LangGraph

framework active open-source

A library for building stateful, multi-actor applications with LLMs, built on top of LangChain. LangGraph models agent workflows as graphs where nodes are functions (LLM calls, tools, logic) and edges are state transitions. It enables complex patterns like cycles, branching, and human-in-the-loop that are hard to express in linear chains.

Implements

Concepts this tool claims to implement:

  • StateGraph class for defining agent workflows as graphs. Built-in support for supervisor patterns, tool nodes, and conditional routing.

  • Native support for multi-agent patterns including supervisor-worker, hierarchical teams, and agent handoffs.

  • Explicit TypedDict state schemas that flow through the graph. State channels, reducers, and checkpointing for persistence.

  • interrupt_before and interrupt_after for pausing execution. Human approval nodes and state modification patterns.

  • Workflow secondary

    Can express both agent-like (dynamic routing) and workflow-like (deterministic paths) patterns in the same graph.

Integration Surfaces

  • Python SDK
  • LangGraph Platform (managed hosting)
  • LangSmith (observability)

Details

Vendor
LangChain Inc.
License
MIT
Runs On
local, cloud
Used By
human, agent, system

Notes

LangGraph emerged as the recommended approach for complex agents in the LangChain ecosystem, addressing limitations of the linear AgentExecutor. Its graph-based model makes state transitions explicit and debuggable.