Open source · ★ 43,232 GitHub stars (as of 2026-10-14)
LangGraph
LangGraph is a low-level orchestration framework for building stateful, long-running agents.
Overview
LangGraph is a low-level orchestration framework for building, managing and deploying long-running, stateful agents. It is a Python library, and LangGraph.js is also available for JS/TS.
It supports durable execution, which resumes from the point of interruption after a failure; human-in-the-loop, in which a person checks and modifies the state during execution; and short- and long-term memory. Single-agent, multi-agent and hierarchical architectures can all be designed within one framework, and it supports token-level streaming. It can be used with LangSmith for debugging and evaluation.
It suits developers who need fine-grained control over an agent's flow; if you want to build something quickly, you can use the higher-level package Deep Agents instead.
What it is used for
- Build long-running agents that resume after a failure.
- Add a flow in which a person reviews and approves agent actions midway.
- Build agents that remember conversation history across sessions.
- Design single-agent, multi-agent and hierarchical workflows.
Pricing and license
MIT license, free
Getting started
- Install with pip install -U langgraph .
- Follow the official documentation to define the agent flow with state and nodes.
- If needed, connect LangSmith to trace and debug execution.
Links
Related terms and companies
Sources
Compiled from public sources (official pages, READMEs). Pricing and features change, so check the official site.