Implementation:BerriAI Litellm Vector Stores API

Property	Value
sources	litellm/vector_stores/main.py
domains	Vector Stores, Search, Retrieval, OpenAI
last_updated	2026-02-15 16:00 GMT

Overview

The Vector Stores API module provides functions for creating vector stores and searching them for relevant content chunks, with built-in mock response support and integration with the vector store registry for credential management.

Description

This module implements two primary operations through sync/async function pairs: create/acreate for creating vector stores and search/asearch for searching vector stores. Both use the @client decorator and default to "openai" as the provider. The create function accepts configuration for file IDs, expiration policies, chunking strategies, and metadata, delegating to BaseLLMHTTPHandler.vector_store_create_handler(). The search function accepts queries (string or list), filters, max results, ranking options, and rewrite query flags, delegating to BaseLLMHTTPHandler.vector_store_search_handler(). The search function integrates with litellm.vector_store_registry for automatic credential injection. Both functions support provider resolution via /-prefixed provider strings (e.g., openai/). The module provides mock_vector_store_search_response() and mock_vector_store_create_response() for testing.

Usage

Import this module when you need to create a new vector store or search an existing one for relevant document chunks. It is the primary interface for vector store lifecycle management and semantic search through LiteLLM.

Code Reference

Source Location

Property	Value
Repository	github.com/BerriAI/litellm
File	litellm/vector_stores/main.py
Lines	482
Module	litellm.vector_stores.main

Signature

@client
def create(
    name: Optional[str] = None,
    file_ids: Optional[List[str]] = None,
    expires_after: Optional[Dict] = None,
    chunking_strategy: Optional[Dict] = None,
    metadata: Optional[Dict[str, str]] = None,
    extra_headers: Optional[Dict[str, Any]] = None,
    extra_query: Optional[Dict[str, Any]] = None,
    extra_body: Optional[Dict[str, Any]] = None,
    timeout: Optional[Union[float, httpx.Timeout]] = None,
    custom_llm_provider: Optional[str] = None,
    **kwargs,
) -> Union[VectorStoreCreateResponse, Coroutine]

@client
def search(
    vector_store_id: str,
    query: Union[str, List[str]],
    filters: Optional[Dict] = None,
    max_num_results: Optional[int] = None,
    ranking_options: Optional[Dict] = None,
    rewrite_query: Optional[bool] = None,
    extra_headers: Optional[Dict[str, Any]] = None,
    extra_query: Optional[Dict[str, Any]] = None,
    extra_body: Optional[Dict[str, Any]] = None,
    timeout: Optional[Union[float, httpx.Timeout]] = None,
    custom_llm_provider: Optional[str] = None,
    **kwargs,
) -> Union[VectorStoreSearchResponse, Coroutine]

Import

from litellm.vector_stores.main import (
    create, acreate,
    search, asearch,
    mock_vector_store_search_response,
    mock_vector_store_create_response,
)

I/O Contract

Inputs

Parameter	Type	Required	Description
name	Optional[str]	No	Name of the vector store to create
file_ids	Optional[List[str]]	No	File IDs to include in the vector store
expires_after	Optional[Dict]	No	Expiration policy for the vector store
chunking_strategy	Optional[Dict]	No	Chunking strategy for files in the store
metadata	Optional[Dict[str, str]]	No	Up to 16 key-value metadata pairs
vector_store_id	str	For search	The ID of the vector store to search
query	Union[str, List[str]]	For search	The search query or array of queries
filters	Optional[Dict]	No	Attribute-based filter for search
max_num_results	Optional[int]	No	Maximum results to return (1-50, default 10)
ranking_options	Optional[Dict]	No	Ranking configuration for search results
rewrite_query	Optional[bool]	No	Whether to rewrite the query for vector search

Outputs

Function	Return Type	Description
create	VectorStoreCreateResponse	Created store with id, status, file_counts, metadata
search	VectorStoreSearchResponse	Search results with scores, content, and query info

Usage Examples

import litellm

# Create a vector store
vs = litellm.vector_stores.create(
    name="My Knowledge Base",
    file_ids=["file_abc123", "file_def456"],
    chunking_strategy={"type": "auto"},
)
print(f"Vector Store ID: {vs.id}, Status: {vs.status}")

import litellm

# Search a vector store
results = litellm.vector_stores.search(
    vector_store_id="vs_abc123",
    query="What is machine learning?",
    max_num_results=5,
)
for result in results.data:
    print(f"Score: {result.score}")
    for content in result.content:
        print(f"  {content.text[:100]}...")

Related Pages

• BerriAI_Litellm_Vector_Store_Files_API -- File management within vector stores (create, list, update, delete files)
• BerriAI_Litellm_Vector_Store_Registry -- Registry that provides credentials and configuration for vector store operations
• BerriAI_Litellm_Rerank_API -- Reranking API that can be combined with vector store search results