LangGraph

LangGraph is an open source framework from LangChain for building long-running, stateful agents. You describe the workflow as a graph of state, nodes and edges, and the runtime carries out the steps in that structure.

Each node in a graph is a step, which can be a model call, a tool call or plain code. Edges decide the order and branching, and shared state keeps data consistent between steps. The runtime adds checkpoints, streaming and human-in-the-loop pauses, so a run can resume after an interruption.

• Graph of state, nodes and edges
• Checkpoints for durable runs
• Streaming of progress
• Human approval pauses
• Works with or without LangChain

Developers who need explicit control over agent flow and state. Teams that want a higher-level agent loop may prefer LangChain's agent tools or Deep Agents.

LangGraph is released under the MIT license and installs with pip or uv. It runs on your own infrastructure, so hosting and model costs are up to you. Learning the graph model takes some time, but the docs walk through the building blocks step by step.