Skip to content

LangGraph OpenAI Serve

Serve registered LangGraph graphs as OpenAI model values through a standard OpenAI-compatible /v1 API.

Build your first app

  • Use the package

    Install LGOS, register your graph, and expose it through /v1.

    Getting started

  • Explore the demo stack

    Run the independent API, graphs, PostgreSQL, Chainlit, Open WebUI, and Bifrost examples.

    Demo capabilities

  • Use your SDK

    Connect from Python or JavaScript, with regular or streaming responses.

    OpenAI clients

  • Register a graph

    Adapt schemas, publish runtime settings, configure streaming, and enable interrupts.

    Custom graphs Runtime settings

  • Understand the contract

    See how OpenAI requests flow through FastAPI into LangGraph and back.

    Architecture

  • Use a gateway

    Preserve metadata, discovery, tool calls, and streaming behavior through an OpenAI-compatible proxy.

    Proxy guide

OpenAI compatibility is the public contract

LGOS keeps client ingestion on the OpenAI SDK path. Graph-specific behavior is adapted behind /v1, so clients do not need a project-specific protocol.

Package versus demo

The package supplies the server integration and public Python API. The repository's demo/ directory supplies example graphs, applications, images, UIs, gateway configuration, and database-backed workflows. Demo components are independently locked and are not installed with LGOS.

Configure And Operate

Use the runtime settings guide to publish safe per-request graph settings, the authentication guide to add bearer tokens, and the proxy guide to preserve the contract through a gateway. See the package reference for endpoints, settings, events, and public classes. Demo-owned models, settings, and commands have their own reference.