Implementation:BerriAI Litellm Provider Endpoints Support

Overview

Description

The provider_endpoints_support.json file is a comprehensive capability matrix that maps every LiteLLM provider to the API endpoints it supports. This file documents which providers can handle chat completions, embeddings, image generation, audio transcription, batches, reranking, and many other endpoint types.

The file serves as both documentation and a validation resource. It is used by CI tests to verify that all provider folders in the codebase have corresponding entries, and that all documented endpoints in the project's documentation sidebar are defined in this file. It also provides display names and documentation URLs for each provider.

Usage

This file is consumed by code coverage tests to ensure documentation completeness:

• tests/code_coverage_tests/check_provider_folders_documented.py -- Verifies every provider folder in litellm/llms/ has a matching entry in this file
• tests/code_coverage_tests/check_endpoint_coverage.py -- Ensures all endpoints listed in sidebars.js are defined in this file

# Loading the file in tests:
file_path = repo_root / "provider_endpoints_support.json"
with open(file_path) as f:
    data = json.load(f)
providers = data.get("providers", {})

Data Schema

Top-Level Structure

{
    "_comment": "This file defines which endpoints are supported by each LiteLLM provider",
    "_schema": {
        "provider_slug": {
            "display_name": "...",
            "url": "...",
            "endpoints": {
                "chat_completions": "description...",
                "messages": "description...",
                ...
            }
        }
    },
    "providers": {
        "openai": { ... },
        "anthropic": { ... },
        "bedrock": { ... },
        ...
    }
}

Self-Documenting Schema

The _schema key provides documentation for the file structure itself, describing each field and endpoint type in human-readable format.

Schema Fields

Provider Entry Object

Field	Type	Required	Description
display_name	String	Yes	Human-readable name with provider slug in backticks (e.g., "Anthropic (`anthropic`)")
url	String	No	URL to the provider's documentation page on LiteLLM docs
endpoints	Object	Yes	Map of endpoint names to boolean support flags

Endpoint Support Flags

Each endpoint in the endpoints object is a boolean (true/false) indicating whether the provider supports that endpoint type:

Endpoint	API Path	Description
chat_completions	/chat/completions	Standard chat completion endpoint (OpenAI format)
messages	/messages	Anthropic Messages API format
responses	/responses	OpenAI/Anthropic unified Responses API
embeddings	/embeddings	Text embedding generation
image_generations	/image/generations	Image generation from text prompts
audio_transcriptions	/audio/transcriptions	Audio-to-text transcription
audio_speech	/audio/speech	Text-to-speech generation
moderations	/moderations	Content moderation classification
batches	/batches	Batch API for asynchronous processing
rerank	/rerank	Document reranking
ocr	/ocr	Optical character recognition
search	/search	Search endpoint
skills	/skills	Skills endpoint
interactions	Varies	Google AI Interactions API
a2a_(Agent Gateway)	/a2a/{agent}/message/send	Agent-to-Agent Protocol support
container	/containers	OpenAI Containers API
container_file	/containers/{id}/files	Container file operations
compact	/responses/compact	Compact responses endpoint
files	/files	File upload/management operations
image_edits	/images/edits	Image editing endpoint
vector_stores_create	/vector_stores	Vector store creation
vector_stores_search	/vector_stores/{id}/search	Vector store search
video_generations	/videos/generations	Video generation
count_tokens	Varies	Token counting support

Sample Provider Entries

Provider	chat_completions	messages	responses	embeddings	batches	Notable Features
openai	true	true	true	true	true	Full endpoint coverage
anthropic	true	true	true	false	true	skills, count_tokens
bedrock	true	true	true	true	false	Broad model support
vertex_ai	true	true	true	true	true	Google Cloud models
azure	true	true	true	true	true	Azure OpenAI
a2a	true	false	false	false	false	Agent-to-Agent only
assemblyai	false	false	false	false	false	Audio transcription only

Usage Examples

CI: Checking Provider Folder Coverage

# From tests/code_coverage_tests/check_provider_folders_documented.py:
def test_all_provider_folders_documented():
    """
    1. Every provider folder in litellm/llms/ has a corresponding entry
       in provider_endpoints_support.json
    2. Every provider in litellm/llms/openai_like/providers.json is
       documented in provider_endpoints_support.json
    """
    data = load_provider_endpoints_support()
    documented_providers = get_documented_provider_slugs(data)
    # Compare against provider folders in litellm/llms/
    ...

CI: Checking Endpoint Coverage

# From tests/code_coverage_tests/check_endpoint_coverage.py:
def test_endpoint_coverage():
    """
    Validates that each endpoint documented in sidebars.js has a
    corresponding entry in provider_endpoints_support.json.
    """
    file_path = repo_root / "provider_endpoints_support.json"
    data = json.load(open(file_path))
    defined_endpoints = get_defined_endpoints(data)
    # Compare against sidebars.js endpoint listings
    ...

Querying Provider Capabilities

The data structure allows simple lookups to determine if a provider supports a specific endpoint:

# Conceptual usage:
import json

with open("provider_endpoints_support.json") as f:
    data = json.load(f)

# Check if Anthropic supports batch API
supports_batches = data["providers"]["anthropic"]["endpoints"].get("batches", False)
# True

# Check if Bedrock supports image generation
supports_images = data["providers"]["bedrock"]["endpoints"].get("image_generations", False)
# True/False depending on entry

Related Pages

• Model Prices Database - Complementary per-model pricing data
• Model Prices Backup - Backup of model pricing data
• CircleCI Config - CI pipeline that runs coverage tests using this file