Implementation:BerriAI Litellm Managed Vector Stores

Attribute	Value
Sources	enterprise/litellm_enterprise/proxy/hooks/managed_vector_stores.py
Domains	Vector Stores, Proxy Hooks, Enterprise, Resource Management
Last Updated	2026-02-15 16:00 GMT

Overview

_PROXY_LiteLLMManagedVectorStores is a proxy hook that manages vector stores across multiple LLM providers using unified IDs, providing transparent CRUD operations, access control, and router integration for vector store search and retrieval.

Description

This class extends both CustomLogger and BaseManagedResource[VectorStoreCreateResponse] to provide managed vector store functionality in the LiteLLM proxy. It follows the same unified ID pattern as Managed Files.

Core Capabilities:

• Cross-Model Vector Store Creation -- Creates vector stores across multiple models and generates a unified base64-encoded ID.
• Unified ID Management -- Encodes resource type, UUID, target model names, provider resource ID, and model ID.
• Access Control -- Validates user access to managed vector stores before operations.
• Pre-Call Hook -- Intercepts avector_store_search, avector_store_retrieve, and avector_store_delete calls to resolve unified IDs and enforce access.
• Deployment Filtering -- Filters deployments to only those that have the vector store available.
• User Listing -- Lists vector stores created by a specific user with pagination support.

The resource_type property returns "vector_store" and the database table is litellm_managedvectorstoretable.

Usage

Instantiated by the proxy server and registered as a proxy hook. Requires InternalUsageCache and PrismaClient.

Code Reference

Source Location

enterprise/litellm_enterprise/proxy/hooks/managed_vector_stores.py

Signature

class _PROXY_LiteLLMManagedVectorStores(CustomLogger, BaseManagedResource[VectorStoreCreateResponse]):
    def __init__(self, internal_usage_cache: InternalUsageCache, prisma_client: PrismaClient): ...

    async def acreate_vector_store(
        self, create_request, llm_router, target_model_names_list, litellm_parent_otel_span, user_api_key_dict
    ) -> VectorStoreCreateResponse: ...
    async def alist_vector_stores(self, user_api_key_dict, limit, after, order) -> Dict[str, Any]: ...
    async def check_vector_store_access(self, vector_store_id, user_api_key_dict) -> bool: ...
    async def check_managed_vector_store_access(self, data, user_api_key_dict) -> bool: ...
    async def async_pre_call_hook(self, user_api_key_dict, cache, data, call_type) -> Union[Exception, str, Dict, None]: ...
    async def async_post_call_success_hook(self, data, user_api_key_dict, response) -> Any: ...
    async def async_filter_deployments(self, model, healthy_deployments, messages, request_kwargs, parent_otel_span) -> List[Dict]: ...

Import

from litellm_enterprise.proxy.hooks.managed_vector_stores import _PROXY_LiteLLMManagedVectorStores

I/O Contract

Inputs

Parameter	Type	Description
internal_usage_cache	InternalUsageCache	Cache for storing vector store mappings.
prisma_client	PrismaClient	Database client for persistent storage.
create_request	VectorStoreCreateOptionalRequestParams	Vector store creation parameters.
target_model_names_list	List[str]	Models to create vector stores across.
user_api_key_dict	UserAPIKeyAuth	Authentication info for access control.

Outputs

Output	Type	Description
Vector store response	VectorStoreCreateResponse	Response with unified vector store ID.
List response	Dict[str, Any]	Paginated list of vector stores with metadata.
Access check	bool	Whether the user has access to the resource.

Usage Examples

from litellm_enterprise.proxy.hooks.managed_vector_stores import _PROXY_LiteLLMManagedVectorStores

managed_vs = _PROXY_LiteLLMManagedVectorStores(
    internal_usage_cache=cache,
    prisma_client=prisma_client,
)

# Create a managed vector store across models
response = await managed_vs.acreate_vector_store(
    create_request={"name": "my-vector-store"},
    llm_router=router,
    target_model_names_list=["openai/gpt-4"],
    litellm_parent_otel_span=None,
    user_api_key_dict=user_key,
)

Related Pages

• BerriAI_Litellm_Vector_Store_Endpoints -- REST management endpoints for vector stores
• BerriAI_Litellm_Managed_Files -- Managed files hook with similar unified ID patterns