Skip to main content

APIs

Llama Stack exposes OpenAI-compatible REST APIs. For the full endpoint list, see the API Reference.

Stability Levels​

APIs are organized by maturity. Stable APIs (/v1/) follow semantic versioning and won't break between minor releases. Experimental APIs (/v1alpha/, /v1beta/) may change based on feedback. See API Leveling for details.

How the APIs work together​

Files + Vector Stores + Responses​

These three APIs combine to give you RAG in a few calls:

  1. Upload a document via /v1/files
  2. Create a vector store and attach the file via /v1/vector_stores
  3. Llama Stack automatically chunks, embeds, and indexes the document
  4. Use file_search in a /v1/responses request to search it

See File Operations and Vector Store Integration for details.

Tool Runtime​

The Responses API orchestrates tools server-side. The model decides which tools to call, Llama Stack executes them and feeds results back automatically:

  • file_search - searches your vector stores
  • web_search - queries configured search providers
  • MCP - connects to any MCP server
  • Custom functions - your own function tools

Providers and External APIs​

Not every provider supports every feature. See API Providers for how providers map to APIs, and External APIs for extending Llama Stack with custom endpoints.