Overview
LangGraph is a low-level orchestration framework from the LangChain team for building, managing, and deploying long-running, stateful agents. Instead of hiding control flow, it lets you define agent workflows explicitly so you decide how steps run, how state is passed along, and where the agent can pause.
It is aimed at developers who need more control than a high-level agent wrapper gives them. You handle persistence, memory, and human review directly, which suits workflows that can fail, run for a long time, or need a person to check a result before continuing. Companies including Klarna, Replit, and Elastic use it in production.
As a general agent framework, LangGraph can be used on its own or paired with the wider LangChain ecosystem. It works with LangChain for model integrations, LangSmith for debugging and observability, and Deep Agents, a higher-level package built on top of it for planning and subagents.
What it does
- Durable execution: agents persist through failures and resume from exactly where they left off
- Human-in-the-loop: inspect and modify agent state at any point during a run
- Short-term working memory plus long-term memory that persists across sessions
- Debugging and tracing through LangSmith, including execution paths and state transitions
- Production deployment for long-running, stateful workflows
- JS/TS support via LangGraph.js for the same patterns in JavaScript
Getting started
LangGraph is a Python package installed from PyPI. Install it, then follow the official quickstart to build your first graph.
Install LangGraph
Install the latest release from PyPI with pip.
pip install -U langgraphFollow the quickstart
The README does not include inline starter code. Use the official LangGraph Quickstart to build a first graph, and see the Python docs for memory, interrupts, and durable execution. For higher-level agents that plan and use subagents, see the Deep Agents package built on LangGraph.
Commands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Building long-running agents that must survive failures and resume from the last completed step
- Adding human review checkpoints so a person can inspect or edit agent state before it continues
- Giving an agent both short-term reasoning memory and long-term memory across sessions
- Orchestrating multi-step agent workflows where you need explicit control over branching and state
How LangGraph compares
LangGraph alongside other open-source agent frameworks & builders tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| AutoGPT | ★ 185k | One of the earliest autonomous agent projects, now a platform for building and running agents from reusable blocks and workflows. |
| Agno | ★ 40.8k | A fast Python framework (formerly Phidata) for building agents with memory, tools, and multimodal inputs, plus a runtime for deploying them in production. |
| AgentGPT | ★ 36.2k | AgentGPT lets you name a custom AI, give it a goal, and watch it plan tasks, run them, and learn from the results, all from a web browser. |
| LangGraph | ★ 35.2k | Low-level orchestration framework for building stateful, long-running AI agents |
| Composio | ★ 28.9k | Composio is an open-source SDK for Python and TypeScript that gives AI agents ready-made tools to act on real apps and APIs across many agent frameworks. |
| Semantic Kernel | ★ 28.2k | Microsoft's SDK for adding agents, plugins, and planning to apps across .NET, Python, and Java. |
| smolagents | ★ 27.9k | A minimal agent library from Hugging Face where the model writes and runs Python code to call tools and complete tasks. |
| Mastra | ★ 25.3k | A TypeScript framework for building AI agents and applications with workflows, RAG, memory, and observability built in. |