (configured on Router init / proxy startup).
@ -983,6 +988,142 @@ follow_up = client.responses.create(
</TabItem>
</Tabs>
## Encrypted Content Affinity (Multi-Region Load Balancing)
When load balancing Responses API across deployments with **different API keys** (e.g., different Azure regions or OpenAI organizations), encrypted content items (like `rs_...` reasoning items) can only be decrypted by the API key that created them.
### The Problem
```json
{
"error": {
"message": "The encrypted content for item rs_0d09d6e56879e76500699d6feee41c8197bd268aae76141f87 could not be verified. Reason: Encrypted content organization_id did not match the target organization.",
"type": "invalid_request_error",
"code": "invalid_encrypted_content"
}
}
```
This error occurs when:
1. Initial request goes to Deployment A (API Key 1) → produces encrypted item `rs_xyz`
2. Follow-up request with `rs_xyz` in input gets load balanced to Deployment B (API Key 2)
3. Deployment B cannot decrypt content created by Deployment A → **request fails**
### The Solution: `encrypted_content_affinity`
The `encrypted_content_affinity` pre-call check routes follow-up requests containing encrypted items to the originating deployment **only when necessary**
**Key Benefits:**
- ✅ **No quota reduction**: Unlike `deployment_affinity`, only pins requests that contain encrypted items
- ✅ **Bypasses rate limits**: When encrypted content requires a specific deployment, RPM/TPM limits are bypassed (the request would fail on any other deployment anyway)
- ✅ **No `previous_response_id` required**: Works by encoding `model_id` directly into item IDs
- ✅ **No cache required**: `model_id` is decoded on-the-fly — no Redis dependency, no TTL to manage
- ✅ **Globally safe**: Can be enabled for all models; non-Responses-API calls (chat, embeddings) are unaffected
### How It Works
1. **Encoding Phase** (on response):
- For each output item that contains `encrypted_content`, LiteLLM rewrites the item ID to embed the originating `model_id`: `rs_xyz` → `encitem_{base64("litellm:model_id:{model_id};item_id:rs_xyz")}`
- The original item ID is restored before forwarding the request to the upstream provider
2. **Routing Phase** (before request):
- Scans request `input` for `encitem_` prefixed IDs
- If found → decodes `model_id`, pins to originating deployment, bypasses rate limits
- If no encoded items → normal load balancing
### Configuration
<Tabs>
<TabItem value="sdk" label="Python SDK">
```python
from litellm import Router
router = Router(
model_list=[
{
"model_name": "gpt-5.1-codex",
"litellm_params": {
"model": "openai/gpt-5.1-codex",
"api_key": "org-1-api-key", # Different API key
},
"model_info": {"id": "deployment-us-east"},
},
{
"model_name": "gpt-5.1-codex",
"litellm_params": {
"model": "openai/gpt-5.1-codex",
"api_key": "org-2-api-key", # Different API key
},
"model_info": {"id": "deployment-eu-west"},
},
],
optional_pre_call_checks=["encrypted_content_affinity"],
)
# Initial request - routes to any deployment
response1 = await router.aresponses(
model="gpt-5.1-codex",
input="Explain quantum computing",
)
# Follow-up with encrypted items - automatically routes to same deployment
response2 = await router.aresponses(
model="gpt-5.1-codex",
input=response1.output, # Contains encrypted items from response1
)
```
</TabItem>
<TabItem value="proxy" label="Proxy Server">
```yaml showLineNumbers title="config.yaml"
model_list:
- model_name: gpt-5.1-codex
litellm_params:
model: azure/gpt-5.1-codex
api_base: https://eastus.openai.azure.com/
api_key: os.environ/AZURE_API_KEY_EASTUS
rpm: 600
tpm: 100000
model_info:
id: "gpt-5.1-codex-eastus"
- model_name: gpt-5.1-codex
litellm_params:
model: azure/gpt-5.1-codex
api_base: https://westeurope.openai.azure.com/
api_key: os.environ/AZURE_API_KEY_WESTEUROPE
rpm: 600
tpm: 100000
model_info:
id: "gpt-5.1-codex-westeurope"
router_settings:
routing_strategy: usage-based-routing-v2
enable_pre_call_checks: true
optional_pre_call_checks:
- encrypted_content_affinity
```
**Start proxy:**
```bash
litellm --config config.yaml
```
</TabItem>
</Tabs>
### When to Use Each Affinity Type
| Affinity Type | Use Case | Scope | Quota Impact |
|---------------|----------|-------|--------------|
| **`encrypted_content_affinity`** | **[Recommended]** Multi-region Responses API with different API keys | Only requests with tracked encrypted items | ✅ None (surgical pinning) |
| `responses_api_deployment_check` | When `previous_response_id` is available | Requests with `previous_response_id` | ✅ None |
| `session_affinity` | Session-based applications | All requests with same `session_id` | ⚠️ Reduces quota by # of sessions |
| `deployment_affinity` | Simple sticky sessions | All requests from same API key | ❌ Reduces quota by # of users |
## Calling non-Responses API endpoints (`/responses` to `/chat/completions` Bridge)
LiteLLM allows you to call non-Responses API models via a bridge to LiteLLM's `/chat/completions` endpoint. This is useful for calling Anthropic, Gemini and even non-Responses API OpenAI models.

7
docs/my-website/docs/search/index.md
View file

@ -2,7 +2,7 @@
| Feature | Supported |
|---------|-----------|
| Supported Providers | `perplexity`, `tavily`, `parallel_ai`, `exa_ai`, `brave`, `google_pse`, `dataforseo`, `firecrawl`, `searxng`, `linkup` |
| Supported Providers | `perplexity`, `tavily`, `parallel_ai`, `exa_ai`, `brave`, `google_pse`, `dataforseo`, `firecrawl`, `searxng`, `linkup`, `duckduckgo`, `searchapi` |
| Cost Tracking | ✅ |
| Logging | ✅ |
| Load Balancing | ❌ |
@ -210,7 +210,7 @@ See the [official Perplexity Search documentation](https://docs.perplexity.ai/ap
| Parameter | Type | Required | Description |
|-----------|------|----------|-------------|
| `query` | string or array | Yes | Search query. Can be a single string or array of strings |
| `search_provider` | string | Yes (SDK) | The search provider to use: `"perplexity"`, `"tavily"`, `"parallel_ai"`, `"exa_ai"`, `"brave"`, `"google_pse"`, `"dataforseo"`, `"firecrawl"`, `"searxng"`, or `"linkup"` |
| `search_provider` | string | Yes (SDK) | The search provider to use: `"perplexity"`, `"tavily"`, `"parallel_ai"`, `"exa_ai"`, `"brave"`, `"google_pse"`, `"dataforseo"`, `"firecrawl"`, `"searxng"`, `"linkup"`, `"duckduckgo"`, or `"searchapi"` |
| `search_tool_name` | string | Yes (Proxy) | Name of the search tool configured in `config.yaml` |
| `max_results` | integer | No | Maximum number of results to return (1-20). Default: 10 |
| `search_domain_filter` | array | No | List of domains to filter results (max 20 domains) |
@ -276,7 +276,8 @@ The response follows Perplexity's search format with the following structure:
| Firecrawl | `FIRECRAWL_API_KEY` | `firecrawl` |
| SearXNG | `SEARXNG_API_BASE` (required) | `searxng` |
| Linkup | `LINKUP_API_KEY` | `linkup` |
| DuckDuckGo | `DUCKDUCKGO_API_BASE` | `duckduckgo` |
| DuckDuckGo | `DUCKDUCKGO_API_BASE` | `duckduckgo` |
| SearchAPI.io | `SEARCHAPI_API_KEY` | `searchapi` |
See the individual provider documentation for detailed setup instructions and provider-specific parameters.

197
docs/my-website/docs/search/searchapi.md Normal file
View file

@ -0,0 +1,197 @@
# SearchAPI.io (Google Search)
Get started by creating a free API key via https://www.searchapi.io/.
SearchAPI.io provides access to Google Search results with a simple API. It supports all Google Search parameters including location, language, time filters, and more.
For complete documentation on all supported parameters, visit https://www.searchapi.io/docs/google.
## LiteLLM Python SDK
```python showLineNumbers title="SearchAPI.io Search"
import os
from litellm import search
os.environ["SEARCHAPI_API_KEY"] = "your-api-key"
response = search(
query="latest AI developments",
search_provider="searchapi",
max_results=10
)
# Access search results
for result in response.results:
print(f"{result.title}: {result.url}")
print(f"Snippet: {result.snippet}\n")
```
### Advanced Usage with SearchAPI.io Parameters
SearchAPI.io supports many Google Search-specific parameters:
```python showLineNumbers title="Advanced SearchAPI.io Parameters"
import os
from litellm import search
os.environ["SEARCHAPI_API_KEY"] = "your-api-key"
response = search(
query="machine learning research",
search_provider="searchapi",
max_results=10,
# Unified parameters
country="US",
search_domain_filter=["arxiv.org", "nature.com"],
# SearchAPI.io specific parameters
gl="us", # Country code
hl="en", # Interface language
time_period="last_month", # Time filter
safe="active", # SafeSearch
device="desktop", # Device type
location="New York" # Geographic location
)
```
## LiteLLM AI Gateway
### 1. Setup config.yaml
```yaml showLineNumbers title="config.yaml"
model_list:
- model_name: gpt-4
litellm_params:
model: gpt-4
api_key: os.environ/OPENAI_API_KEY
search_tools:
- search_tool_name: google-search
litellm_params:
search_provider: searchapi
api_key: os.environ/SEARCHAPI_API_KEY
```
### 2. Start the proxy
```bash
litellm --config /path/to/config.yaml
# RUNNING on http://0.0.0.0:4000
```
### 3. Test the search endpoint
```bash showLineNumbers title="Test Request"
curl http://0.0.0.0:4000/v1/search/google-search \
-H "Authorization: Bearer sk-1234" \
-H "Content-Type: application/json" \
-d '{
"query": "latest AI developments",
"max_results": 10,
"country": "US"
}'
```
## SearchAPI.io Specific Parameters
SearchAPI.io supports many Google Search parameters. Here are some commonly used ones:
| Parameter | Type | Description |
|-----------|------|-------------|
| `gl` | string | Country code (e.g., 'us', 'uk', 'de') |
| `hl` | string | Interface language (e.g., 'en', 'es', 'fr') |
| `location` | string | Geographic location (e.g., 'New York', 'London') |
| `device` | string | Device type: 'desktop', 'mobile', 'tablet' |
| `time_period` | string | Time filter: 'last_hour', 'last_day', 'last_week', 'last_month', 'last_year' |
| `time_period_min` | string | Start date (MM/DD/YYYY) |
| `time_period_max` | string | End date (MM/DD/YYYY) |
| `safe` | string | SafeSearch: 'active' or 'off' |
| `lr` | string | Language restriction (e.g., 'lang_en', 'lang_es') |
| `cr` | string | Country restriction |
| `page` | integer | Page number for pagination |
### Example with Time Filters
```python showLineNumbers title="Search with Time Filter"
response = search(
query="AI breakthroughs",
search_provider="searchapi",
max_results=10,
time_period="last_month"
)
```
### Example with Custom Date Range
```python showLineNumbers title="Search with Custom Date Range"
response = search(
query="AI research papers",
search_provider="searchapi",
max_results=10,
time_period_min="01/01/2024",
time_period_max="03/01/2024"
)
```
### Example with Location
```python showLineNumbers title="Search with Location"
response = search(
query="AI conferences",
search_provider="searchapi",
max_results=10,
location="San Francisco",
gl="us"
)
```
## Response Format
SearchAPI.io returns results in the standard LiteLLM search format:
```json
{
"object": "search",
"results": [
{
"title": "Latest AI Developments",
"url": "https://example.com/ai-news",
"snippet": "Recent breakthroughs in artificial intelligence...",
"date": "2024-01-15"
}
]
}
```
## Rate Limits
SearchAPI.io has different rate limits based on your plan:
- Free tier: 100 requests/month
- Paid plans: Higher limits available
Check your current usage at https://www.searchapi.io/dashboard.
## Error Handling
```python showLineNumbers title="Error Handling"
from litellm import search
import os
os.environ["SEARCHAPI_API_KEY"] = "your-api-key"
try:
response = search(
query="test query",
search_provider="searchapi",
max_results=10
)
print(f"Found {len(response.results)} results")
except Exception as e:
print(f"Search failed: {str(e)}")
```
## Additional Resources
- SearchAPI.io Documentation: https://www.searchapi.io/docs
- API Dashboard: https://www.searchapi.io/dashboard
- Pricing: https://www.searchapi.io/pricing

37
enterprise/litellm_enterprise/integrations/custom_guardrail.py
View file

@ -10,10 +10,15 @@ class EnterpriseCustomGuardrailHelper:
event_hook: Optional[
Union[GuardrailEventHooks, List[GuardrailEventHooks], Mode]
],
event_type: Optional[GuardrailEventHooks] = None,
) -> Optional[bool]:
"""
Assumes check for event match is done in `should_run_guardrail`
Returns True if the guardrail should be run by tag
Returns True if the guardrail should be run for this request and event_type.
Logic:
- If a request tag matches a Mode tag key, only run if event_type matches
the tag's value (the mode for that tag).
- If no request tag matches, fall back to default mode(s).
"""
from litellm.litellm_core_utils.litellm_logging import (
StandardLoggingPayloadSetup,
@ -36,11 +41,29 @@ class EnterpriseCustomGuardrailHelper:
proxy_server_request=proxy_server_request,
)
if request_tags and any(tag in event_hook.tags for tag in request_tags):
return True
elif event_hook.default and any(
tag in event_hook.default for tag in request_tags
):
# Check if any request tag matches a Mode tag key
matched_mode = None
if request_tags:
for tag in request_tags:
if tag in event_hook.tags:
matched_mode = event_hook.tags[tag]
break
if matched_mode is not None:
# Tag matched: only run if event_type matches the tag's mode value
if event_type is not None:
return event_type.value == matched_mode
return True
# No tag matched: fall back to default mode(s)
if event_hook.default is not None:
if event_type is not None:
default_list = (
event_hook.default
if isinstance(event_hook.default, list)
else [event_hook.default]
)
return event_type.value in default_list
return False
return False

53
enterprise/litellm_enterprise/proxy/audit_logging_endpoints.py
View file

@ -1,13 +1,13 @@
"""
AUDIT LOGGING
All /audit logging endpoints. Attempting to write these as CRUD endpoints.
All /audit logging endpoints. Attempting to write these as CRUD endpoints.
GET - /audit/{id} - Get audit log by id
GET - /audit - Get all audit logs
"""
from typing import Any, Dict, Optional
from typing import Any, Dict, List, Optional
#### AUDIT LOGGING ####
from fastapi import APIRouter, Depends, HTTPException, Query
@ -22,6 +22,27 @@ from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
router = APIRouter()
def _build_json_field_or_condition(json_key: str, value: str) -> Dict[str, Any]:
"""
Build an OR condition that matches a value inside a JSON column at the
given key, checking both before_value and updated_values.
Uses Prisma's JSON path filtering (PostgreSQL only).
Example result (team_id="t1"):
{"OR": [
{"before_value": {"path": ["team_id"], "string_contains": "t1"}},
{"updated_values": {"path": ["team_id"], "string_contains": "t1"}},
]}
"""
return {
"OR": [
{"before_value": {"path": [json_key], "string_contains": value}},
{"updated_values": {"path": [json_key], "string_contains": value}},
]
}
@router.get(
"/audit",
tags=["Audit Logging"],
@ -49,6 +70,14 @@ async def get_audit_logs(
),
start_date: Optional[str] = Query(None, description="Filter logs after this date"),
end_date: Optional[str] = Query(None, description="Filter logs before this date"),
object_team_id: Optional[str] = Query(
None,
description="Filter by team_id present in before_value or updated_values JSON (PostgreSQL only)",
),
object_key_hash: Optional[str] = Query(
None,
description="Filter by token (key hash) present in before_value or updated_values JSON (PostgreSQL only)",
),
# Sorting parameters
sort_by: Optional[str] = Query(
None,
@ -60,6 +89,9 @@ async def get_audit_logs(
Get all audit logs with filtering and pagination.
Returns a paginated response of audit logs matching the specified filters.
Note: object_team_id and object_key_hash use Prisma JSON path filtering,
which requires PostgreSQL.
"""
from litellm.proxy.proxy_server import prisma_client
@ -82,18 +114,29 @@ async def get_audit_logs(
if object_id:
where_conditions["object_id"] = object_id
if start_date or end_date:
date_filter = {}
date_filter: Dict[str, Any] = {}
if start_date:
date_filter["gte"] = start_date
if end_date:
date_filter["lte"] = end_date
where_conditions["updated_at"] = date_filter
# JSON field filters (PostgreSQL only) — each filter is AND'd with the
# others, but checks both before_value and updated_values internally (OR).
if object_team_id:
where_conditions["AND"] = where_conditions.get("AND", []) + [
_build_json_field_or_condition("team_id", object_team_id)
]
if object_key_hash:
where_conditions["AND"] = where_conditions.get("AND", []) + [
_build_json_field_or_condition("token", object_key_hash)
]
# Build sort conditions
order_by = {}
order_by: Dict[str, Any] = {}
if sort_by and isinstance(sort_by, str):
order_by[sort_by] = sort_order
elif sort_order and isinstance(sort_order, str):
else:
order_by["updated_at"] = sort_order # Default sort by updated_at
# Get paginated results

7
enterprise/litellm_enterprise/proxy/hooks/managed_files.py
View file

@ -589,7 +589,14 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints):
model_file_id_mapping = cast(
Optional[Dict[str, Dict[str, str]]], kwargs.get("model_file_id_mapping")
)
# model_info may be at top-level or nested under litellm_metadata
# (batch/file operations use litellm_metadata)
model_id = cast(Optional[str], kwargs.get("model_info", {}).get("id", None))
if model_id is None:
model_id = cast(
Optional[str],
kwargs.get("litellm_metadata", {}).get("model_info", {}).get("id", None),
)
mapped_file_id: Optional[str] = None
if input_file_id and model_file_id_mapping and model_id:
mapped_file_id = model_file_id_mapping.get(input_file_id, {}).get(

2
litellm-proxy-extras/litellm_proxy_extras/migrations/20260226000000_add_blocked_tools_to_object_permission/migration.sql Normal file
View file

@ -0,0 +1,2 @@
-- AlterTable
ALTER TABLE "LiteLLM_ObjectPermissionTable" ADD COLUMN "blocked_tools" TEXT[] DEFAULT ARRAY[]::TEXT[];

11
litellm-proxy-extras/litellm_proxy_extras/migrations/20260226120000_add_spend_log_tool_index/migration.sql Normal file
View file

@ -0,0 +1,11 @@
-- CreateTable
CREATE TABLE "LiteLLM_SpendLogToolIndex" (
"request_id" TEXT NOT NULL,
"tool_name" TEXT NOT NULL,
"start_time" TIMESTAMP(3) NOT NULL,
CONSTRAINT "LiteLLM_SpendLogToolIndex_pkey" PRIMARY KEY ("request_id","tool_name")
);
-- CreateIndex
CREATE INDEX "LiteLLM_SpendLogToolIndex_tool_name_start_time_idx" ON "LiteLLM_SpendLogToolIndex"("tool_name", "start_time");

46
litellm-proxy-extras/litellm_proxy_extras/schema.prisma
View file

@ -260,6 +260,7 @@ model LiteLLM_ObjectPermissionTable {
vector_stores String[] @default([])
agents String[] @default([])
agent_access_groups String[] @default([])
blocked_tools String[] @default([]) // Tool names blocked for any key/team/user with this permission
teams LiteLLM_TeamTable[]
projects LiteLLM_ProjectTable[]
verification_tokens LiteLLM_VerificationToken[]
@ -928,6 +929,16 @@ model LiteLLM_SpendLogGuardrailIndex {
@@index([policy_id, start_time])
}
// Index for fast "last N logs for tool" from SpendLogs – see how a tool is called in production
model LiteLLM_SpendLogToolIndex {
request_id String
tool_name String // matches LiteLLM_ToolTable.tool_name; join for input_policy/output_policy etc.
start_time DateTime
@@id([request_id, tool_name])
@@index([tool_name, start_time])
}
// Prompt table for storing prompt configurations
model LiteLLM_PromptTable {
id String @id @default(uuid())
@ -1065,26 +1076,31 @@ model LiteLLM_PolicyAttachmentTable {
updated_by String?
}
// Global tool registry - auto-discovered from LLM responses; admins set call_policy here
// Global tool registry - auto-discovered from LLM responses; admins set input_policy/output_policy here
model LiteLLM_ToolTable {
tool_id String @id @default(uuid())
tool_name String @unique // e.g. "huggingface_remote-mcp__dynamic_space"
origin String? // MCP server name or "user_defined"
call_policy String @default("untrusted") // "trusted" | "untrusted" | "dual_llm" | "blocked"
call_count Int @default(0) // cumulative number of times this tool was seen
assignments Json? @default("{}")
key_hash String? // hash of the virtual key that first called this tool
team_id String? // team that first called this tool
key_alias String? // human-readable alias of the virtual key
created_at DateTime @default(now())
created_by String?
updated_at DateTime @default(now()) @updatedAt
updated_by String?
tool_id String @id @default(uuid())
tool_name String @unique // e.g. "huggingface_remote-mcp__dynamic_space"
origin String? // MCP server name or "user_defined"
input_policy String @default("untrusted") // "trusted" | "untrusted" | "blocked"
output_policy String @default("untrusted") // "trusted" | "untrusted"
call_count Int @default(0) // cumulative number of times this tool was seen
assignments Json? @default("{}")
key_hash String? // hash of the virtual key that first called this tool
team_id String? // team that first called this tool
key_alias String? // human-readable alias of the virtual key
user_agent String? // user-agent of the first request that discovered this tool
last_used_at DateTime? // timestamp of the most recent call
created_at DateTime @default(now())
created_by String?
updated_at DateTime @default(now()) @updatedAt
updated_by String?
@@index([call_policy])
@@index([input_policy])
@@index([output_policy])
@@index([team_id])
}
// Per-(tool, team/key) policy overrides. When present, override replaces global tool policy for that scope.
//Unified Access Groups table for storing unified access groups
model LiteLLM_AccessGroupTable {
access_group_id String @id @default(uuid())

1
litellm/__init__.py
View file

@ -1246,6 +1246,7 @@ from .ocr.main import *
from .rag.main import *
from .search.main import *
from .realtime_api.main import _arealtime
from .responses.main import _aresponses_websocket
from .fine_tuning.main import *
from .files.main import *
from .vector_store_files.main import (

91
litellm/a2a_protocol/main.py
View file

@ -24,11 +24,7 @@ from litellm.utils import client
if TYPE_CHECKING:
from a2a.client import A2AClient as A2AClientType
from a2a.types import (
AgentCard,
SendMessageRequest,
SendStreamingMessageRequest,
)
from a2a.types import AgentCard, SendMessageRequest, SendStreamingMessageRequest
# Runtime imports with availability check
A2A_SDK_AVAILABLE = False
@ -124,13 +120,48 @@ def _get_a2a_model_info(a2a_client: Any, kwargs: Dict[str, Any]) -> str:
litellm_logging_obj.model = model
litellm_logging_obj.custom_llm_provider = custom_llm_provider
litellm_logging_obj.model_call_details["model"] = model
litellm_logging_obj.model_call_details[
"custom_llm_provider"
] = custom_llm_provider
litellm_logging_obj.model_call_details["custom_llm_provider"] = (
custom_llm_provider
)
return agent_name
async def _send_message_via_completion_bridge(
request: "SendMessageRequest",
custom_llm_provider: str,
api_base: Optional[str],
litellm_params: Dict[str, Any],
) -> LiteLLMSendMessageResponse:
"""
Route a send_message through the LiteLLM completion bridge (e.g. LangGraph, Bedrock AgentCore).
Requires request; api_base is optional for providers that derive endpoint from model.
"""
verbose_logger.info(
f"A2A using completion bridge: provider={custom_llm_provider}, api_base={api_base}"
)
from litellm.a2a_protocol.litellm_completion_bridge.handler import (
A2ACompletionBridgeHandler,
)
params = (
request.params.model_dump(mode="json")
if hasattr(request.params, "model_dump")
else dict(request.params)
)
response_dict = await A2ACompletionBridgeHandler.handle_non_streaming(
request_id=str(request.id),
params=params,
litellm_params=litellm_params,
api_base=api_base,
)
return LiteLLMSendMessageResponse.from_dict(response_dict)
@client
async def asend_message(
a2a_client: Optional["A2AClientType"] = None,
@ -193,39 +224,21 @@ async def asend_message(
```
"""
litellm_params = litellm_params or {}
logging_obj = kwargs.get("litellm_logging_obj")
trace_id = getattr(logging_obj, "litellm_trace_id", None) if logging_obj else None
custom_llm_provider = litellm_params.get("custom_llm_provider")
# Route through completion bridge if custom_llm_provider is set
if custom_llm_provider:
if request is None:
raise ValueError("request is required for completion bridge")
# api_base is optional for providers that derive endpoint from model (e.g., bedrock/agentcore)
verbose_logger.info(
f"A2A using completion bridge: provider={custom_llm_provider}, api_base={api_base}"
)
from litellm.a2a_protocol.litellm_completion_bridge.handler import (
A2ACompletionBridgeHandler,
)
# Extract params from request
params = (
request.params.model_dump(mode="json")
if hasattr(request.params, "model_dump")
else dict(request.params)
)
response_dict = await A2ACompletionBridgeHandler.handle_non_streaming(
request_id=str(request.id),
params=params,
litellm_params=litellm_params,
return await _send_message_via_completion_bridge(
request=request,
custom_llm_provider=custom_llm_provider,
api_base=api_base,
litellm_params=litellm_params,
)
# Convert to LiteLLMSendMessageResponse
return LiteLLMSendMessageResponse.from_dict(response_dict)
# Standard A2A client flow
if request is None:
raise ValueError("request is required")
@ -236,11 +249,13 @@ async def asend_message(
raise ValueError(
"Either a2a_client or api_base is required for standard A2A flow"
)
trace_id = str(uuid.uuid4())
trace_id = trace_id or str(uuid.uuid4())
extra_headers = {"X-LiteLLM-Trace-Id": trace_id}
if agent_id:
extra_headers["X-LiteLLM-Agent-Id"] = agent_id
a2a_client = await create_a2a_client(base_url=api_base, extra_headers=extra_headers)
a2a_client = await create_a2a_client(
base_url=api_base, extra_headers=extra_headers
)
# Type assertion: a2a_client is guaranteed to be non-None here
assert a2a_client is not None
@ -255,6 +270,10 @@ async def asend_message(
)
card_url = getattr(agent_card, "url", None) if agent_card else None
context_id = trace_id or str(uuid.uuid4())
if request.params.message.context_id is None:
request.params.message.context_id = context_id
# Retry loop: if connection fails due to localhost URL in agent card, retry with fixed URL
a2a_response = None
for _ in range(2): # max 2 attempts: original + 1 retry
@ -606,7 +625,9 @@ async def create_a2a_client(
if extra_headers:
httpx_client.headers.update(extra_headers)
verbose_proxy_logger.debug(f"A2A client created with extra_headers={extra_headers}")
verbose_proxy_logger.debug(
f"A2A client created with extra_headers={extra_headers}"
)
# Resolve agent card
resolver = A2ACardResolver(

107
litellm/batches/batch_utils.py
View file

@ -1,14 +1,10 @@
import json
import time
from typing import Any, List, Literal, Optional, Tuple
import httpx
import litellm
from litellm._logging import verbose_logger
from litellm._uuid import uuid
from litellm.types.llms.openai import Batch
from litellm.types.utils import CallTypes, ModelInfo, ModelResponse, Usage
from litellm.types.utils import CallTypes, ModelInfo, Usage
from litellm.utils import token_counter
@ -128,73 +124,58 @@ def calculate_vertex_ai_batch_cost_and_usage(
model_name: Optional[str] = None,
) -> Tuple[float, Usage]:
"""
Calculate both cost and usage from Vertex AI batch responses
Calculate both cost and usage from Vertex AI batch responses.
Vertex AI batch output lines have format:
{"request": ..., "status": "", "response": {"candidates": [...], "usageMetadata": {...}}}
usageMetadata contains promptTokenCount, candidatesTokenCount, totalTokenCount.
"""
from litellm.litellm_core_utils.litellm_logging import Logging
from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
VertexGeminiConfig,
)
from litellm.cost_calculator import batch_cost_calculator
total_cost = 0.0
total_tokens = 0
prompt_tokens = 0
completion_tokens = 0
for response in vertex_ai_batch_responses:
if response.get("status") == "JOB_STATE_SUCCEEDED": # Check if response was successful
# Transform Vertex AI response to OpenAI format if needed
actual_model_name = model_name or "gemini-2.0-flash-001"
# Create required arguments for the transformation method
model_response = ModelResponse()
# Ensure model_name is not None
actual_model_name = model_name or "gemini-2.5-flash"
# Create a real LiteLLM logging object
logging_obj = Logging(
for response in vertex_ai_batch_responses:
response_body = response.get("response")
if response_body is None:
continue
usage_metadata = response_body.get("usageMetadata", {})
_prompt = usage_metadata.get("promptTokenCount", 0) or 0
_completion = usage_metadata.get("candidatesTokenCount", 0) or 0
_total = usage_metadata.get("totalTokenCount", 0) or (_prompt + _completion)
line_usage = Usage(
prompt_tokens=_prompt,
completion_tokens=_completion,
total_tokens=_total,
)
try:
p_cost, c_cost = batch_cost_calculator(
usage=line_usage,
model=actual_model_name,
messages=[{"role": "user", "content": "batch_request"}],
stream=False,
call_type=CallTypes.aretrieve_batch,
start_time=time.time(),
litellm_call_id="batch_" + str(uuid.uuid4()),
function_id="batch_processing",
litellm_trace_id=str(uuid.uuid4()),
kwargs={"optional_params": {}}
)
# Add the optional_params attribute that the Vertex AI transformation expects
logging_obj.optional_params = {}
raw_response = httpx.Response(200) # Mock response object
openai_format_response = VertexGeminiConfig()._transform_google_generate_content_to_openai_model_response(
completion_response=response["response"],
model_response=model_response,
model=actual_model_name,
logging_obj=logging_obj,
raw_response=raw_response,
)
# Calculate cost using existing function
cost = litellm.completion_cost(
completion_response=openai_format_response,
custom_llm_provider="vertex_ai",
call_type=CallTypes.aretrieve_batch.value,
)
total_cost += cost
# Extract usage from the transformed response
usage_obj = getattr(openai_format_response, 'usage', None)
if usage_obj:
usage = usage_obj
else:
# Fallback: create usage from response dict
response_dict = openai_format_response.dict() if hasattr(openai_format_response, 'dict') else {}
usage = _get_batch_job_usage_from_response_body(response_dict)
total_tokens += usage.total_tokens
prompt_tokens += usage.prompt_tokens
completion_tokens += usage.completion_tokens
total_cost += p_cost + c_cost
except Exception as e:
verbose_logger.debug(
"vertex_ai batch cost calculation error for line: %s", str(e)
)
prompt_tokens += _prompt
completion_tokens += _completion
total_tokens += _total
verbose_logger.info(
"vertex_ai batch cost: cost=%s, prompt=%d, completion=%d, total=%d",
total_cost, prompt_tokens, completion_tokens, total_tokens,
)
return total_cost, Usage(
total_tokens=total_tokens,
prompt_tokens=prompt_tokens,

7
litellm/batches/main.py
View file

@ -112,6 +112,7 @@ async def acreate_batch(
metadata: Optional[Dict[str, str]] = None,
extra_headers: Optional[Dict[str, str]] = None,
extra_body: Optional[Dict[str, str]] = None,
output_expires_after: Optional[Dict[str, Any]] = None,
**kwargs,
) -> LiteLLMBatch:
"""
@ -133,6 +134,7 @@ async def acreate_batch(
metadata,
extra_headers,
extra_body,
output_expires_after,
**kwargs,
)
@ -152,7 +154,7 @@ async def acreate_batch(
@client
def create_batch(
def create_batch( # noqa: PLR0915
completion_window: Literal["24h"],
endpoint: Literal["/v1/chat/completions", "/v1/embeddings", "/v1/completions"],
input_file_id: str,
@ -160,6 +162,7 @@ def create_batch(
metadata: Optional[Dict[str, str]] = None,
extra_headers: Optional[Dict[str, str]] = None,
extra_body: Optional[Dict[str, str]] = None,
output_expires_after: Optional[Dict[str, Any]] = None,
**kwargs,
) -> Union[LiteLLMBatch, Coroutine[Any, Any, LiteLLMBatch]]:
"""
@ -215,6 +218,8 @@ def create_batch(
extra_headers=extra_headers,
extra_body=extra_body,
)
if output_expires_after is not None:
_create_batch_request["output_expires_after"] = output_expires_after
if model is not None:
provider_config = ProviderConfigManager.get_provider_batches_config(
model=model,

347
litellm/files/main.py
View file

@ -7,7 +7,6 @@ https://platform.openai.com/docs/api-reference/files
import asyncio
import contextvars
import os
import time
import uuid as uuid_module
from functools import partial
@ -20,10 +19,12 @@ from litellm import get_secret_str
from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
from litellm.llms.anthropic.files.handler import AnthropicFilesHandler
from litellm.llms.azure.common_utils import get_azure_credentials
from litellm.llms.azure.files.handler import AzureOpenAIFilesAPI
from litellm.llms.bedrock.files.handler import BedrockFilesHandler
from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler
from litellm.llms.custom_httpx.llm_http_handler import BaseLLMHTTPHandler
from litellm.llms.openai.common_utils import get_openai_credentials
from litellm.llms.openai.openai import FileDeleted, FileObject, OpenAIFilesAPI
from litellm.llms.vertex_ai.files.handler import VertexAIFilesHandler
from litellm.types.llms.openai import (
@ -185,95 +186,36 @@ def create_file(
timeout=timeout,
)
elif custom_llm_provider in OPENAI_COMPATIBLE_BATCH_AND_FILES_PROVIDERS:
# for deepinfra/perplexity/anyscale/groq we check in get_llm_provider and pass in the api base from there
api_base = (
optional_params.api_base
or litellm.api_base
or os.getenv("OPENAI_BASE_URL")
or os.getenv("OPENAI_API_BASE")
or "https://api.openai.com/v1"
openai_creds = get_openai_credentials(
api_base=optional_params.api_base,
api_key=optional_params.api_key,
organization=optional_params.organization,
)
organization = (
optional_params.organization
or litellm.organization
or os.getenv("OPENAI_ORGANIZATION", None)
or None # default - https://github.com/openai/openai-python/blob/284c1799070c723c6a553337134148a7ab088dd8/openai/util.py#L105
)
# set API KEY
api_key = (
optional_params.api_key
or litellm.api_key # for deepinfra/perplexity/anyscale we check in get_llm_provider and pass in the api key from there
or litellm.openai_key
or os.getenv("OPENAI_API_KEY")
)
response = openai_files_instance.create_file(
_is_async=_is_async,
api_base=api_base,
api_key=api_key,
api_base=openai_creds.api_base,
api_key=openai_creds.api_key,
timeout=timeout,
max_retries=optional_params.max_retries,
organization=organization,
organization=openai_creds.organization,
create_file_data=_create_file_request,
)
elif custom_llm_provider == "azure":
api_base = optional_params.api_base or litellm.api_base or get_secret_str("AZURE_API_BASE") # type: ignore
api_version = (
optional_params.api_version
or litellm.api_version
or get_secret_str("AZURE_API_VERSION")
) # type: ignore
api_key = (
optional_params.api_key
or litellm.api_key
or litellm.azure_key
or get_secret_str("AZURE_OPENAI_API_KEY")
or get_secret_str("AZURE_API_KEY")
) # type: ignore
extra_body = optional_params.get("extra_body", {})
if extra_body is not None:
extra_body.pop("azure_ad_token", None)
else:
get_secret_str("AZURE_AD_TOKEN") # type: ignore
azure_creds = get_azure_credentials(
api_base=optional_params.api_base,
api_key=optional_params.api_key,
api_version=optional_params.api_version,
)
response = azure_files_instance.create_file(
_is_async=_is_async,
api_base=api_base,
api_key=api_key,
api_version=api_version,
api_base=azure_creds.api_base,
api_key=azure_creds.api_key,
api_version=azure_creds.api_version,
timeout=timeout,
max_retries=optional_params.max_retries,
create_file_data=_create_file_request,
litellm_params=litellm_params_dict,
)
elif custom_llm_provider == "vertex_ai":
api_base = optional_params.api_base or ""
vertex_ai_project = (
optional_params.vertex_project
or litellm.vertex_project
or get_secret_str("VERTEXAI_PROJECT")
)
vertex_ai_location = (
optional_params.vertex_location
or litellm.vertex_location
or get_secret_str("VERTEXAI_LOCATION")
)
vertex_credentials = optional_params.vertex_credentials or get_secret_str(
"VERTEXAI_CREDENTIALS"
)
response = vertex_ai_files_instance.create_file(
_is_async=_is_async,
api_base=api_base,
vertex_project=vertex_ai_project,
vertex_location=vertex_ai_location,
vertex_credentials=vertex_credentials,
timeout=timeout,
max_retries=optional_params.max_retries,
create_file_data=_create_file_request,
)
else:
raise litellm.exceptions.BadRequestError(
message="LiteLLM doesn't support {} for 'create_file'. Only ['openai', 'azure', 'vertex_ai', 'manus'] are supported.".format(
@ -295,7 +237,7 @@ def create_file(
@client
async def afile_retrieve(
file_id: str,
custom_llm_provider: Literal["openai", "azure", "gemini", "hosted_vllm", "manus"] = "openai",
custom_llm_provider: Literal["openai", "azure", "gemini", "vertex_ai", "hosted_vllm", "manus"] = "openai",
extra_headers: Optional[Dict[str, str]] = None,
extra_body: Optional[Dict[str, str]] = None,
**kwargs,
@ -336,7 +278,7 @@ async def afile_retrieve(
@client
def file_retrieve(
file_id: str,
custom_llm_provider: Literal["openai", "azure", "hosted_vllm", "manus"] = "openai",
custom_llm_provider: Literal["openai", "azure", "gemini", "vertex_ai", "hosted_vllm", "manus"] = "openai",
extra_headers: Optional[Dict[str, str]] = None,
extra_body: Optional[Dict[str, str]] = None,
**kwargs,
@ -367,64 +309,31 @@ def file_retrieve(
_is_async = kwargs.pop("is_async", False) is True
if custom_llm_provider in OPENAI_COMPATIBLE_BATCH_AND_FILES_PROVIDERS:
# for deepinfra/perplexity/anyscale/groq we check in get_llm_provider and pass in the api base from there
api_base = (
optional_params.api_base
or litellm.api_base
or os.getenv("OPENAI_BASE_URL")
or os.getenv("OPENAI_API_BASE")
or "https://api.openai.com/v1"
openai_creds = get_openai_credentials(
api_base=optional_params.api_base,
api_key=optional_params.api_key,
organization=optional_params.organization,
)
organization = (
optional_params.organization
or litellm.organization
or os.getenv("OPENAI_ORGANIZATION", None)
or None # default - https://github.com/openai/openai-python/blob/284c1799070c723c6a553337134148a7ab088dd8/openai/util.py#L105
)
# set API KEY
api_key = (
optional_params.api_key
or litellm.api_key # for deepinfra/perplexity/anyscale we check in get_llm_provider and pass in the api key from there
or litellm.openai_key
or os.getenv("OPENAI_API_KEY")
)
response = openai_files_instance.retrieve_file(
file_id=file_id,
_is_async=_is_async,
api_base=api_base,
api_key=api_key,
api_base=openai_creds.api_base,
api_key=openai_creds.api_key,
timeout=timeout,
max_retries=optional_params.max_retries,
organization=organization,
organization=openai_creds.organization,
)
elif custom_llm_provider == "azure":
api_base = optional_params.api_base or litellm.api_base or get_secret_str("AZURE_API_BASE") # type: ignore
api_version = (
optional_params.api_version
or litellm.api_version
or get_secret_str("AZURE_API_VERSION")
) # type: ignore
api_key = (
optional_params.api_key
or litellm.api_key
or litellm.azure_key
or get_secret_str("AZURE_OPENAI_API_KEY")
or get_secret_str("AZURE_API_KEY")
) # type: ignore
extra_body = optional_params.get("extra_body", {})
if extra_body is not None:
extra_body.pop("azure_ad_token", None)
else:
get_secret_str("AZURE_AD_TOKEN") # type: ignore
azure_creds = get_azure_credentials(
api_base=optional_params.api_base,
api_key=optional_params.api_key,
api_version=optional_params.api_version,
)
response = azure_files_instance.retrieve_file(
_is_async=_is_async,
api_base=api_base,
api_key=api_key,
api_version=api_version,
api_base=azure_creds.api_base,
api_key=azure_creds.api_key,
api_version=azure_creds.api_version,
timeout=timeout,
max_retries=optional_params.max_retries,
file_id=file_id,
@ -576,63 +485,31 @@ def file_delete(
timeout = 600.0
_is_async = kwargs.pop("is_async", False) is True
if custom_llm_provider in OPENAI_COMPATIBLE_BATCH_AND_FILES_PROVIDERS:
# for deepinfra/perplexity/anyscale/groq we check in get_llm_provider and pass in the api base from there
api_base = (
optional_params.api_base
or litellm.api_base
or os.getenv("OPENAI_BASE_URL")
or os.getenv("OPENAI_API_BASE")
or "https://api.openai.com/v1"
)
organization = (
optional_params.organization
or litellm.organization
or os.getenv("OPENAI_ORGANIZATION", None)
or None # default - https://github.com/openai/openai-python/blob/284c1799070c723c6a553337134148a7ab088dd8/openai/util.py#L105
)
# set API KEY
api_key = (
optional_params.api_key
or litellm.api_key # for deepinfra/perplexity/anyscale we check in get_llm_provider and pass in the api key from there
or litellm.openai_key
or os.getenv("OPENAI_API_KEY")
openai_creds = get_openai_credentials(
api_base=optional_params.api_base,
api_key=optional_params.api_key,
organization=optional_params.organization,
)
response = openai_files_instance.delete_file(
file_id=file_id,
_is_async=_is_async,
api_base=api_base,
api_key=api_key,
api_base=openai_creds.api_base,
api_key=openai_creds.api_key,
timeout=timeout,
max_retries=optional_params.max_retries,
organization=organization,
organization=openai_creds.organization,
)
elif custom_llm_provider == "azure":
api_base = optional_params.api_base or litellm.api_base or get_secret_str("AZURE_API_BASE") # type: ignore
api_version = (
optional_params.api_version
or litellm.api_version
or get_secret_str("AZURE_API_VERSION")
) # type: ignore
api_key = (
optional_params.api_key
or litellm.api_key
or litellm.azure_key
or get_secret_str("AZURE_OPENAI_API_KEY")
or get_secret_str("AZURE_API_KEY")
) # type: ignore
extra_body = optional_params.get("extra_body", {})
if extra_body is not None:
extra_body.pop("azure_ad_token", None)
else:
get_secret_str("AZURE_AD_TOKEN") # type: ignore
azure_creds = get_azure_credentials(
api_base=optional_params.api_base,
api_key=optional_params.api_key,
api_version=optional_params.api_version,
)
response = azure_files_instance.delete_file(
_is_async=_is_async,
api_base=api_base,
api_key=api_key,
api_version=api_version,
api_base=azure_creds.api_base,
api_key=azure_creds.api_key,
api_version=azure_creds.api_version,
timeout=timeout,
max_retries=optional_params.max_retries,
file_id=file_id,
@ -815,64 +692,31 @@ def file_list(
)
return response
elif custom_llm_provider in OPENAI_COMPATIBLE_BATCH_AND_FILES_PROVIDERS:
# for deepinfra/perplexity/anyscale/groq we check in get_llm_provider and pass in the api base from there
api_base = (
optional_params.api_base
or litellm.api_base
or os.getenv("OPENAI_BASE_URL")
or os.getenv("OPENAI_API_BASE")
or "https://api.openai.com/v1"
openai_creds = get_openai_credentials(
api_base=optional_params.api_base,
api_key=optional_params.api_key,
organization=optional_params.organization,
)
organization = (
optional_params.organization
or litellm.organization
or os.getenv("OPENAI_ORGANIZATION", None)
or None # default - https://github.com/openai/openai-python/blob/284c1799070c723c6a553337134148a7ab088dd8/openai/util.py#L105
)
# set API KEY
api_key = (
optional_params.api_key
or litellm.api_key # for deepinfra/perplexity/anyscale we check in get_llm_provider and pass in the api key from there
or litellm.openai_key
or os.getenv("OPENAI_API_KEY")
)
response = openai_files_instance.list_files(
purpose=purpose,
_is_async=_is_async,
api_base=api_base,
api_key=api_key,
api_base=openai_creds.api_base,
api_key=openai_creds.api_key,
timeout=timeout,
max_retries=optional_params.max_retries,
organization=organization,
organization=openai_creds.organization,
)
elif custom_llm_provider == "azure":
api_base = optional_params.api_base or litellm.api_base or get_secret_str("AZURE_API_BASE") # type: ignore
api_version = (
optional_params.api_version
or litellm.api_version
or get_secret_str("AZURE_API_VERSION")
) # type: ignore
api_key = (
optional_params.api_key
or litellm.api_key
or litellm.azure_key
or get_secret_str("AZURE_OPENAI_API_KEY")
or get_secret_str("AZURE_API_KEY")
) # type: ignore
extra_body = optional_params.get("extra_body", {})
if extra_body is not None:
extra_body.pop("azure_ad_token", None)
else:
get_secret_str("AZURE_AD_TOKEN") # type: ignore
azure_creds = get_azure_credentials(
api_base=optional_params.api_base,
api_key=optional_params.api_key,
api_version=optional_params.api_version,
)
response = azure_files_instance.list_files(
_is_async=_is_async,
api_base=api_base,
api_key=api_key,
api_version=api_version,
api_base=azure_creds.api_base,
api_key=azure_creds.api_key,
api_version=azure_creds.api_version,
timeout=timeout,
max_retries=optional_params.max_retries,
purpose=purpose,
@ -1003,64 +847,31 @@ def file_content(
return response
if custom_llm_provider in OPENAI_COMPATIBLE_BATCH_AND_FILES_PROVIDERS:
# for deepinfra/perplexity/anyscale/groq we check in get_llm_provider and pass in the api base from there
api_base = (
optional_params.api_base
or litellm.api_base
or os.getenv("OPENAI_BASE_URL")
or os.getenv("OPENAI_API_BASE")
or "https://api.openai.com/v1"
openai_creds = get_openai_credentials(
api_base=optional_params.api_base,
api_key=optional_params.api_key,
organization=optional_params.organization,
)
organization = (
optional_params.organization
or litellm.organization
or os.getenv("OPENAI_ORGANIZATION", None)
or None # default - https://github.com/openai/openai-python/blob/284c1799070c723c6a553337134148a7ab088dd8/openai/util.py#L105
)
# set API KEY
api_key = (
optional_params.api_key
or litellm.api_key # for deepinfra/perplexity/anyscale we check in get_llm_provider and pass in the api key from there
or litellm.openai_key
or os.getenv("OPENAI_API_KEY")
)
response = openai_files_instance.file_content(
_is_async=_is_async,
file_content_request=_file_content_request,
api_base=api_base,
api_key=api_key,
api_base=openai_creds.api_base,
api_key=openai_creds.api_key,
timeout=timeout,
max_retries=optional_params.max_retries,
organization=organization,
organization=openai_creds.organization,
)
elif custom_llm_provider == "azure":
api_base = optional_params.api_base or litellm.api_base or get_secret_str("AZURE_API_BASE") # type: ignore
api_version = (
optional_params.api_version
or litellm.api_version
or get_secret_str("AZURE_API_VERSION")
) # type: ignore
api_key = (
optional_params.api_key
or litellm.api_key
or litellm.azure_key
or get_secret_str("AZURE_OPENAI_API_KEY")
or get_secret_str("AZURE_API_KEY")
) # type: ignore
extra_body = optional_params.get("extra_body", {})
if extra_body is not None:
extra_body.pop("azure_ad_token", None)
else:
get_secret_str("AZURE_AD_TOKEN") # type: ignore
azure_creds = get_azure_credentials(
api_base=optional_params.api_base,
api_key=optional_params.api_key,
api_version=optional_params.api_version,
)
response = azure_files_instance.file_content(
_is_async=_is_async,
api_base=api_base,
api_key=api_key,
api_version=api_version,
api_base=azure_creds.api_base,
api_key=azure_creds.api_key,
api_version=azure_creds.api_version,
timeout=timeout,
max_retries=optional_params.max_retries,
file_content_request=_file_content_request,

55
litellm/fine_tuning/main.py
View file

@ -34,6 +34,44 @@ vertex_fine_tuning_apis_instance = VertexFineTuningAPI()
#################################################
def _prepare_azure_extra_body(
extra_body: Optional[Dict[str, Any]],
kwargs: Dict[str, Any],
azure_specific_hyperparams: Dict[str, Any],
) -> Dict[str, Any]:
"""
Prepare extra_body for Azure fine-tuning API by combining Azure-specific parameters.
Azure fine-tuning API accepts additional parameters beyond the standard OpenAI spec:
- trainingType: Type of training (e.g., 1 for supervised fine-tuning)
- prompt_loss_weight: Weight for prompt loss in training
These parameters must be passed in the extra_body field when calling the Azure OpenAI SDK.
Args:
extra_body: Optional existing extra_body dict
kwargs: Request kwargs that may contain Azure-specific parameters
azure_specific_hyperparams: Dict of Azure-specific hyperparameters already extracted
Returns:
Dict containing all Azure-specific parameters to be passed in extra_body
"""
if extra_body is None:
extra_body = {}
# Azure-specific root-level parameters
azure_specific_params = ["trainingType"]
for param in azure_specific_params:
if param in kwargs:
extra_body[param] = kwargs[param]
# Add Azure-specific hyperparameters
if azure_specific_hyperparams:
extra_body.update(azure_specific_hyperparams)
return extra_body
@client
async def acreate_fine_tuning_job(
model: str,
@ -114,6 +152,15 @@ def create_fine_tuning_job(
# handle hyperparameters
hyperparameters = hyperparameters or {} # original hyperparameters
# For Azure, extract Azure-specific hyperparameters before creating OpenAI-spec hyperparameters
azure_specific_hyperparams = {}
if custom_llm_provider == "azure":
azure_hyperparameter_keys = ["prompt_loss_weight"]
for key in azure_hyperparameter_keys:
if key in hyperparameters:
azure_specific_hyperparams[key] = hyperparameters.pop(key)
_oai_hyperparameters: Hyperparameters = Hyperparameters(
**hyperparameters
) # Typed Hyperparameters for OpenAI Spec
@ -207,6 +254,10 @@ def create_fine_tuning_job(
extra_body.pop("azure_ad_token", None)
else:
get_secret_str("AZURE_AD_TOKEN") # type: ignore
# Prepare Azure-specific parameters for extra_body
extra_body = _prepare_azure_extra_body(extra_body, kwargs, azure_specific_hyperparams)
create_fine_tuning_job_data = FineTuningJobCreate(
model=model,
training_file=training_file,
@ -220,6 +271,10 @@ def create_fine_tuning_job(
create_fine_tuning_job_data_dict = create_fine_tuning_job_data.model_dump(
exclude_none=True
)
# Add extra_body if it has Azure-specific parameters
if extra_body:
create_fine_tuning_job_data_dict["extra_body"] = extra_body
response = azure_fine_tuning_apis_instance.create_fine_tuning_job(
api_base=api_base,

22
litellm/integrations/custom_guardrail.py
View file

@ -235,8 +235,13 @@ class CustomGuardrail(CustomLogger):
list(event_hook.tags.values()), supported_event_hooks
)
if event_hook.default:
default_list = (
event_hook.default
if isinstance(event_hook.default, list)
else [event_hook.default]
)
_validate_event_hook_list_is_in_supported_event_hooks(
[event_hook.default], supported_event_hooks
default_list, supported_event_hooks
)
elif isinstance(event_hook, GuardrailEventHooks):
if event_hook not in supported_event_hooks:
@ -415,7 +420,7 @@ class CustomGuardrail(CustomLogger):
"Setting tag-based guardrails is only available in litellm-enterprise. You must be a premium user to use this feature."
)
result = EnterpriseCustomGuardrailHelper._should_run_if_mode_by_tag(
data, self.event_hook
data, self.event_hook, event_type
)
if result is not None:
return result
@ -442,7 +447,7 @@ class CustomGuardrail(CustomLogger):
"Setting tag-based guardrails is only available in litellm-enterprise. You must be a premium user to use this feature."
)
result = EnterpriseCustomGuardrailHelper._should_run_if_mode_by_tag(
data, self.event_hook
data, self.event_hook, event_type
)
if result is not None:
return result
@ -461,7 +466,16 @@ class CustomGuardrail(CustomLogger):
if isinstance(self.event_hook, list):
return event_type.value in self.event_hook
if isinstance(self.event_hook, Mode):
return event_type.value in self.event_hook.tags.values()
if event_type.value in self.event_hook.tags.values():
return True
if self.event_hook.default:
default_list = (
self.event_hook.default
if isinstance(self.event_hook.default, list)
else [self.event_hook.default]
)
return event_type.value in default_list
return False
return self.event_hook == event_type.value
def get_guardrail_dynamic_request_body_params(self, request_data: dict) -> dict:

6
litellm/integrations/helicone.py
View file

@ -167,12 +167,12 @@ class HeliconeLogger:
if "claude" in model and not is_vertex_ai:
url = f"{self.api_base}/anthropic/v1/log"
provider_url = "https://api.anthropic.com/v1/messages"
elif "gemini" in model:
url = f"{self.api_base}/custom/v1/log"
provider_url = "https://generativelanguage.googleapis.com/v1beta"
elif is_vertex_ai:
url = f"{self.api_base}/custom/v1/log"
provider_url = "https://aiplatform.googleapis.com/v1"
elif "gemini" in model:
url = f"{self.api_base}/custom/v1/log"
provider_url = "https://generativelanguage.googleapis.com/v1beta"
headers = {
"Authorization": f"Bearer {self.key}",
"Content-Type": "application/json",

45
litellm/litellm_core_utils/exception_mapping_utils.py
View file

@ -1,9 +1,9 @@
import json
import re
import traceback
from typing import Any, Optional
import httpx
import re
import litellm
from litellm._logging import verbose_logger
@ -443,6 +443,27 @@ def exception_type( # type: ignore # noqa: PLR0915
response=getattr(original_exception, "response", None),
litellm_debug_info=extra_information,
)
elif "invalid_encrypted_content" in error_str or "could not be verified" in error_str:
exception_mapping_worked = True
helpful_message = (
f"{exception_provider} - {message}\n\n"
" This error occurs when load balancing Responses API across deployments with different API keys.\n"
" Encrypted content is tied to the organization that created it and cannot be decrypted by other organizations.\n\n"
" Solution: Enable 'encrypted_content_affinity' to route follow-up requests to the correct deployment:\n\n"
" router_settings:\n"
" enable_pre_call_checks: true\n"
" optional_pre_call_checks:\n"
" - encrypted_content_affinity\n\n"
" Learn more: https://docs.litellm.ai/docs/response_api#encrypted-content-affinity-multi-region-load-balancing"
)
raise BadRequestError(
message=helpful_message,
llm_provider=custom_llm_provider,
model=model,
response=getattr(original_exception, "response", None),
litellm_debug_info=extra_information,
body=getattr(original_exception, "body", None),
)
elif (
"invalid_request_error" in error_str
and "Incorrect API key provided" not in error_str
@ -2126,7 +2147,27 @@ def exception_type( # type: ignore # noqa: PLR0915
extra_information=extra_information,
original_exception=original_exception,
)
elif azure_error_code == "invalid_encrypted_content" or "could not be verified" in error_str:
exception_mapping_worked = True
helpful_message = (
f"AzureException - {message}\n\n"
"This error occurs when load balancing Responses API across deployments with different API keys.\n"
" Encrypted content is tied to the organization that created it and cannot be decrypted by other organizations.\n\n"
" Solution: Enable 'encrypted_content_affinity' to route follow-up requests to the correct deployment:\n\n"
" router_settings:\n"
" enable_pre_call_checks: true\n"
" optional_pre_call_checks:\n"
" - encrypted_content_affinity\n\n"
" Learn more: https://docs.litellm.ai/docs/response_api#encrypted-content-affinity-multi-region-load-balancing"
)
raise BadRequestError(
message=helpful_message,
llm_provider="azure",
model=model,
litellm_debug_info=extra_information,
response=getattr(original_exception, "response", None),
body=getattr(original_exception, "body", None),
)
elif "invalid_request_error" in error_str:
exception_mapping_worked = True
raise BadRequestError(

8
litellm/litellm_core_utils/get_litellm_params.py
View file

@ -1,6 +1,5 @@
from typing import Optional
# Pre-define optional kwargs keys as frozenset for O(1) lookups
# These are extracted from kwargs only if present, avoiding unnecessary .get() calls
_OPTIONAL_KWARGS_KEYS = frozenset({
@ -95,6 +94,13 @@ def get_litellm_params(
litellm_request_debug: Optional[bool] = None,
**kwargs,
) -> dict:
# Derive litellm_session_id / litellm_trace_id from metadata when not provided (call chaining)
_meta = metadata or {}
if litellm_session_id is None:
litellm_session_id = _meta.get("session_id") or _meta.get("trace_id")
if litellm_trace_id is None:
litellm_trace_id = _meta.get("trace_id") or _meta.get("session_id")
# Build base dict with explicit parameters (always included)
litellm_params = {
"acompletion": acompletion,

306
litellm/litellm_core_utils/litellm_logging.py
View file

@ -133,8 +133,8 @@ from ..integrations.azure_sentinel.azure_sentinel import AzureSentinelLogger
from ..integrations.azure_storage.azure_storage import AzureBlobStorageLogger
from ..integrations.custom_prompt_management import CustomPromptManagement
from ..integrations.datadog.datadog import DataDogLogger
from ..integrations.datadog.datadog_metrics import DatadogMetricsLogger
from ..integrations.datadog.datadog_llm_obs import DataDogLLMObsLogger
from ..integrations.datadog.datadog_metrics import DatadogMetricsLogger
from ..integrations.dotprompt import DotpromptManager
from ..integrations.dynamodb import DyanmoDBLogger
from ..integrations.galileo import GalileoObserve
@ -352,9 +352,9 @@ class Logging(LiteLLMLoggingBaseClass):
)
self.function_id = function_id
self.streaming_chunks: List[Any] = [] # for generating complete stream response
self.sync_streaming_chunks: List[
Any
] = [] # for generating complete stream response
self.sync_streaming_chunks: List[Any] = (
[]
) # for generating complete stream response
self.log_raw_request_response = log_raw_request_response
# Initialize dynamic callbacks
@ -746,9 +746,9 @@ class Logging(LiteLLMLoggingBaseClass):
prompt_spec=prompt_spec,
dynamic_callback_params=dynamic_callback_params,
):
self.model_call_details[
"prompt_integration"
] = logger.__class__.__name__
self.model_call_details["prompt_integration"] = (
logger.__class__.__name__
)
return logger
except Exception:
# If check fails, continue to next logger
@ -816,9 +816,9 @@ class Logging(LiteLLMLoggingBaseClass):
if anthropic_cache_control_logger := AnthropicCacheControlHook.get_custom_logger_for_anthropic_cache_control_hook(
non_default_params
):
self.model_call_details[
"prompt_integration"
] = anthropic_cache_control_logger.__class__.__name__
self.model_call_details["prompt_integration"] = (
anthropic_cache_control_logger.__class__.__name__
)
return anthropic_cache_control_logger
#########################################################
@ -830,9 +830,9 @@ class Logging(LiteLLMLoggingBaseClass):
internal_usage_cache=None,
llm_router=None,
)
self.model_call_details[
"prompt_integration"
] = vector_store_custom_logger.__class__.__name__
self.model_call_details["prompt_integration"] = (
vector_store_custom_logger.__class__.__name__
)
# Add to global callbacks so post-call hooks are invoked
if (
vector_store_custom_logger
@ -892,9 +892,9 @@ class Logging(LiteLLMLoggingBaseClass):
model
): # if model name was changes pre-call, overwrite the initial model call name with the new one
self.model_call_details["model"] = model
self.model_call_details["litellm_params"][
"api_base"
] = self._get_masked_api_base(additional_args.get("api_base", ""))
self.model_call_details["litellm_params"]["api_base"] = (
self._get_masked_api_base(additional_args.get("api_base", ""))
)
def pre_call(self, input, api_key, model=None, additional_args={}): # noqa: PLR0915
# Log the exact input to the LLM API
@ -923,10 +923,10 @@ class Logging(LiteLLMLoggingBaseClass):
try:
# [Non-blocking Extra Debug Information in metadata]
if turn_off_message_logging is True:
_metadata[
"raw_request"
] = "redacted by litellm. \
_metadata["raw_request"] = (
"redacted by litellm. \
'litellm.turn_off_message_logging=True'"
)
else:
curl_command = self._get_request_curl_command(
api_base=additional_args.get("api_base", ""),
@ -937,34 +937,34 @@ class Logging(LiteLLMLoggingBaseClass):
_metadata["raw_request"] = str(curl_command)
# split up, so it's easier to parse in the UI
self.model_call_details[
"raw_request_typed_dict"
] = RawRequestTypedDict(
raw_request_api_base=str(
additional_args.get("api_base") or ""
),
raw_request_body=self._get_raw_request_body(
additional_args.get("complete_input_dict", {})
),
# NOTE: setting ignore_sensitive_headers to True will cause
# the Authorization header to be leaked when calls to the health
# endpoint are made and fail.
raw_request_headers=self._get_masked_headers(
additional_args.get("headers", {}) or {},
),
error=None,
self.model_call_details["raw_request_typed_dict"] = (
RawRequestTypedDict(
raw_request_api_base=str(
additional_args.get("api_base") or ""
),
raw_request_body=self._get_raw_request_body(
additional_args.get("complete_input_dict", {})
),
# NOTE: setting ignore_sensitive_headers to True will cause
# the Authorization header to be leaked when calls to the health
# endpoint are made and fail.
raw_request_headers=self._get_masked_headers(
additional_args.get("headers", {}) or {},
),
error=None,
)
)
except Exception as e:
self.model_call_details[
"raw_request_typed_dict"
] = RawRequestTypedDict(
error=str(e),
self.model_call_details["raw_request_typed_dict"] = (
RawRequestTypedDict(
error=str(e),
)
)
_metadata[
"raw_request"
] = "Unable to Log \
_metadata["raw_request"] = (
"Unable to Log \
raw request: {}".format(
str(e)
str(e)
)
)
if getattr(self, "logger_fn", None) and callable(self.logger_fn):
try:
@ -1265,13 +1265,13 @@ class Logging(LiteLLMLoggingBaseClass):
for callback in callbacks:
try:
if isinstance(callback, CustomLogger):
response: Optional[
MCPPostCallResponseObject
] = await callback.async_post_mcp_tool_call_hook(
kwargs=kwargs,
response_obj=post_mcp_tool_call_response_obj,
start_time=start_time,
end_time=end_time,
response: Optional[MCPPostCallResponseObject] = (
await callback.async_post_mcp_tool_call_hook(
kwargs=kwargs,
response_obj=post_mcp_tool_call_response_obj,
start_time=start_time,
end_time=end_time,
)
)
######################################################################
# if any of the callbacks modify the response, use the modified response
@ -1466,9 +1466,9 @@ class Logging(LiteLLMLoggingBaseClass):
verbose_logger.debug(
f"response_cost_failure_debug_information: {debug_info}"
)
self.model_call_details[
"response_cost_failure_debug_information"
] = debug_info
self.model_call_details["response_cost_failure_debug_information"] = (
debug_info
)
return None
try:
@ -1494,9 +1494,9 @@ class Logging(LiteLLMLoggingBaseClass):
verbose_logger.debug(
f"response_cost_failure_debug_information: {debug_info}"
)
self.model_call_details[
"response_cost_failure_debug_information"
] = debug_info
self.model_call_details["response_cost_failure_debug_information"] = (
debug_info
)
return None
@ -1652,10 +1652,8 @@ class Logging(LiteLLMLoggingBaseClass):
result=logging_result
)
self.model_call_details[
"standard_logging_object"
] = self._build_standard_logging_payload(
logging_result, start_time, end_time
self.model_call_details["standard_logging_object"] = (
self._build_standard_logging_payload(logging_result, start_time, end_time)
)
if (
@ -1734,9 +1732,9 @@ class Logging(LiteLLMLoggingBaseClass):
end_time = datetime.datetime.now()
if self.completion_start_time is None:
self.completion_start_time = end_time
self.model_call_details[
"completion_start_time"
] = self.completion_start_time
self.model_call_details["completion_start_time"] = (
self.completion_start_time
)
self.model_call_details["log_event_type"] = "successful_api_call"
self.model_call_details["end_time"] = end_time
@ -1773,10 +1771,10 @@ class Logging(LiteLLMLoggingBaseClass):
end_time=end_time,
)
elif isinstance(result, dict) or isinstance(result, list):
self.model_call_details[
"standard_logging_object"
] = self._build_standard_logging_payload(
result, start_time, end_time
self.model_call_details["standard_logging_object"] = (
self._build_standard_logging_payload(
result, start_time, end_time
)
)
if (
standard_logging_payload := self.model_call_details.get(
@ -1785,9 +1783,9 @@ class Logging(LiteLLMLoggingBaseClass):
) is not None:
emit_standard_logging_payload(standard_logging_payload)
elif standard_logging_object is not None:
self.model_call_details[
"standard_logging_object"
] = standard_logging_object
self.model_call_details["standard_logging_object"] = (
standard_logging_object
)
else:
self.model_call_details["response_cost"] = None
@ -1945,17 +1943,17 @@ class Logging(LiteLLMLoggingBaseClass):
verbose_logger.debug(
"Logging Details LiteLLM-Success Call streaming complete"
)
self.model_call_details[
"complete_streaming_response"
] = complete_streaming_response
self.model_call_details[
"response_cost"
] = self._response_cost_calculator(result=complete_streaming_response)
self.model_call_details["complete_streaming_response"] = (
complete_streaming_response
)
self.model_call_details["response_cost"] = (
self._response_cost_calculator(result=complete_streaming_response)
)
## STANDARDIZED LOGGING PAYLOAD
self.model_call_details[
"standard_logging_object"
] = self._build_standard_logging_payload(
complete_streaming_response, start_time, end_time
self.model_call_details["standard_logging_object"] = (
self._build_standard_logging_payload(
complete_streaming_response, start_time, end_time
)
)
if (
standard_logging_payload := self.model_call_details.get(
@ -2289,10 +2287,10 @@ class Logging(LiteLLMLoggingBaseClass):
)
else:
if self.stream and complete_streaming_response:
self.model_call_details[
"complete_response"
] = self.model_call_details.get(
"complete_streaming_response", {}
self.model_call_details["complete_response"] = (
self.model_call_details.get(
"complete_streaming_response", {}
)
)
result = self.model_call_details["complete_response"]
openMeterLogger.log_success_event(
@ -2316,10 +2314,10 @@ class Logging(LiteLLMLoggingBaseClass):
)
else:
if self.stream and complete_streaming_response:
self.model_call_details[
"complete_response"
] = self.model_call_details.get(
"complete_streaming_response", {}
self.model_call_details["complete_response"] = (
self.model_call_details.get(
"complete_streaming_response", {}
)
)
result = self.model_call_details["complete_response"]
@ -2458,9 +2456,9 @@ class Logging(LiteLLMLoggingBaseClass):
if complete_streaming_response is not None:
print_verbose("Async success callbacks: Got a complete streaming response")
self.model_call_details[
"async_complete_streaming_response"
] = complete_streaming_response
self.model_call_details["async_complete_streaming_response"] = (
complete_streaming_response
)
try:
if self.model_call_details.get("cache_hit", False) is True:
@ -2471,10 +2469,10 @@ class Logging(LiteLLMLoggingBaseClass):
model_call_details=self.model_call_details
)
# base_model defaults to None if not set on model_info
self.model_call_details[
"response_cost"
] = self._response_cost_calculator(
result=complete_streaming_response
self.model_call_details["response_cost"] = (
self._response_cost_calculator(
result=complete_streaming_response
)
)
verbose_logger.debug(
@ -2487,10 +2485,10 @@ class Logging(LiteLLMLoggingBaseClass):
self.model_call_details["response_cost"] = None
## STANDARDIZED LOGGING PAYLOAD
self.model_call_details[
"standard_logging_object"
] = self._build_standard_logging_payload(
complete_streaming_response, start_time, end_time
self.model_call_details["standard_logging_object"] = (
self._build_standard_logging_payload(
complete_streaming_response, start_time, end_time
)
)
# print standard logging payload
@ -2517,10 +2515,8 @@ class Logging(LiteLLMLoggingBaseClass):
# _success_handler_helper_fn
if self.model_call_details.get("standard_logging_object") is None:
## STANDARDIZED LOGGING PAYLOAD
self.model_call_details[
"standard_logging_object"
] = self._build_standard_logging_payload(
result, start_time, end_time
self.model_call_details["standard_logging_object"] = (
self._build_standard_logging_payload(result, start_time, end_time)
)
# print standard logging payload
@ -2764,18 +2760,18 @@ class Logging(LiteLLMLoggingBaseClass):
## STANDARDIZED LOGGING PAYLOAD
self.model_call_details[
"standard_logging_object"
] = get_standard_logging_object_payload(
kwargs=self.model_call_details,
init_response_obj={},
start_time=start_time,
end_time=end_time,
logging_obj=self,
status="failure",
error_str=str(exception),
original_exception=exception,
standard_built_in_tools_params=self.standard_built_in_tools_params,
self.model_call_details["standard_logging_object"] = (
get_standard_logging_object_payload(
kwargs=self.model_call_details,
init_response_obj={},
start_time=start_time,
end_time=end_time,
logging_obj=self,
status="failure",
error_str=str(exception),
original_exception=exception,
standard_built_in_tools_params=self.standard_built_in_tools_params,
)
)
return start_time, end_time
@ -3739,9 +3735,9 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915
service_name=arize_config.project_name,
)
os.environ[
"OTEL_EXPORTER_OTLP_TRACES_HEADERS"
] = f"space_id={arize_config.space_key or arize_config.space_id},api_key={arize_config.api_key}"
os.environ["OTEL_EXPORTER_OTLP_TRACES_HEADERS"] = (
f"space_id={arize_config.space_key or arize_config.space_id},api_key={arize_config.api_key}"
)
for callback in _in_memory_loggers:
if (
isinstance(callback, ArizeLogger)
@ -3767,13 +3763,13 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915
existing_attrs = os.environ.get("OTEL_RESOURCE_ATTRIBUTES", "")
# Add openinference.project.name attribute
if existing_attrs:
os.environ[
"OTEL_RESOURCE_ATTRIBUTES"
] = f"{existing_attrs},openinference.project.name={arize_phoenix_config.project_name}"
os.environ["OTEL_RESOURCE_ATTRIBUTES"] = (
f"{existing_attrs},openinference.project.name={arize_phoenix_config.project_name}"
)
else:
os.environ[
"OTEL_RESOURCE_ATTRIBUTES"
] = f"openinference.project.name={arize_phoenix_config.project_name}"
os.environ["OTEL_RESOURCE_ATTRIBUTES"] = (
f"openinference.project.name={arize_phoenix_config.project_name}"
)
# Set Phoenix project name from environment variable
phoenix_project_name = os.environ.get("PHOENIX_PROJECT_NAME", None)
@ -3781,19 +3777,19 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915
existing_attrs = os.environ.get("OTEL_RESOURCE_ATTRIBUTES", "")
# Add openinference.project.name attribute
if existing_attrs:
os.environ[
"OTEL_RESOURCE_ATTRIBUTES"
] = f"{existing_attrs},openinference.project.name={phoenix_project_name}"
os.environ["OTEL_RESOURCE_ATTRIBUTES"] = (
f"{existing_attrs},openinference.project.name={phoenix_project_name}"
)
else:
os.environ[
"OTEL_RESOURCE_ATTRIBUTES"
] = f"openinference.project.name={phoenix_project_name}"
os.environ["OTEL_RESOURCE_ATTRIBUTES"] = (
f"openinference.project.name={phoenix_project_name}"
)
# auth can be disabled on local deployments of arize phoenix
if arize_phoenix_config.otlp_auth_headers is not None:
os.environ[
"OTEL_EXPORTER_OTLP_TRACES_HEADERS"
] = arize_phoenix_config.otlp_auth_headers
os.environ["OTEL_EXPORTER_OTLP_TRACES_HEADERS"] = (
arize_phoenix_config.otlp_auth_headers
)
for callback in _in_memory_loggers:
if (
@ -3969,9 +3965,9 @@ def _init_custom_logger_compatible_class( # noqa: PLR0915
exporter="otlp_http",
endpoint="https://langtrace.ai/api/trace",
)
os.environ[
"OTEL_EXPORTER_OTLP_TRACES_HEADERS"
] = f"api_key={os.getenv('LANGTRACE_API_KEY')}"
os.environ["OTEL_EXPORTER_OTLP_TRACES_HEADERS"] = (
f"api_key={os.getenv('LANGTRACE_API_KEY')}"
)
for callback in _in_memory_loggers:
if (
isinstance(callback, OpenTelemetry)
@ -4204,8 +4200,7 @@ def _maybe_auto_initialize_arize_phoenix(_in_memory_loggers: list) -> None:
litellm.logging_callback_manager.add_litellm_callback(phoenix_logger)
verbose_logger.info(
"Auto-initialized Arize Phoenix logger alongside otel "
"(endpoint=%s)",
"Auto-initialized Arize Phoenix logger alongside otel " "(endpoint=%s)",
arize_phoenix_config.endpoint,
)
except Exception as e:
@ -4768,9 +4763,11 @@ class StandardLoggingPayloadSetup:
).model_dump()
if isinstance(_raw, dict):
if ResponseAPILoggingUtils._is_response_api_usage(_raw):
return ResponseAPILoggingUtils._transform_response_api_usage_to_chat_usage(
_raw
).model_dump()
return (
ResponseAPILoggingUtils._transform_response_api_usage_to_chat_usage(
_raw
).model_dump()
)
return _raw
if isinstance(_raw, Usage):
return _raw.model_dump()
@ -4884,10 +4881,10 @@ class StandardLoggingPayloadSetup:
for key in StandardLoggingHiddenParams.__annotations__.keys():
if key in hidden_params:
if key == "additional_headers":
clean_hidden_params[
"additional_headers"
] = StandardLoggingPayloadSetup.get_additional_headers(
hidden_params[key]
clean_hidden_params["additional_headers"] = (
StandardLoggingPayloadSetup.get_additional_headers(
hidden_params[key]
)
)
else:
clean_hidden_params[key] = hidden_params[key] # type: ignore
@ -5039,14 +5036,22 @@ class StandardLoggingPayloadSetup:
dynamic_litellm_session_id = litellm_params.get("litellm_session_id")
dynamic_litellm_trace_id = litellm_params.get("litellm_trace_id")
# Note: we recommend using `litellm_session_id` for session tracking
# `litellm_trace_id` is an internal litellm param
if dynamic_litellm_session_id:
return str(dynamic_litellm_session_id)
elif dynamic_litellm_trace_id:
return str(dynamic_litellm_trace_id)
else:
return logging_obj.litellm_trace_id
# Fallback: use metadata.session_id or metadata.trace_id for call chaining
metadata = litellm_params.get("metadata") or {}
metadata_session_id = metadata.get("session_id")
metadata_trace_id = metadata.get("trace_id")
if metadata_session_id:
return str(metadata_session_id)
if metadata_trace_id:
return str(metadata_trace_id)
return logging_obj.litellm_trace_id
@staticmethod
def _get_user_agent_tags(proxy_server_request: dict) -> Optional[List[str]]:
@ -5502,9 +5507,9 @@ def scrub_sensitive_keys_in_metadata(litellm_params: Optional[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"
cleaned_user_api_key_metadata[k] = (
"scrubbed_by_litellm_for_sensitive_keys"
)
else:
cleaned_user_api_key_metadata[k] = v
@ -5616,4 +5621,3 @@ def create_dummy_standard_logging_payload() -> StandardLoggingPayload:
model_parameters={"stream": True},
hidden_params=hidden_params,
)

10
litellm/llms/anthropic/chat/guardrail_translation/handler.py
View file

@ -75,7 +75,7 @@ class AnthropicMessagesHandler(BaseTranslation):
if messages is None:
return data
chat_completion_compatible_request, tool_name_mapping = (
chat_completion_compatible_request, _tool_name_mapping = (
LiteLLMAnthropicMessagesAdapter().translate_anthropic_to_openai(
# Use a shallow copy to avoid mutating request data (pop on litellm_metadata).
anthropic_message_request=cast(AnthropicMessagesRequest, data.copy())
@ -141,6 +141,14 @@ class AnthropicMessagesHandler(BaseTranslation):
return data
def extract_request_tool_names(self, data: dict) -> List[str]:
"""Extract tool names from Anthropic messages request (tools[].name)."""
names: List[str] = []
for tool in data.get("tools") or []:
if isinstance(tool, dict) and tool.get("name"):
names.append(str(tool["name"]))
return names
def _extract_input_text_and_images(
self,
message: Dict[str, Any],

136
litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py
View file

@ -41,7 +41,6 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper):
type="text",
text="",
)
pending_new_content_block: bool = False
chunk_queue: deque = deque() # Queue for buffering multiple chunks
def __init__(
@ -80,38 +79,40 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper):
from .transformation import LiteLLMAnthropicMessagesAdapter
try:
# Always return queued chunks first
if self.chunk_queue:
return self.chunk_queue.popleft()
# Queue initial chunks if not sent yet
if self.sent_first_chunk is False:
self.sent_first_chunk = True
return {
"type": "message_start",
"message": {
"id": "msg_{}".format(uuid.uuid4()),
"type": "message",
"role": "assistant",
"content": [],
"model": self.model,
"stop_reason": None,
"stop_sequence": None,
"usage": self._create_initial_usage_delta(),
},
}
self.chunk_queue.append(
{
"type": "message_start",
"message": {
"id": "msg_{}".format(uuid.uuid4()),
"type": "message",
"role": "assistant",
"content": [],
"model": self.model,
"stop_reason": None,
"stop_sequence": None,
"usage": self._create_initial_usage_delta(),
},
}
)
return self.chunk_queue.popleft()
if self.sent_content_block_start is False:
self.sent_content_block_start = True
return {
"type": "content_block_start",
"index": self.current_content_block_index,
"content_block": {"type": "text", "text": ""},
}
# Handle pending new content block start
if self.pending_new_content_block:
self.pending_new_content_block = False
self.sent_content_block_finish = False # Reset for new block
return {
"type": "content_block_start",
"index": self.current_content_block_index,
"content_block": self.current_content_block_start,
}
self.chunk_queue.append(
{
"type": "content_block_start",
"index": self.current_content_block_index,
"content_block": {"type": "text", "text": ""},
}
)
return self.chunk_queue.popleft()
for chunk in self.completion_stream:
if chunk == "None" or chunk is None:
@ -126,45 +127,65 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper):
current_content_block_index=self.current_content_block_index,
)
# Check if we need to start a new content block
# This is where you'd add your logic to detect when a new content block should start
# For example, if the chunk indicates a tool call or different content type
if should_start_new_block and not self.sent_content_block_finish:
# End current content block and prepare for new one
self.holding_chunk = processed_chunk
self.sent_content_block_finish = True
self.pending_new_content_block = True
return {
"type": "content_block_stop",
"index": max(self.current_content_block_index - 1, 0),
}
# Queue the sequence: content_block_stop -> content_block_start
# The trigger chunk itself is not emitted as a delta since the
# content_block_start already carries the relevant information.
self.chunk_queue.append(
{
"type": "content_block_stop",
"index": max(self.current_content_block_index - 1, 0),
}
)
self.chunk_queue.append(
{
"type": "content_block_start",
"index": self.current_content_block_index,
"content_block": self.current_content_block_start,
}
)
self.sent_content_block_finish = False
return self.chunk_queue.popleft()
if (
processed_chunk["type"] == "message_delta"
and self.sent_content_block_finish is False
):
self.holding_chunk = processed_chunk
# Queue both the content_block_stop and the message_delta
self.chunk_queue.append(
{
"type": "content_block_stop",
"index": self.current_content_block_index,
}
)
self.sent_content_block_finish = True
return {
"type": "content_block_stop",
"index": self.current_content_block_index,
}
self.chunk_queue.append(processed_chunk)
return self.chunk_queue.popleft()
elif self.holding_chunk is not None:
return_chunk = self.holding_chunk
self.holding_chunk = processed_chunk
return return_chunk
self.chunk_queue.append(self.holding_chunk)
self.chunk_queue.append(processed_chunk)
self.holding_chunk = None
return self.chunk_queue.popleft()
else:
return processed_chunk
self.chunk_queue.append(processed_chunk)
return self.chunk_queue.popleft()
# Handle any remaining held chunks after stream ends
if self.holding_chunk is not None:
return_chunk = self.holding_chunk
self.chunk_queue.append(self.holding_chunk)
self.holding_chunk = None
return return_chunk
if self.sent_last_message is False:
if not self.sent_last_message:
self.sent_last_message = True
return {"type": "message_stop"}
self.chunk_queue.append({"type": "message_stop"})
if self.chunk_queue:
return self.chunk_queue.popleft()
raise StopIteration
except StopIteration:
if self.chunk_queue:
return self.chunk_queue.popleft()
if self.sent_last_message is False:
self.sent_last_message = True
return {"type": "message_stop"}
@ -265,7 +286,9 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper):
if not self.queued_usage_chunk:
if should_start_new_block and not self.sent_content_block_finish:
# Queue the sequence: content_block_stop -> content_block_start -> current_chunk
# Queue the sequence: content_block_stop -> content_block_start
# The trigger chunk itself is not emitted as a delta since the
# content_block_start already carries the relevant information.
# 1. Stop current content block
self.chunk_queue.append(
@ -284,9 +307,6 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper):
}
)
# 3. Queue the current chunk (don't lose it!)
self.chunk_queue.append(processed_chunk)
# Reset state for new block
self.sent_content_block_finish = False

8
litellm/llms/azure/chat/gpt_5_transformation.py
View file

@ -43,8 +43,12 @@ class AzureOpenAIGPT5Config(AzureOpenAIConfig, OpenAIGPT5Config):
if "tool_choice" not in params:
params.append("tool_choice")
# Only gpt-5.2 has been verified to support logprobs on Azure
if self.is_model_gpt_5_2_model(model):
# Only gpt-5.2 has been verified to support logprobs on Azure.
# The base OpenAI class includes logprobs for gpt-5.1+, but Azure
# hasn't verified support for gpt-5.1, so remove them unless gpt-5.2.
if self.is_model_gpt_5_1_model(model) and not self.is_model_gpt_5_2_model(model):
params = [p for p in params if p not in ["logprobs", "top_logprobs"]]
elif self.is_model_gpt_5_2_model(model):
azure_supported_params = ["logprobs", "top_logprobs"]
params.extend(azure_supported_params)

38
litellm/llms/azure/common_utils.py
View file

@ -1,6 +1,6 @@
import json
import os
from typing import Any, Callable, Dict, Literal, Optional, Union, cast
from typing import Any, Callable, Dict, Literal, NamedTuple, Optional, Union, cast
import httpx
from openai import AsyncAzureOpenAI, AsyncOpenAI, AzureOpenAI, OpenAI
@ -789,3 +789,39 @@ class BaseAzureLLM(BaseOpenAILLM):
return param_value
return os.getenv(env_var_key)
class AzureCredentials(NamedTuple):
api_base: Optional[str]
api_key: Optional[str]
api_version: Optional[str]
def get_azure_credentials(
api_base: Optional[str] = None,
api_key: Optional[str] = None,
api_version: Optional[str] = None,
) -> AzureCredentials:
"""Resolve Azure credentials from params, litellm globals, and env vars."""
resolved_api_base = (
api_base
or litellm.api_base
or get_secret_str("AZURE_API_BASE")
)
resolved_api_version = (
api_version
or litellm.api_version
or get_secret_str("AZURE_API_VERSION")
)
resolved_api_key = (
api_key
or litellm.api_key
or litellm.azure_key
or get_secret_str("AZURE_OPENAI_API_KEY")
or get_secret_str("AZURE_API_KEY")
)
return AzureCredentials(
api_base=resolved_api_base,
api_key=resolved_api_key,
api_version=resolved_api_version,
)

7
litellm/llms/base_llm/guardrail_translation/base_translation.py
View file

@ -98,3 +98,10 @@ class BaseTranslation(ABC):
Optional to override in subclasses.
"""
return responses_so_far
def extract_request_tool_names(self, data: dict) -> List[str]:
"""
Extract tool names from the request body for allowlist/policy checks.
Override in tool-capable handlers; default returns [].
"""
return []

93
litellm/llms/custom_httpx/llm_http_handler.py
View file

@ -69,6 +69,7 @@ from litellm.responses.streaming_iterator import (
BaseResponsesAPIStreamingIterator,
MockResponsesAPIStreamingIterator,
ResponsesAPIStreamingIterator,
ResponsesWebSocketStreaming,
SyncResponsesAPIStreamingIterator,
)
from litellm.types.containers.main import (
@ -4731,6 +4732,98 @@ class BaseLLMHTTPHandler:
f"Unexpected error while closing WebSocket: {close_error}"
)
async def async_responses_websocket(
self,
model: str,
websocket: Any,
logging_obj: LiteLLMLoggingObj,
responses_api_provider_config: BaseResponsesAPIConfig,
api_base: Optional[str] = None,
api_key: Optional[str] = None,
timeout: Optional[float] = None,
user_api_key_dict: Optional[Any] = None,
litellm_metadata: Optional[Dict[str, Any]] = None,
):
"""
Handles Responses API WebSocket mode.
Opens a persistent WebSocket to the provider's /v1/responses endpoint
and proxies response.create events bidirectionally for lower-latency
agentic workflows.
"""
import websockets
from websockets.asyncio.client import ClientConnection
litellm_params = GenericLiteLLMParams()
headers = responses_api_provider_config.validate_environment(
headers={},
model=model,
litellm_params=litellm_params,
)
if api_key:
headers["Authorization"] = f"Bearer {api_key}"
http_url = responses_api_provider_config.get_complete_url(
api_base=api_base,
litellm_params={},
)
# /responses -> wss:// URL
ws_url = http_url.replace("https://", "wss://").replace("http://", "ws://")
try:
ssl_context = get_shared_realtime_ssl_context()
if ws_url.startswith("wss://") and ssl_context is False:
ssl_context = ssl.SSLContext(ssl.PROTOCOL_TLS_CLIENT)
ssl_context.check_hostname = False
ssl_context.verify_mode = ssl.CERT_NONE
logging_obj.pre_call(
input=None,
api_key=api_key or "",
additional_args={
"api_base": ws_url,
"headers": headers,
"complete_input_dict": {"mode": "responses_websocket"},
},
)
async with websockets.connect( # type: ignore
ws_url,
additional_headers=headers,
max_size=REALTIME_WEBSOCKET_MAX_MESSAGE_SIZE_BYTES,
ssl=ssl_context,
) as backend_ws:
_request_data: Dict[str, Any] = {}
if litellm_metadata:
_request_data["litellm_metadata"] = litellm_metadata
streaming = ResponsesWebSocketStreaming(
websocket=websocket,
backend_ws=cast(ClientConnection, backend_ws),
logging_obj=logging_obj,
user_api_key_dict=user_api_key_dict,
request_data=_request_data,
)
await streaming.bidirectional_forward()
except websockets.exceptions.InvalidStatusCode as e: # type: ignore
verbose_logger.exception(f"Error connecting to responses WS backend: {e}")
await websocket.close(code=e.status_code, reason=str(e))
except Exception as e:
verbose_logger.exception(f"Error in responses WS: {e}")
try:
await websocket.close(
code=1011, reason=f"Internal server error: {str(e)}"
)
except RuntimeError as close_error:
if "already completed" in str(close_error) or "websocket.close" in str(
close_error
):
pass
else:
raise Exception(
f"Unexpected error while closing WebSocket: {close_error}"
)
def image_edit_handler(
self,
model: str,

3
litellm/llms/gemini/common_utils.py
View file

@ -166,7 +166,8 @@ class GoogleAIStudioTokenCounter(BaseTokenCounter):
contents: Optional[List[Dict[str, Any]]],
deployment: Optional[Dict[str, Any]] = None,
request_model: str = "",
**kwargs,
tools: Optional[List[Dict[str, Any]]] = None,
system: Optional[Any] = None,
) -> Optional[TokenCountResponse]:
import copy

13
litellm/llms/openai/chat/guardrail_translation/handler.py
View file

@ -135,6 +135,19 @@ class OpenAIChatCompletionsHandler(BaseTranslation):
return data
def extract_request_tool_names(self, data: dict) -> List[str]:
"""Extract tool names from OpenAI chat completions request (tools[].function.name, functions[].name)."""
names: List[str] = []
for tool in data.get("tools") or []:
if isinstance(tool, dict) and tool.get("type") == "function":
fn = tool.get("function")
if isinstance(fn, dict) and fn.get("name"):
names.append(str(fn["name"]))
for fn in data.get("functions") or []:
if isinstance(fn, dict) and fn.get("name"):
names.append(str(fn["name"]))
return names
def _extract_inputs(
self,
message: Dict[str, Any],

39
litellm/llms/openai/common_utils.py
View file

@ -5,8 +5,9 @@ Common helpers / utils across al OpenAI endpoints
import hashlib
import inspect
import json
import os
import ssl
from typing import TYPE_CHECKING, Any, Dict, List, Literal, Optional, Tuple, Union
from typing import TYPE_CHECKING, Any, Dict, List, Literal, NamedTuple, Optional, Tuple, Union
import httpx
import openai
@ -244,3 +245,39 @@ class BaseOpenAILLM:
)
class OpenAICredentials(NamedTuple):
api_base: str
api_key: Optional[str]
organization: Optional[str]
def get_openai_credentials(
api_base: Optional[str] = None,
api_key: Optional[str] = None,
organization: Optional[str] = None,
) -> OpenAICredentials:
"""Resolve OpenAI credentials from params, litellm globals, and env vars."""
resolved_api_base = (
api_base
or litellm.api_base
or os.getenv("OPENAI_BASE_URL")
or os.getenv("OPENAI_API_BASE")
or "https://api.openai.com/v1"
)
resolved_organization = (
organization
or litellm.organization
or os.getenv("OPENAI_ORGANIZATION", None)
or None
)
resolved_api_key = (
api_key
or litellm.api_key
or litellm.openai_key
or os.getenv("OPENAI_API_KEY")
)
return OpenAICredentials(
api_base=resolved_api_base,
api_key=resolved_api_key,
organization=resolved_organization,
)

39
litellm/llms/openai/responses/guardrail_translation/handler.py
View file

@ -30,27 +30,22 @@ Output: response.output is List[GenericResponseOutputItem] where each has:
from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Union, cast
from openai.types.responses.response_function_tool_call import ResponseFunctionToolCall
from openai.types.responses.response_function_tool_call import \
ResponseFunctionToolCall
from pydantic import BaseModel
from litellm._logging import verbose_proxy_logger
from litellm.completion_extras.litellm_responses_transformation.transformation import (
LiteLLMResponsesTransformationHandler,
OpenAiResponsesToChatCompletionStreamIterator,
)
from litellm.llms.base_llm.guardrail_translation.base_translation import BaseTranslation
from litellm.responses.litellm_completion_transformation.transformation import (
LiteLLMCompletionResponsesConfig,
)
from litellm.types.llms.openai import (
ChatCompletionToolCallChunk,
ChatCompletionToolParam,
)
from litellm.types.responses.main import (
GenericResponseOutputItem,
OutputFunctionToolCall,
OutputText,
)
OpenAiResponsesToChatCompletionStreamIterator)
from litellm.llms.base_llm.guardrail_translation.base_translation import \
BaseTranslation
from litellm.responses.litellm_completion_transformation.transformation import \
LiteLLMCompletionResponsesConfig
from litellm.types.llms.openai import (ChatCompletionToolCallChunk,
ChatCompletionToolParam)
from litellm.types.responses.main import (GenericResponseOutputItem,
OutputFunctionToolCall, OutputText)
from litellm.types.utils import GenericGuardrailAPIInputs
if TYPE_CHECKING:
@ -188,6 +183,18 @@ class OpenAIResponsesHandler(BaseTranslation):
return data
def extract_request_tool_names(self, data: dict) -> List[str]:
"""Extract tool names from Responses API request (tools[].name for function, tools[].server_label for mcp)."""
names: List[str] = []
for tool in data.get("tools") or []:
if not isinstance(tool, dict):
continue
if tool.get("type") == "function" and tool.get("name"):
names.append(str(tool["name"]))
elif tool.get("type") == "mcp" and tool.get("server_label"):
names.append(str(tool["server_label"]))
return names
def _extract_and_transform_tools(
self,
tools: List[Dict[str, Any]],

11
litellm/llms/openrouter/image_edit/__init__.py Normal file
View file

@ -0,0 +1,11 @@
from litellm.llms.base_llm.image_edit.transformation import BaseImageEditConfig
from .transformation import OpenRouterImageEditConfig
__all__ = [
"OpenRouterImageEditConfig",
]
def get_openrouter_image_edit_config(model: str) -> BaseImageEditConfig:
return OpenRouterImageEditConfig()

367
litellm/llms/openrouter/image_edit/transformation.py Normal file
View file

@ -0,0 +1,367 @@
"""
OpenRouter Image Edit Support
OpenRouter provides image editing through chat completion endpoints.
The source image is sent as a base64 data URL in the message content,
and the response contains edited images in the message's images array.
Request format:
{
"model": "google/gemini-2.5-flash-image",
"messages": [{
"role": "user",
"content": [
{"type": "image_url", "image_url": {"url": "data:image/png;base64,..."}},
{"type": "text", "text": "Edit this image by..."}
]
}],
"modalities": ["image", "text"]
}
Response format:
{
"choices": [{
"message": {
"content": "Here is the edited image.",
"role": "assistant",
"images": [{
"image_url": {"url": "data:image/png;base64,..."},
"type": "image_url"
}]
}
}],
"usage": {
"completion_tokens": 1299,
"prompt_tokens": 300,
"total_tokens": 1599,
"completion_tokens_details": {"image_tokens": 1290},
"cost": 0.0387243
}
}
"""
import base64
from io import BufferedReader, BytesIO
from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Union, cast
import httpx
from httpx._types import RequestFiles
import litellm
from litellm.images.utils import ImageEditRequestUtils
from litellm.llms.base_llm.chat.transformation import BaseLLMException
from litellm.llms.base_llm.image_edit.transformation import BaseImageEditConfig
from litellm.llms.openrouter.common_utils import OpenRouterException
from litellm.secret_managers.main import get_secret_str
from litellm.types.images.main import ImageEditOptionalRequestParams
from litellm.types.router import GenericLiteLLMParams
from litellm.types.utils import FileTypes, ImageObject, ImageResponse, ImageUsage, ImageUsageInputTokensDetails
if TYPE_CHECKING:
from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj
LiteLLMLoggingObj = _LiteLLMLoggingObj
else:
LiteLLMLoggingObj = Any
class OpenRouterImageEditConfig(BaseImageEditConfig):
"""
Configuration for OpenRouter image editing via chat completions.
OpenRouter uses the chat completions endpoint for image editing.
The source image is sent as a base64 data URL in the message content,
and the response contains edited images in the message's images array.
"""
def get_supported_openai_params(self, model: str) -> list:
return ["size", "quality", "n"]
def map_openai_params(
self,
image_edit_optional_params: ImageEditOptionalRequestParams,
model: str,
drop_params: bool,
) -> Dict:
supported_params = self.get_supported_openai_params(model)
mapped_params: Dict[str, Any] = {}
for key, value in image_edit_optional_params.items():
if key in supported_params:
if key == "size":
if "image_config" not in mapped_params:
mapped_params["image_config"] = {}
mapped_params["image_config"]["aspect_ratio"] = self._map_size_to_aspect_ratio(value)
elif key == "quality":
image_size = self._map_quality_to_image_size(value)
if image_size:
if "image_config" not in mapped_params:
mapped_params["image_config"] = {}
mapped_params["image_config"]["image_size"] = image_size
else:
mapped_params[key] = value
return mapped_params
def validate_environment(
self,
headers: dict,
model: str,
api_key: Optional[str] = None,
) -> dict:
api_key = (
api_key
or litellm.api_key
or get_secret_str("OPENROUTER_API_KEY")
)
if not api_key:
raise ValueError("OPENROUTER_API_KEY is not set")
headers.update(
{
"Authorization": f"Bearer {api_key}",
}
)
return headers
def use_multipart_form_data(self) -> bool:
"""OpenRouter uses JSON requests, not multipart/form-data."""
return False
def get_complete_url(
self,
model: str,
api_base: Optional[str],
litellm_params: dict,
) -> str:
base_url = api_base or get_secret_str("OPENROUTER_API_BASE") or "https://openrouter.ai/api/v1"
base_url = base_url.rstrip("/")
if not base_url.endswith("/chat/completions"):
return f"{base_url}/chat/completions"
return base_url
def transform_image_edit_request(
self,
model: str,
prompt: Optional[str],
image: Optional[FileTypes],
image_edit_optional_request_params: Dict,
litellm_params: GenericLiteLLMParams,
headers: dict,
) -> Tuple[Dict, RequestFiles]:
content_parts: List[Dict[str, Any]] = []
# Add source image(s) as base64 data URLs
if image is not None:
images = image if isinstance(image, list) else [image]
for img in images:
if img is None:
continue
mime_type = ImageEditRequestUtils.get_image_content_type(img)
image_bytes = self._read_image_bytes(img)
b64_data = base64.b64encode(image_bytes).decode("utf-8")
content_parts.append(
{
"type": "image_url",
"image_url": {
"url": f"data:{mime_type};base64,{b64_data}"
},
}
)
# Add the text prompt
if prompt:
content_parts.append({"type": "text", "text": prompt})
request_body: Dict[str, Any] = {
"model": model,
"messages": [
{
"role": "user",
"content": content_parts,
}
],
"modalities": ["image", "text"],
}
# Add mapped optional params (image_config, n, etc.)
for key, value in image_edit_optional_request_params.items():
if key not in ("model", "messages", "modalities"):
request_body[key] = value
empty_files = cast(RequestFiles, [])
return request_body, empty_files
def transform_image_edit_response(
self,
model: str,
raw_response: httpx.Response,
logging_obj: LiteLLMLoggingObj,
) -> ImageResponse:
try:
response_json = raw_response.json()
except Exception as e:
raise OpenRouterException(
message=f"Error parsing OpenRouter response: {str(e)}",
status_code=raw_response.status_code,
headers=raw_response.headers,
)
model_response = ImageResponse()
model_response.data = []
try:
choices = response_json.get("choices", [])
for choice in choices:
message = choice.get("message", {})
images = message.get("images", [])
for image_data in images:
image_url_obj = image_data.get("image_url", {})
image_url = image_url_obj.get("url")
if image_url:
if image_url.startswith("data:"):
# Extract base64 data from data URL
parts = image_url.split(",", 1)
b64_data = parts[1] if len(parts) > 1 else None
model_response.data.append(
ImageObject(
b64_json=b64_data,
url=None,
revised_prompt=None,
)
)
else:
model_response.data.append(
ImageObject(
b64_json=None,
url=image_url,
revised_prompt=None,
)
)
except Exception as e:
raise OpenRouterException(
message=f"Error transforming OpenRouter image edit response: {str(e)}",
status_code=500,
headers={},
)
self._set_usage_and_cost(model_response, response_json, model)
return model_response
def get_error_class(
self, error_message: str, status_code: int, headers: Union[dict, httpx.Headers]
) -> BaseLLMException:
return OpenRouterException(
message=error_message,
status_code=status_code,
headers=headers,
)
# Private helper methods
def _map_size_to_aspect_ratio(self, size: str) -> str:
"""
Map OpenAI size format to OpenRouter aspect_ratio format.
Uses the same mapping as image generation since OpenRouter
handles both through the same chat completions endpoint.
"""
size_to_aspect_ratio = {
"256x256": "1:1",
"512x512": "1:1",
"1024x1024": "1:1",
"1536x1024": "3:2",
"1792x1024": "16:9",
"1024x1536": "2:3",
"1024x1792": "9:16",
"auto": "1:1",
}
return size_to_aspect_ratio.get(size, "1:1")
def _map_quality_to_image_size(self, quality: str) -> Optional[str]:
"""
Map OpenAI quality to OpenRouter image_size format.
Uses the same mapping as image generation since OpenRouter
handles both through the same chat completions endpoint.
"""
quality_to_image_size = {
"low": "1K",
"standard": "1K",
"medium": "2K",
"high": "4K",
"hd": "4K",
"auto": "1K",
}
return quality_to_image_size.get(quality)
def _set_usage_and_cost(
self,
model_response: ImageResponse,
response_json: dict,
model: str,
) -> None:
"""Extract and set usage and cost information from OpenRouter response."""
usage_data = response_json.get("usage", {})
if usage_data:
prompt_tokens = usage_data.get("prompt_tokens", 0)
total_tokens = usage_data.get("total_tokens", 0)
completion_tokens_details = usage_data.get("completion_tokens_details", {})
image_tokens = completion_tokens_details.get("image_tokens", 0)
# For image edit, input may include image tokens
input_image_tokens = 0
prompt_tokens_details = usage_data.get("prompt_tokens_details", {})
if prompt_tokens_details:
input_image_tokens = prompt_tokens_details.get("image_tokens", 0)
model_response.usage = ImageUsage(
input_tokens=prompt_tokens,
input_tokens_details=ImageUsageInputTokensDetails(
image_tokens=input_image_tokens,
text_tokens=prompt_tokens - input_image_tokens,
),
output_tokens=image_tokens,
total_tokens=total_tokens,
)
cost = usage_data.get("cost")
if cost is not None:
if not hasattr(model_response, "_hidden_params"):
model_response._hidden_params = {}
if "additional_headers" not in model_response._hidden_params:
model_response._hidden_params["additional_headers"] = {}
model_response._hidden_params["additional_headers"][
"llm_provider-x-litellm-response-cost"
] = float(cost)
cost_details = usage_data.get("cost_details", {})
if cost_details:
if "response_cost_details" not in model_response._hidden_params:
model_response._hidden_params["response_cost_details"] = {}
model_response._hidden_params["response_cost_details"].update(cost_details)
model_response._hidden_params["model"] = response_json.get("model", model)
def _read_image_bytes(self, image: FileTypes) -> bytes:
"""Read raw bytes from various image input types."""
if isinstance(image, bytes):
return image
if isinstance(image, BytesIO):
current_pos = image.tell()
image.seek(0)
data = image.read()
image.seek(current_pos)
return data
if isinstance(image, BufferedReader):
current_pos = image.tell()
image.seek(0)
data = image.read()
image.seek(current_pos)
return data
raise ValueError("Unsupported image type for OpenRouter image edit.")

1
litellm/llms/searchapi/__init__.py Normal file
View file

@ -0,0 +1 @@
"""SearchAPI.io integration for LiteLLM."""

4
litellm/llms/searchapi/search/__init__.py Normal file
View file

@ -0,0 +1,4 @@
"""SearchAPI.io search integration for LiteLLM."""
from litellm.llms.searchapi.search.transformation import SearchAPIConfig
__all__ = ["SearchAPIConfig"]

232
litellm/llms/searchapi/search/transformation.py Normal file
View file

@ -0,0 +1,232 @@
"""
Calls SearchAPI.io's Google Search API endpoint.
SearchAPI.io API Reference: https://www.searchapi.io/docs/google
"""
from typing import Dict, List, Literal, Optional, TypedDict, Union
from urllib.parse import urlencode
import httpx
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
from litellm.llms.base_llm.search.transformation import (
BaseSearchConfig,
SearchResponse,
SearchResult,
)
from litellm.secret_managers.main import get_secret_str
class _SearchAPIRequestRequired(TypedDict):
"""Required fields for SearchAPI.io request."""
engine: str # Required - search engine (e.g., 'google')
q: str # Required - search query
class SearchAPIRequest(_SearchAPIRequestRequired, total=False):
"""
SearchAPI.io request format for Google Search.
Based on: https://www.searchapi.io/docs/google
"""
kgmid: str # Optional - Knowledge Graph identifier
device: str # Optional - device type ('desktop', 'mobile', 'tablet')
location: str # Optional - geographic location
uule: str # Optional - Google-encoded location
google_domain: str # Optional - Google domain (deprecated)
gl: str # Optional - country code (e.g., 'us', 'uk')
hl: str # Optional - interface language (e.g., 'en', 'es')
lr: str # Optional - language restriction (e.g., 'lang_en')
cr: str # Optional - country restriction
nfpr: int # Optional - exclude auto-corrected results (0 or 1)
filter: int # Optional - duplicate/host crowding filter (0 or 1)
safe: str # Optional - SafeSearch ('active', 'off')
time_period: str # Optional - time period ('last_hour', 'last_day', 'last_week', 'last_month', 'last_year')
time_period_min: str # Optional - start date (MM/DD/YYYY)
time_period_max: str # Optional - end date (MM/DD/YYYY)
num: int # Optional - number of results (phased out by Google, constant 10)
page: int # Optional - page number for pagination
optimization_strategy: str # Optional - 'performance' or 'ads'
class SearchAPIConfig(BaseSearchConfig):
SEARCHAPI_API_BASE = "https://www.searchapi.io/api/v1/search"
@staticmethod
def ui_friendly_name() -> str:
return "SearchAPI.io (Google Search)"
def get_http_method(self) -> Literal["GET", "POST"]:
"""
SearchAPI.io uses GET requests for search.
"""
return "GET"
def validate_environment(
self,
headers: Dict,
api_key: Optional[str] = None,
api_base: Optional[str] = None,
**kwargs,
) -> Dict:
"""
Validate environment and return headers.
"""
api_key = api_key or get_secret_str("SEARCHAPI_API_KEY")
if not api_key:
raise ValueError(
"SEARCHAPI_API_KEY is not set. Set `SEARCHAPI_API_KEY` environment variable."
)
headers["Content-Type"] = "application/json"
return headers
def get_complete_url(
self,
api_base: Optional[str],
optional_params: dict,
data: Optional[Union[Dict, List[Dict]]] = None,
**kwargs,
) -> str:
"""
Get complete URL for Search endpoint with query parameters.
SearchAPI.io uses GET requests and includes api_key in query params.
"""
api_base = api_base or get_secret_str("SEARCHAPI_API_BASE") or self.SEARCHAPI_API_BASE
# Build query parameters from the transformed request body
if data and isinstance(data, dict) and "_searchapi_params" in data:
params = data["_searchapi_params"]
query_string = urlencode(params, doseq=True)
return f"{api_base}?{query_string}"
return api_base
def transform_search_request(
self,
query: Union[str, List[str]],
optional_params: dict,
api_key: Optional[str] = None,
search_engine_id: Optional[str] = None,
**kwargs,
) -> Dict:
"""
Transform Search request to SearchAPI.io format.
Transforms unified spec parameters:
- query → q
- max_results → num (limited to 10 by Google)
- search_domain_filter → q (append site: filters)
- country → gl
Args:
query: Search query (string or list of strings)
optional_params: Optional parameters for the request
api_key: API key for authentication
Returns:
Dict with typed request data following SearchAPI.io spec
"""
if isinstance(query, list):
query = " ".join(query)
# Get API key from parameter or environment
api_key = api_key or get_secret_str("SEARCHAPI_API_KEY")
if not api_key:
raise ValueError(
"SEARCHAPI_API_KEY is not set. Set `SEARCHAPI_API_KEY` environment variable."
)
request_data: SearchAPIRequest = {
"engine": "google",
"q": query,
}
# Add API key to request
result_data = dict(request_data)
result_data["api_key"] = api_key
# Transform unified spec parameters to SearchAPI.io format
if "max_results" in optional_params:
# Google now returns constant 10 results, but we can still set num
num_results = min(optional_params["max_results"], 10)
result_data["num"] = num_results
if "search_domain_filter" in optional_params:
# Convert to multiple "site:domain" clauses
domains = optional_params["search_domain_filter"]
if isinstance(domains, list) and len(domains) > 0:
result_data["q"] = self._append_domain_filters(
result_data["q"], domains
)
if "country" in optional_params:
# Map to gl parameter
result_data["gl"] = optional_params["country"].lower()
# Pass through all other SearchAPI.io-specific parameters
for param, value in optional_params.items():
if (
param not in self.get_supported_perplexity_optional_params()
and param not in result_data
):
result_data[param] = value
# Store params in special key for URL building (GET request)
return {
"_searchapi_params": result_data,
}
@staticmethod
def _append_domain_filters(query: str, domains: List[str]) -> str:
"""
Add site: filters to restrict search to specific domains.
"""
domain_clauses = [f"site:{domain}" for domain in domains]
domain_query = " OR ".join(domain_clauses)
return f"({query}) AND ({domain_query})"
def transform_search_response(
self,
raw_response: httpx.Response,
logging_obj: Optional[LiteLLMLoggingObj],
**kwargs,
) -> SearchResponse:
"""
Transform SearchAPI.io response to LiteLLM unified SearchResponse format.
SearchAPI.io → LiteLLM mappings:
- organic_results[].title → SearchResult.title
- organic_results[].link → SearchResult.url
- organic_results[].snippet → SearchResult.snippet
- organic_results[].date → SearchResult.date
"""
response_json = raw_response.json()
# Transform results to SearchResult objects
results: List[SearchResult] = []
# Process organic results
for result in response_json.get("organic_results", []):
title = result.get("title", "")
url = result.get("link", "")
snippet = result.get("snippet", "")
date = result.get("date") # SearchAPI.io provides date in some results
search_result = SearchResult(
title=title,
url=url,
snippet=snippet,
date=date,
last_updated=None, # SearchAPI.io doesn't provide last_updated
)
results.append(search_result)
return SearchResponse(
results=results,
object="search",
)

18
litellm/llms/vertex_ai/batches/handler.py
View file

@ -108,11 +108,19 @@ class VertexAIBatchPrediction(VertexLLM):
client = get_async_httpx_client(
llm_provider=litellm.LlmProviders.VERTEX_AI,
)
response = await client.post(
url=api_base,
headers=headers,
data=json.dumps(vertex_batch_request),
)
try:
response = await client.post(
url=api_base,
headers=headers,
data=json.dumps(vertex_batch_request),
)
except httpx.HTTPStatusError as e:
error_body = e.response.text
litellm.verbose_logger.error(
"Vertex AI batch create failed: status=%s, body=%s",
e.response.status_code, error_body[:1000],
)
raise
if response.status_code != 200:
raise Exception(f"Error: {response.status_code} {response.text}")

2
litellm/llms/vertex_ai/batches/transformation.py
View file

@ -29,7 +29,7 @@ class VertexAIBatchTransformation:
if input_file_id is None:
raise ValueError("input_file_id is required, but not provided")
input_config: InputConfig = InputConfig(
gcsSource=GcsSource(uris=input_file_id), instancesFormat="jsonl"
gcsSource=GcsSource(uris=[input_file_id]), instancesFormat="jsonl"
)
model: str = cls._get_model_from_gcs_file(input_file_id)
output_config: OutputConfig = OutputConfig(

39
litellm/llms/vertex_ai/common_utils.py
View file

@ -571,14 +571,38 @@ def _filter_anyof_fields(schema_dict: Dict[str, Any]) -> Dict[str, Any]:
return schema_dict
def _is_any_type_schema(schema: dict) -> bool:
"""
Detect schemas that represent "any JSON value" (no type constraints).
In JSON Schema, an empty schema {} means "any value is valid".
Schemas with only metadata keys (title, description, default, examples)
but no type-constraining keywords also represent "any type".
Gemini's Schema proto uses TYPE_UNSPECIFIED (0) as default,
so omitting the type field is valid and means "any type".
"""
type_constraining_keys = {
"type",
"properties",
"items",
"anyOf",
"oneOf",
"allOf",
"enum",
"required",
"$ref",
"$schema",
}
return not any(key in type_constraining_keys for key in schema.keys())
def process_items(schema, depth=0):
if depth > DEFAULT_MAX_RECURSE_DEPTH:
raise ValueError(
f"Max depth of {DEFAULT_MAX_RECURSE_DEPTH} exceeded while processing schema. Please check the schema for excessive nesting."
)
if isinstance(schema, dict):
if "items" in schema and schema["items"] == {}:
schema["items"] = {"type": "object"}
for key, value in schema.items():
if isinstance(value, dict):
process_items(value, depth + 1)
@ -677,9 +701,8 @@ def convert_anyof_null_to_nullable(schema, depth=0):
# remove null type
anyof.remove(atype)
contains_null = True
elif "type" not in atype and len(atype) == 0:
# Handle empty object case
atype["type"] = "object"
elif isinstance(atype, dict) and _is_any_type_schema(atype):
pass # preserve "any type" semantics — don't coerce to object
if len(anyof) == 0:
# Edge case: response schema with only null type present is invalid in Vertex AI
@ -714,7 +737,8 @@ def add_object_type(schema):
# Gemini requires all function parameters to be type OBJECT
# Handle case where schema has no properties and no type (e.g. tools with no arguments)
if "type" not in schema and "anyOf" not in schema and "oneOf" not in schema and "allOf" not in schema:
schema["type"] = "object"
if not _is_any_type_schema(schema):
schema["type"] = "object"
properties = schema.get("properties", None)
if properties is not None:
@ -1030,7 +1054,8 @@ class VertexAITokenCounter(BaseTokenCounter):
contents: Optional[List[Dict[str, Any]]],
deployment: Optional[Dict[str, Any]] = None,
request_model: str = "",
**kwargs,
tools: Optional[List[Dict[str, Any]]] = None,
system: Optional[Any] = None,
) -> Optional[TokenCountResponse]:
import copy

63
litellm/llms/vertex_ai/files/transformation.py
View file

@ -335,13 +335,37 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig):
status_code=status_code, message=error_message, headers=headers
)
def _parse_gcs_uri(self, file_id: str) -> Tuple[str, str]:
"""
Parse a GCS URI (gs://bucket/path/to/object) into (bucket, url-encoded-object-path).
Handles both raw and URL-encoded input.
"""
import urllib.parse
decoded = urllib.parse.unquote(file_id)
if decoded.startswith("gs://"):
full_path = decoded[5:]
else:
full_path = decoded
if "/" in full_path:
bucket_name, object_path = full_path.split("/", 1)
else:
bucket_name = full_path
object_path = ""
encoded_object = urllib.parse.quote(object_path, safe="")
return bucket_name, encoded_object
def transform_retrieve_file_request(
self,
file_id: str,
optional_params: dict,
litellm_params: dict,
) -> tuple[str, dict]:
raise NotImplementedError("VertexAIFilesConfig does not support file retrieval")
bucket, encoded_object = self._parse_gcs_uri(file_id)
url = f"https://storage.googleapis.com/storage/v1/b/{bucket}/o/{encoded_object}"
return url, {}
def transform_retrieve_file_response(
self,
@ -349,7 +373,21 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig):
logging_obj: LiteLLMLoggingObj,
litellm_params: dict,
) -> OpenAIFileObject:
raise NotImplementedError("VertexAIFilesConfig does not support file retrieval")
response_json = raw_response.json()
gcs_id = response_json.get("id", "")
gcs_id = "/".join(gcs_id.split("/")[:-1]) if gcs_id else ""
return OpenAIFileObject(
id=f"gs://{gcs_id}",
bytes=int(response_json.get("size", 0)),
created_at=_convert_vertex_datetime_to_openai_datetime(
vertex_datetime=response_json.get("timeCreated", "")
),
filename=response_json.get("name", ""),
object="file",
purpose=response_json.get("metadata", {}).get("purpose", "batch"),
status="processed",
status_details=None,
)
def transform_delete_file_request(
self,
@ -357,7 +395,9 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig):
optional_params: dict,
litellm_params: dict,
) -> tuple[str, dict]:
raise NotImplementedError("VertexAIFilesConfig does not support file deletion")
bucket, encoded_object = self._parse_gcs_uri(file_id)
url = f"https://storage.googleapis.com/storage/v1/b/{bucket}/o/{encoded_object}"
return url, {}
def transform_delete_file_response(
self,
@ -365,7 +405,15 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig):
logging_obj: LiteLLMLoggingObj,
litellm_params: dict,
) -> FileDeleted:
raise NotImplementedError("VertexAIFilesConfig does not support file deletion")
file_id = "deleted"
if hasattr(raw_response, "request") and raw_response.request:
url = str(raw_response.request.url)
if "/b/" in url and "/o/" in url:
import urllib.parse
bucket_part = url.split("/b/")[-1].split("/o/")[0]
encoded_name = url.split("/o/")[-1].split("?")[0]
file_id = f"gs://{bucket_part}/{urllib.parse.unquote(encoded_name)}"
return FileDeleted(id=file_id, deleted=True, object="file")
def transform_list_files_request(
self,
@ -389,7 +437,10 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig):
optional_params: dict,
litellm_params: dict,
) -> tuple[str, dict]:
raise NotImplementedError("VertexAIFilesConfig does not support file content retrieval")
file_id = file_content_request.get("file_id", "")
bucket, encoded_object = self._parse_gcs_uri(file_id)
url = f"https://storage.googleapis.com/storage/v1/b/{bucket}/o/{encoded_object}?alt=media"
return url, {}
def transform_file_content_response(
self,
@ -397,7 +448,7 @@ class VertexAIFilesConfig(VertexBase, BaseFilesConfig):
logging_obj: LiteLLMLoggingObj,
litellm_params: dict,
) -> HttpxBinaryResponseContent:
raise NotImplementedError("VertexAIFilesConfig does not support file content retrieval")
return HttpxBinaryResponseContent(response=raw_response)
class VertexAIJsonlFilesTransformation(VertexGeminiConfig):

52
litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
View file

@ -1136,23 +1136,6 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
if VertexGeminiConfig._is_gemini_3_or_newer(model):
if "temperature" not in optional_params:
optional_params["temperature"] = 1.0
# Only add thinkingLevel if model supports it (exclude image models)
if "image" not in model.lower():
thinking_config = optional_params.get("thinkingConfig", {})
if (
"thinkingLevel" not in thinking_config
and "thinkingBudget" not in thinking_config
):
# For gemini-3-flash-preview, default to "minimal" to match Gemini 2.5 Flash behavior
# For other Gemini 3 models, default to "low"
is_gemini3flash = (
"gemini-3-flash-preview" in model.lower()
or "gemini-3-flash" in model.lower()
)
thinking_config["thinkingLevel"] = (
"minimal" if is_gemini3flash else "low"
)
optional_params["thinkingConfig"] = thinking_config
return optional_params
@ -2922,6 +2905,7 @@ class ModelResponseIterator:
self.logging_obj = logging_obj
self.is_function_call = check_is_function_call(logging_obj)
self.cumulative_tool_call_index: int = 0
self.has_seen_tool_calls: bool = False
def chunk_parser(self, chunk: dict) -> Optional["ModelResponseStream"]:
try:
@ -2960,6 +2944,40 @@ class ModelResponseIterator:
cumulative_tool_call_index=self.cumulative_tool_call_index,
)
# Track whether tool_calls have been seen across streaming chunks.
# Gemini sends tool_calls and finishReason in separate chunks,
# so we need to remember if earlier chunks contained tool_calls
# to correctly set finish_reason="tool_calls" per the OpenAI spec.
if not self.has_seen_tool_calls:
for choice in model_response.choices:
if hasattr(choice, "delta") and choice.delta and choice.delta.tool_calls:
self.has_seen_tool_calls = True
break
# Handle final chunk with finishReason but no content.
# _process_candidates skips candidates without "content",
# so the finish_reason from the final chunk is lost.
if not model_response.choices and _candidates:
from litellm.types.utils import Delta, StreamingChoices
for candidate in _candidates:
finish_reason_str = candidate.get("finishReason")
if finish_reason_str is not None:
if self.has_seen_tool_calls:
mapped_finish_reason = "tool_calls"
else:
mapped_finish_reason = VertexGeminiConfig._check_finish_reason(
None, finish_reason_str
)
choice = StreamingChoices(
finish_reason=mapped_finish_reason,
index=candidate.get("index", 0),
delta=Delta(content=None, role=None),
logprobs=None,
enhancements=None,
)
model_response.choices.append(choice)
setattr(model_response, "vertex_ai_grounding_metadata", grounding_metadata) # type: ignore
setattr(model_response, "vertex_ai_url_context_metadata", url_context_metadata) # type: ignore
setattr(model_response, "vertex_ai_safety_ratings", safety_ratings) # type: ignore

86
litellm/main.py
View file

@ -107,6 +107,7 @@ from litellm.realtime_api.main import _realtime_health_check
from litellm.secret_managers.main import get_secret_bool, get_secret_str
from litellm.types.router import GenericLiteLLMParams
from litellm.types.utils import (
CustomPricingLiteLLMParams,
ModelResponseStream,
RawRequestTypedDict,
StreamingChoices,
@ -418,6 +419,8 @@ async def acompletion( # noqa: PLR0915
web_search_options: Optional[OpenAIWebSearchOptions] = None,
# Session management
shared_session: Optional["ClientSession"] = None,
# Per-request JSON schema validation (overrides litellm.enable_json_schema_validation)
enable_json_schema_validation: Optional[bool] = None,
**kwargs,
) -> Union[ModelResponse, CustomStreamWrapper]:
"""
@ -562,6 +565,7 @@ async def acompletion( # noqa: PLR0915
"thinking": thinking,
"web_search_options": web_search_options,
"shared_session": shared_session,
"enable_json_schema_validation": enable_json_schema_validation,
}
if custom_llm_provider is None:
_, custom_llm_provider, _, _ = get_llm_provider(
@ -996,6 +1000,32 @@ def _drop_input_examples_from_tools(
return cleaned_tools
def _build_custom_pricing_entry(
custom_llm_provider: str,
kwargs: dict,
model_info: Optional[dict] = None,
) -> dict:
"""Build a complete model cost entry from kwargs and model_info.
Collects all CustomPricingLiteLLMParams fields present in kwargs and
merges metadata from model_info (mode, supports_prompt_caching, max_tokens)
so that register_model() receives the full pricing configuration.
"""
entry: dict = {"litellm_provider": custom_llm_provider}
for field_name in CustomPricingLiteLLMParams.model_fields:
value = kwargs.get(field_name)
if value is not None:
entry[field_name] = value
if model_info and isinstance(model_info, dict):
for key in ("mode", "supports_prompt_caching", "max_tokens"):
if key in model_info and model_info[key] is not None:
entry.setdefault(key, model_info[key])
return entry
@tracer.wrap()
@client
def completion( # type: ignore # noqa: PLR0915
@ -1047,6 +1077,8 @@ def completion( # type: ignore # noqa: PLR0915
thinking: Optional[AnthropicThinkingParam] = None,
# Session management
shared_session: Optional["ClientSession"] = None,
# Per-request JSON schema validation (overrides litellm.enable_json_schema_validation)
enable_json_schema_validation: Optional[bool] = None,
**kwargs,
) -> Union[ModelResponse, CustomStreamWrapper]:
"""
@ -1167,6 +1199,7 @@ def completion( # type: ignore # noqa: PLR0915
thinking=thinking,
web_search_options=web_search_options,
shared_session=shared_session,
enable_json_schema_validation=enable_json_schema_validation,
**kwargs,
)
api_base = kwargs.get("api_base", None)
@ -1351,27 +1384,16 @@ def completion( # type: ignore # noqa: PLR0915
timeout = float(timeout) # type: ignore
### REGISTER CUSTOM MODEL PRICING -- IF GIVEN ###
if input_cost_per_token is not None and output_cost_per_token is not None:
if (
input_cost_per_token is not None and output_cost_per_token is not None
) or input_cost_per_second is not None:
litellm.register_model(
{
f"{custom_llm_provider}/{model}": {
"input_cost_per_token": input_cost_per_token,
"output_cost_per_token": output_cost_per_token,
"litellm_provider": custom_llm_provider,
}
}
)
elif (
input_cost_per_second is not None
): # time based pricing just needs cost in place
output_cost_per_second = output_cost_per_second
litellm.register_model(
{
f"{custom_llm_provider}/{model}": {
"input_cost_per_second": input_cost_per_second,
"output_cost_per_second": output_cost_per_second,
"litellm_provider": custom_llm_provider,
}
f"{custom_llm_provider}/{model}": _build_custom_pricing_entry(
custom_llm_provider=custom_llm_provider,
kwargs=kwargs,
model_info=model_info,
)
}
)
### BUILD CUSTOM PROMPT TEMPLATE -- IF GIVEN ###
@ -4644,7 +4666,6 @@ def embedding( # noqa: PLR0915
input_cost_per_token = kwargs.get("input_cost_per_token", None)
output_cost_per_token = kwargs.get("output_cost_per_token", None)
input_cost_per_second = kwargs.get("input_cost_per_second", None)
output_cost_per_second = kwargs.get("output_cost_per_second", None)
openai_params = [
"user",
"dimensions",
@ -4694,25 +4715,16 @@ def embedding( # noqa: PLR0915
)
### REGISTER CUSTOM MODEL PRICING -- IF GIVEN ###
if input_cost_per_token is not None and output_cost_per_token is not None:
if (
input_cost_per_token is not None and output_cost_per_token is not None
) or input_cost_per_second is not None:
litellm.register_model(
{
f"{custom_llm_provider}/{model}": {
"input_cost_per_token": input_cost_per_token,
"output_cost_per_token": output_cost_per_token,
"litellm_provider": custom_llm_provider,
}
}
)
if input_cost_per_second is not None: # time based pricing just needs cost in place
output_cost_per_second = output_cost_per_second or 0.0
litellm.register_model(
{
f"{custom_llm_provider}/{model}": {
"input_cost_per_second": input_cost_per_second,
"output_cost_per_second": output_cost_per_second,
"litellm_provider": custom_llm_provider,
}
f"{custom_llm_provider}/{model}": _build_custom_pricing_entry(
custom_llm_provider=custom_llm_provider,
kwargs=kwargs,
model_info=kwargs.get("model_info"),
)
}
)

972
litellm/model_prices_and_context_window_backup.json
View file

File diff suppressed because it is too large Load diff

11
litellm/proxy/_experimental/mcp_server/mcp_server_manager.py
View file

@ -642,6 +642,7 @@ class MCPServerManager:
available_on_public_internet=bool(
getattr(mcp_server, "available_on_public_internet", True)
),
created_at=getattr(mcp_server, "created_at", None),
updated_at=getattr(mcp_server, "updated_at", None),
)
return new_server
@ -2540,8 +2541,8 @@ class MCPServerManager:
url=server.url,
transport=server.transport,
auth_type=server.auth_type,
created_at=datetime.now(),
updated_at=datetime.now(),
created_at=server.created_at,
updated_at=server.updated_at,
teams=[],
mcp_access_groups=server.access_groups or [],
allowed_tools=server.allowed_tools or [],
@ -2620,8 +2621,6 @@ class MCPServerManager:
return list_mcp_servers
def _build_mcp_server_table(self, server: MCPServer) -> LiteLLM_MCPServerTable:
from datetime import datetime
return LiteLLM_MCPServerTable(
server_id=server.server_id,
server_name=server.server_name,
@ -2633,8 +2632,8 @@ class MCPServerManager:
spec_path=server.spec_path,
transport=server.transport,
auth_type=server.auth_type,
created_at=datetime.now(),
updated_at=datetime.now(),
created_at=server.created_at,
updated_at=server.updated_at,
teams=[],
mcp_access_groups=server.access_groups or [],
allowed_tools=server.allowed_tools or [],

17
litellm/proxy/_experimental/mcp_server/server.py
View file

@ -5,7 +5,6 @@ LiteLLM MCP Server Routes
import asyncio
import contextlib
import traceback
import uuid
from datetime import datetime
@ -44,7 +43,10 @@ from litellm.proxy._experimental.mcp_server.utils import (
)
from litellm.proxy._types import UserAPIKeyAuth
from litellm.proxy.auth.ip_address_utils import IPAddressUtils
from litellm.proxy.litellm_pre_call_utils import LiteLLMProxyRequestSetup
from litellm.proxy.litellm_pre_call_utils import (
LiteLLMProxyRequestSetup,
get_chain_id_from_headers,
)
from litellm.types.mcp import MCPAuth
from litellm.types.mcp_server.mcp_server_manager import MCPInfo, MCPServer
from litellm.types.utils import CallTypes, StandardLoggingMCPToolCall
@ -331,6 +333,11 @@ if MCP_AVAILABLE:
try:
# Create a body date for logging
body_data = {"name": name, "arguments": arguments}
# Set trace/session id from raw_headers so spend logs and logging_obj stay consistent (same as A2A)
chain_id = get_chain_id_from_headers(raw_headers)
if chain_id:
body_data["litellm_trace_id"] = chain_id
body_data["litellm_session_id"] = chain_id
request = Request(
scope={
@ -884,6 +891,10 @@ if MCP_AVAILABLE:
# This is intentionally minimal: only async_success_handler / post_call_failure_hook
rules_obj = Rules()
list_tools_call_id = str(uuid.uuid4())
# Derive trace_id from raw_headers when not explicitly passed (same as A2A / MCP call_tool)
effective_litellm_trace_id = litellm_trace_id or get_chain_id_from_headers(
raw_headers
)
spend_logs_metadata: Dict[str, Any] = {
"mcp_operation": "list_tools",
}
@ -896,7 +907,7 @@ if MCP_AVAILABLE:
"model": "MCP: list_tools",
"call_type": CallTypes.list_mcp_tools.value,
"litellm_call_id": list_tools_call_id,
"litellm_trace_id": litellm_trace_id,
"litellm_trace_id": effective_litellm_trace_id,
"metadata": {
"spend_logs_metadata": spend_logs_metadata,
},

0
litellm/proxy/_experimental/out/404.html → litellm/proxy/_experimental/out/404/index.html
View file

0
litellm/proxy/_experimental/out/_not-found.html → litellm/proxy/_experimental/out/_not-found/index.html
View file

0
litellm/proxy/_experimental/out/api-reference.html → litellm/proxy/_experimental/out/api-reference/index.html
View file

0
litellm/proxy/_experimental/out/experimental/api-playground.html → litellm/proxy/_experimental/out/experimental/api-playground/index.html
View file

0
litellm/proxy/_experimental/out/experimental/budgets.html → litellm/proxy/_experimental/out/experimental/budgets/index.html
View file

0
litellm/proxy/_experimental/out/experimental/caching.html → litellm/proxy/_experimental/out/experimental/caching/index.html
View file

0
litellm/proxy/_experimental/out/experimental/claude-code-plugins.html → litellm/proxy/_experimental/out/experimental/claude-code-plugins/index.html
View file

0
litellm/proxy/_experimental/out/experimental/old-usage.html → litellm/proxy/_experimental/out/experimental/old-usage/index.html
View file

0
litellm/proxy/_experimental/out/experimental/prompts.html → litellm/proxy/_experimental/out/experimental/prompts/index.html
View file

0
litellm/proxy/_experimental/out/experimental/tag-management.html → litellm/proxy/_experimental/out/experimental/tag-management/index.html
View file

0
litellm/proxy/_experimental/out/guardrails.html → litellm/proxy/_experimental/out/guardrails/index.html
View file

0
litellm/proxy/_experimental/out/login.html → litellm/proxy/_experimental/out/login/index.html
View file

0
litellm/proxy/_experimental/out/logs.html → litellm/proxy/_experimental/out/logs/index.html
View file

0
litellm/proxy/_experimental/out/mcp/oauth/callback.html → litellm/proxy/_experimental/out/mcp/oauth/callback/index.html
View file

0
litellm/proxy/_experimental/out/model-hub.html → litellm/proxy/_experimental/out/model-hub/index.html
View file

0
litellm/proxy/_experimental/out/model_hub.html → litellm/proxy/_experimental/out/model_hub/index.html
View file

0
litellm/proxy/_experimental/out/model_hub_table.html → litellm/proxy/_experimental/out/model_hub_table/index.html
View file

0
litellm/proxy/_experimental/out/models-and-endpoints.html → litellm/proxy/_experimental/out/models-and-endpoints/index.html
View file

0
litellm/proxy/_experimental/out/onboarding.html → litellm/proxy/_experimental/out/onboarding/index.html
View file

0
litellm/proxy/_experimental/out/organizations.html → litellm/proxy/_experimental/out/organizations/index.html
View file

0
litellm/proxy/_experimental/out/playground.html → litellm/proxy/_experimental/out/playground/index.html
View file

0
litellm/proxy/_experimental/out/policies.html → litellm/proxy/_experimental/out/policies/index.html
View file

0
litellm/proxy/_experimental/out/settings/admin-settings.html → litellm/proxy/_experimental/out/settings/admin-settings/index.html
View file

0
litellm/proxy/_experimental/out/settings/logging-and-alerts.html → litellm/proxy/_experimental/out/settings/logging-and-alerts/index.html
View file

0
litellm/proxy/_experimental/out/settings/router-settings.html → litellm/proxy/_experimental/out/settings/router-settings/index.html
View file

0
litellm/proxy/_experimental/out/settings/ui-theme.html → litellm/proxy/_experimental/out/settings/ui-theme/index.html
View file

0
litellm/proxy/_experimental/out/teams.html → litellm/proxy/_experimental/out/teams/index.html
View file

0
litellm/proxy/_experimental/out/test-key.html → litellm/proxy/_experimental/out/test-key/index.html
View file

0
litellm/proxy/_experimental/out/tools/mcp-servers.html → litellm/proxy/_experimental/out/tools/mcp-servers/index.html
View file

0
litellm/proxy/_experimental/out/tools/vector-stores.html → litellm/proxy/_experimental/out/tools/vector-stores/index.html
View file

0
litellm/proxy/_experimental/out/usage.html → litellm/proxy/_experimental/out/usage/index.html
View file

0
litellm/proxy/_experimental/out/users.html → litellm/proxy/_experimental/out/users/index.html
View file

0
litellm/proxy/_experimental/out/virtual-keys.html → litellm/proxy/_experimental/out/virtual-keys/index.html
View file

30
litellm/proxy/_new_secret_config.yaml
View file

@ -23,33 +23,11 @@ model_list:
guardrails:
- guardrail_name: "airline-competitor-intent"
guardrail_id: "airline-competitor-intent"
- guardrail_name: "tool_policy"
litellm_params:
guardrail: litellm_content_filter
mode: pre_call
default_on: false
competitor_intent_config:
brand_self:
- emirates
- ek
competitors:
- qatar airways
- qatar
- etihad
locations:
- qatar
- doha
- doh
competitor_aliases:
qatar airways: [qr, doha airline]
qatar: [qr]
policy:
competitor_comparison: refuse
possible_competitor_comparison: reframe
threshold_high: 0.70
threshold_medium: 0.45
threshold_low: 0.30
guardrail: tool_policy
mode: [pre_call, post_call]
default_on: true
mcp_servers:
my_http_server:

16
litellm/proxy/_types.py
View file

@ -77,6 +77,7 @@ class SupportedDBObjectType(str, enum.Enum):
PASS_THROUGH_ENDPOINTS = "pass_through_endpoints"
PROMPTS = "prompts"
MODEL_COST_MAP = "model_cost_map"
TOOLS = "tools"
def __str__(self):
return str(self.value)
@ -512,6 +513,7 @@ class LiteLLMRoutes(enum.Enum):
KeyManagementRoutes.KEY_UNBLOCK.value,
KeyManagementRoutes.KEY_BULK_UPDATE.value,
KeyManagementRoutes.TEAM_DAILY_ACTIVITY.value,
KeyManagementRoutes.KEY_RESET_SPEND.value,
]
management_routes = [
@ -1551,6 +1553,8 @@ class NewTeamRequest(TeamBase):
] = None # allow user to set TPM limit for all team members
team_member_key_duration: Optional[str] = None # e.g. "1d", "1w", "1m"
allowed_vector_store_indexes: Optional[List[AllowedVectorStoreIndexItem]] = None
enforced_batch_output_expires_after: Optional[dict] = None
enforced_file_expires_after: Optional[dict] = None
model_config = ConfigDict(protected_namespaces=())
@ -1606,6 +1610,8 @@ class UpdateTeamRequest(LiteLLMPydanticObjectBase):
model_rpm_limit: Optional[Dict[str, int]] = None
model_tpm_limit: Optional[Dict[str, int]] = None
allowed_vector_store_indexes: Optional[List[AllowedVectorStoreIndexItem]] = None
enforced_batch_output_expires_after: Optional[dict] = None
enforced_file_expires_after: Optional[dict] = None
router_settings: Optional[dict] = None
access_group_ids: Optional[List[str]] = None
@ -2128,7 +2134,7 @@ class ConfigGeneralSettings(LiteLLMPydanticObjectBase):
user_header_mappings: Optional[List[UserHeaderMapping]] = None
supported_db_objects: Optional[List[SupportedDBObjectType]] = Field(
None,
description="Fine-grained control over which object types to load from the database when store_model_in_db is True. Available types: 'models', 'mcp', 'guardrails', 'vector_stores', 'pass_through_endpoints', 'prompts', 'model_cost_map'. If not set, all objects are loaded (default behavior).",
description="Fine-grained control over which object types to load from the database when store_model_in_db is True. Available types: 'models', 'mcp', 'guardrails', 'vector_stores', 'pass_through_endpoints', 'prompts', 'model_cost_map', 'tools'. If not set, all objects are loaded (default behavior).",
)
user_mcp_management_mode: Optional[UserMCPManagementMode] = Field(
None,
@ -3372,6 +3378,11 @@ class ProxyErrorTypes(str, enum.Enum):
Team member is already in team
"""
tool_access_denied = "tool_access_denied"
"""
Tool is not in the allowed tools list for this key/team
"""
@classmethod
def get_model_access_error_type_for_object(
cls, object_type: Literal["key", "user", "team", "org", "project"]
@ -3783,6 +3794,8 @@ LiteLLM_ManagementEndpoint_MetadataFields = [
"temp_budget_increase",
"temp_budget_expiry",
"allowed_vector_store_indexes",
"enforced_batch_output_expires_after",
"enforced_file_expires_after",
]
LiteLLM_ManagementEndpoint_MetadataFields_Premium = [
@ -4154,6 +4167,7 @@ class ToolDiscoveryQueueItem(TypedDict, total=False):
key_hash: Optional[str] # hash of virtual key that triggered discovery
team_id: Optional[str] # team that triggered discovery
key_alias: Optional[str] # human-readable key alias
user_agent: Optional[str] # HTTP User-Agent of the caller
class LiteLLM_ManagedFileTable(LiteLLMPydanticObjectBase):

49
litellm/proxy/agent_endpoints/a2a_endpoints.py
View file

@ -69,6 +69,7 @@ async def _handle_stream_message(
from litellm.a2a_protocol.main import A2A_SDK_AVAILABLE
if not A2A_SDK_AVAILABLE:
async def _error_stream():
yield json.dumps(
{
@ -106,7 +107,12 @@ async def _handle_stream_message(
proxy_server_request=proxy_server_request,
)
if use_proxy_hooks and user_api_key_dict is not None and request_data is not None and proxy_logging_obj is not None:
if (
use_proxy_hooks
and user_api_key_dict is not None
and request_data is not None
and proxy_logging_obj is not None
):
from litellm.proxy.common_request_processing import (
ProxyBaseLLMRequestProcessing,
)
@ -119,20 +125,27 @@ async def _handle_stream_message(
return json.dumps(obj) + "\n"
def _ndjson_error(proxy_exc: Any) -> str:
return json.dumps(
{
"jsonrpc": "2.0",
"id": request_id,
"error": {
"code": -32603,
"message": getattr(
proxy_exc, "message", f"Streaming error: {proxy_exc!s}"
),
},
}
) + "\n"
return (
json.dumps(
{
"jsonrpc": "2.0",
"id": request_id,
"error": {
"code": -32603,
"message": getattr(
proxy_exc,
"message",
f"Streaming error: {proxy_exc!s}",
),
},
}
)
+ "\n"
)
async for line in ProxyBaseLLMRequestProcessing.async_streaming_data_generator(
async for (
line
) in ProxyBaseLLMRequestProcessing.async_streaming_data_generator(
response=a2a_stream,
user_api_key_dict=user_api_key_dict,
request_data=request_data,
@ -151,7 +164,12 @@ async def _handle_stream_message(
yield json.dumps(chunk) + "\n"
except Exception as e:
verbose_proxy_logger.exception(f"Error streaming A2A response: {e}")
if use_proxy_hooks and proxy_logging_obj is not None and user_api_key_dict is not None and request_data is not None:
if (
use_proxy_hooks
and proxy_logging_obj is not None
and user_api_key_dict is not None
and request_data is not None
):
transformed_exception = await proxy_logging_obj.post_call_failure_hook(
user_api_key_dict=user_api_key_dict,
original_exception=e,
@ -382,6 +400,7 @@ async def invoke_agent_a2a(
agent_id=agent.agent_id,
metadata=data.get("metadata", {}),
proxy_server_request=data.get("proxy_server_request"),
litellm_logging_obj=logging_obj,
)
response = await proxy_logging_obj.post_call_success_hook(

55
litellm/proxy/auth/auth_checks.py
View file

@ -58,6 +58,10 @@ from litellm.proxy._types import (
)
from litellm.proxy.auth.route_checks import RouteChecks
from litellm.proxy.db.exception_handler import PrismaDBExceptionHandler
from litellm.proxy.guardrails.tool_name_extraction import (
TOOL_CAPABLE_CALL_TYPES,
extract_request_tool_names,
)
from litellm.proxy.route_llm_request import route_request
from litellm.proxy.utils import PrismaClient, ProxyLogging, log_db_metrics
from litellm.router import Router
@ -220,7 +224,48 @@ async def _run_project_checks(
)
async def common_checks(
async def check_tools_allowlist(
request_body: dict,
valid_token: Optional[UserAPIKeyAuth],
team_object: Optional[LiteLLM_TeamTable],
route: str,
) -> None:
"""
Enforce key/team tool allowlist (metadata.allowed_tools). No DB in hot path —
effective allowlist is read from valid_token.metadata and valid_token.team_metadata.
Raises ProxyException with tool_access_denied if a tool is not allowed.
"""
from litellm.litellm_core_utils.api_route_to_call_types import (
get_call_types_for_route,
)
if valid_token is None:
return
call_types = get_call_types_for_route(route)
if not call_types or not any(ct.value in TOOL_CAPABLE_CALL_TYPES for ct in call_types):
return
tool_names = extract_request_tool_names(route, request_body)
if not tool_names:
return
key_meta = (valid_token.metadata or {}) if isinstance(valid_token.metadata, dict) else {}
team_meta = (valid_token.team_metadata or {}) if isinstance(valid_token.team_metadata, dict) else {}
key_allowed = key_meta.get("allowed_tools")
team_allowed = team_meta.get("allowed_tools")
effective = key_allowed if (isinstance(key_allowed, list) and len(key_allowed) > 0) else team_allowed
if not isinstance(effective, list) or len(effective) == 0:
return
allowed_set = {str(t) for t in effective}
disallowed = [n for n in tool_names if n not in allowed_set]
if disallowed:
raise ProxyException(
message=f"Tool(s) {disallowed} are not in the allowed tools list for this key/team.",
type=ProxyErrorTypes.tool_access_denied,
param="tools",
code=status.HTTP_403_FORBIDDEN,
)
async def common_checks( # noqa: PLR0915
request_body: dict,
team_object: Optional[LiteLLM_TeamTable],
user_object: Optional[LiteLLM_UserTable],
@ -473,6 +518,14 @@ async def common_checks(
valid_token=valid_token,
)
# 12. [OPTIONAL] Tool allowlist - key/team allowed_tools (no DB in hot path)
await check_tools_allowlist(
request_body=request_body,
valid_token=valid_token,
team_object=team_object,
route=route,
)
return True

31
litellm/proxy/auth/user_api_key_auth.py
View file

@ -1752,20 +1752,21 @@ async def _run_post_custom_auth_checks(
if _project_obj is not None:
valid_token.project_metadata = _project_obj.metadata
_ = await common_checks(
request=request,
request_body=request_data,
team_object=_team_obj,
user_object=user_object,
end_user_object=end_user_object,
general_settings=general_settings,
global_proxy_spend=None,
route=route,
llm_router=llm_router,
proxy_logging_obj=proxy_logging_obj,
valid_token=valid_token,
skip_budget_checks=False,
project_object=_project_obj,
)
if general_settings.get("custom_auth_run_common_checks", False):
_ = await common_checks(
request=request,
request_body=request_data,
team_object=_team_obj,
user_object=user_object,
end_user_object=end_user_object,
general_settings=general_settings,
global_proxy_spend=None,
route=route,
llm_router=llm_router,
proxy_logging_obj=proxy_logging_obj,
valid_token=valid_token,
skip_budget_checks=False,
project_object=_project_obj,
)
return valid_token

38
litellm/proxy/batches_endpoints/endpoints.py
View file

@ -23,6 +23,7 @@ from litellm.proxy.common_utils.openai_endpoint_utils import (
from litellm.proxy.openai_files_endpoints.common_utils import (
_is_base64_encoded_unified_file_id,
decode_model_from_file_id,
encode_batch_response_ids,
encode_file_id_with_model,
get_batch_from_database,
get_credentials_for_model,
@ -118,6 +119,22 @@ async def create_batch( # noqa: PLR0915
or "openai"
)
_create_batch_data = LiteLLMBatchCreateRequest(**data)
# Apply team-level batch output expiry enforcement
team_metadata = user_api_key_dict.team_metadata or {}
enforced_batch_expiry = team_metadata.get(
"enforced_batch_output_expires_after"
)
if enforced_batch_expiry is not None:
if "anchor" not in enforced_batch_expiry or "seconds" not in enforced_batch_expiry:
raise HTTPException(
status_code=400,
detail={
"error": "enforced_batch_output_expires_after must contain 'anchor' and 'seconds' keys",
},
)
_create_batch_data["output_expires_after"] = enforced_batch_expiry
input_file_id = _create_batch_data.get("input_file_id", None)
unified_file_id: Union[str, Literal[False]] = False
@ -242,7 +259,9 @@ async def create_batch( # noqa: PLR0915
custom_llm_provider=credentials["custom_llm_provider"],
**_create_batch_data # type: ignore
)
encode_batch_response_ids(response, model=model_param)
verbose_proxy_logger.debug(f"Created batch using model: {model_param}")
else:
# SCENARIO 3: Fallback to custom_llm_provider (uses env variables)
@ -440,8 +459,9 @@ async def retrieve_batch( # noqa: PLR0915
custom_llm_provider=credentials["custom_llm_provider"],
**data # type: ignore
)
encode_batch_response_ids(response, model=model_from_id)
verbose_proxy_logger.debug(
f"Retrieved batch using model: {model_from_id}, original_id: {original_batch_id}"
)
@ -633,7 +653,13 @@ async def list_batches(
limit=limit,
**data # type: ignore
)
# Encode batch IDs in the list response so clients can use
# them for retrieve/cancel/file downloads through the proxy.
if response and hasattr(response, "data") and response.data:
for batch in response.data:
encode_batch_response_ids(batch, model=model_param)
verbose_proxy_logger.debug(f"Listed batches using model: {model_param}")
# SCENARIO 2 (alternative): target_model_names based routing
@ -809,7 +835,9 @@ async def cancel_batch(
custom_llm_provider=credentials["custom_llm_provider"],
**data # type: ignore
)
encode_batch_response_ids(response, model=model_from_id)
verbose_proxy_logger.debug(
f"Cancelled batch using model: {model_from_id}, original_id: {original_batch_id}"
)

30
litellm/proxy/common_request_processing.py
View file

@ -29,7 +29,7 @@ from litellm.constants import (
MAX_PAYLOAD_SIZE_FOR_DEBUG_LOG,
STREAM_SSE_DATA_PREFIX,
)
from litellm.litellm_core_utils.dd_tracing import set_active_span_tag, tracer
from litellm.litellm_core_utils.dd_tracing import tracer
from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
from litellm.litellm_core_utils.llm_response_utils.get_headers import (
get_response_headers,
@ -41,6 +41,7 @@ from litellm.proxy.common_utils.callback_utils import (
get_logging_caching_headers,
get_remaining_tokens_and_requests_from_request_data,
)
from litellm.proxy.dd_span_tagger import DDSpanTagger
from litellm.proxy.route_llm_request import route_request
from litellm.proxy.utils import ProxyLogging
from litellm.router import Router
@ -245,26 +246,6 @@ async def create_response(
)
def _add_dd_apm_tags_for_litellm_call_id(litellm_call_id: Optional[str]) -> None:
"""
Attach LiteLLM call id to the active Datadog APM span.
This enables searching APM traces by LiteLLM call id returned in
`x-litellm-call-id`.
"""
if not litellm_call_id:
return
try:
set_active_span_tag("litellm.call_id", str(litellm_call_id))
except Exception:
# Tagging is best-effort and should never impact request processing.
verbose_proxy_logger.debug(
"Failed to tag active ddtrace span with litellm.call_id",
exc_info=True,
)
def _override_openai_response_model(
*,
response_obj: Any,
@ -518,6 +499,7 @@ class ProxyBaseLLMRequestProcessing:
"aembedding",
"aresponses",
"_arealtime",
"_aresponses_websocket",
"aget_responses",
"adelete_responses",
"acancel_responses",
@ -662,7 +644,11 @@ class ProxyBaseLLMRequestProcessing:
self.data["litellm_call_id"] = request.headers.get(
"x-litellm-call-id", str(uuid.uuid4())
)
_add_dd_apm_tags_for_litellm_call_id(self.data.get("litellm_call_id"))
DDSpanTagger.tag_call_id(self.data.get("litellm_call_id"))
DDSpanTagger.tag_request(
user_api_key_dict=user_api_key_dict,
requested_model=self.data.get("model"),
)
### AUTO STREAM USAGE TRACKING ###
# If always_include_stream_usage is enabled and this is a streaming request

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