LangGraph is an orchestration framework and runtime for building long-running, stateful agents, developed by LangChain Inc. Its central idea is to model an agent’s control flow as a graph: nodes are units of work - a model call, a tool call, a function - and edges define how control and a shared state object move between them. Unlike a simple chain that runs once from start to finish, a graph can contain cycles, which is what lets an agent loop, retry, and revise until a task is done.
The state passes through the graph as nodes read and update it, which makes branching and parallel paths natural to express. LangGraph emphasizes durable execution: agents can persist through failures and resume from where they stopped, run for extended periods, and pause for human review. Its human-in-the-loop support lets a person inspect and modify the agent’s state at any point, and its persistence layer is what enables both long-running tasks and the ability to rewind. The design takes inspiration from graph-processing systems like Pregel and Apache Beam, with an interface influenced by the NetworkX graph library. Although built by the makers of LangChain, it runs independently and does not require LangChain.
LangGraph is one of the most widely adopted ways to build production agents, including multi-agent setups where several agents are nodes that hand off work through shared state. It is frequently compared with CrewAI and AutoGen as the leading framework for moving agents from a prototype script to a reliable, resumable system.