Skip to main content
LingxiGraph is a provider-neutral durable graph runtime. It represents agent workflows as typed state graphs and adds checkpoints, streaming events, human approval, version pinning, and a multi-tenant Agent Server.

Install and choose a runtime

Choose the dependency-free embedded core, local development server, or complete production stack.

Build your first graph

Create an executable state graph from TypedDict schemas, nodes, and edges.

Start Agent Server

Run graphs through Studio, REST, SSE, and the Python SDK.

Browse the API

Explore resource models, authentication, errors, and every endpoint.

​
What you can build

  • Long-running agent workflows with checkpoints and human approval.
  • Supervisor, handoff, swarm, group-chat, and parallel-review systems.
  • Reliable task execution with stable idempotency keys, budgets, retries, and dead letters.
  • Hybrid agent networks connected through Coze, OpenAI-compatible APIs, A2A, or MCP.
  • Enterprise platforms that require OIDC, RBAC, tenant isolation, RLS, and OpenTelemetry.

​
Two product layers

For a first evaluation, follow Installation → Build your first graph → Agent Server. It takes about ten minutes.

​
Version and compatibility

This documentation targets LingxiGraph 2.0.0 on Python 3.11–3.13. Public server endpoints live under /v1. Each run pins a graph ID and version, so a deployment cannot change queued or paused work. Continue with Architecture and Durable execution when you need the underlying guarantees.