Implementation:BerriAI Litellm Vector Store Endpoints

Attribute	Value
Sources	enterprise/litellm_enterprise/proxy/vector_stores/endpoints.py
Domains	Vector Stores, REST API, Management Endpoints, Enterprise
Last Updated	2026-02-15 16:00 GMT

Overview

This module defines FastAPI management endpoints for creating, listing, retrieving, updating, and deleting vector store configurations in the LiteLLM proxy database and in-memory registry.

Description

The module provides a FastAPI APIRouter with five endpoints for vector store CRUD operations:

• POST /vector_store/new -- Creates a new vector store entry in the database and registers it in the in-memory VectorStoreRegistry.
• GET /vector_store/list -- Lists all vector stores with pagination, using the database as the source of truth and syncing the in-memory registry.
• POST /vector_store/delete -- Deletes a vector store from both the database and the in-memory registry.
• POST /vector_store/info -- Retrieves detailed information about a specific vector store by ID.
• POST /vector_store/update -- Updates vector store fields (name, description, metadata, litellm_params).

All endpoints require authentication via user_api_key_auth. Vector store metadata is JSON-serialized for database storage. The litellm_params field is validated through GenericLiteLLMParams before storage.

Usage

These endpoints are mounted on the LiteLLM proxy FastAPI application to provide administrative management of vector store configurations.

Code Reference

Source Location

enterprise/litellm_enterprise/proxy/vector_stores/endpoints.py

Signature

router = APIRouter()

@router.post("/vector_store/new")
async def new_vector_store(
    vector_store: LiteLLM_ManagedVectorStore,
    user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth),
): ...

@router.get("/vector_store/list", response_model=LiteLLM_ManagedVectorStoreListResponse)
async def list_vector_stores(
    user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth),
    page: int = 1, page_size: int = 100,
): ...

@router.post("/vector_store/delete")
async def delete_vector_store(
    data: VectorStoreDeleteRequest,
    user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth),
): ...

@router.post("/vector_store/info", response_model=ResponseLiteLLM_ManagedVectorStore)
async def get_vector_store_info(
    data: VectorStoreInfoRequest,
    user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth),
): ...

@router.post("/vector_store/update")
async def update_vector_store(
    data: VectorStoreUpdateRequest,
    user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth),
): ...

Import

from litellm_enterprise.proxy.vector_stores.endpoints import router

I/O Contract

Inputs

Endpoint	Parameter	Type	Description
POST /vector_store/new	vector_store	LiteLLM_ManagedVectorStore	Vector store ID, provider, name, description, metadata, litellm_params.
GET /vector_store/list	page, page_size	int	Pagination parameters (defaults: 1, 100).
POST /vector_store/delete	data	VectorStoreDeleteRequest	Contains vector_store_id.
POST /vector_store/info	data	VectorStoreInfoRequest	Contains vector_store_id.
POST /vector_store/update	data	VectorStoreUpdateRequest	Fields to update including vector_store_id.

Outputs

Endpoint	Response Type	Description
POST /vector_store/new	dict	{"status": "success", "message": ..., "vector_store": ...}
GET /vector_store/list	LiteLLM_ManagedVectorStoreListResponse	Paginated list with data, total_count, current_page, total_pages.
POST /vector_store/delete	dict	{"message": "Vector store ... deleted successfully"}
POST /vector_store/info	ResponseLiteLLM_ManagedVectorStore	{"vector_store": ...}
POST /vector_store/update	dict	{"vector_store": ...} with updated values.

Usage Examples

import httpx

# Create a new vector store
response = httpx.post(
    "http://localhost:4000/vector_store/new",
    json={
        "vector_store_id": "vs-openai-001",
        "custom_llm_provider": "openai",
        "vector_store_name": "Product Knowledge Base",
    },
    headers={"Authorization": "Bearer sk-..."},
)

# List all vector stores
response = httpx.get(
    "http://localhost:4000/vector_store/list?page=1&page_size=50",
    headers={"Authorization": "Bearer sk-..."},
)

Related Pages

• BerriAI_Litellm_Managed_Vector_Stores -- Proxy hook for managed vector store operations