LangGraph is a low-level orchestration framework for building long-running, stateful agents and workflows. Rather than a high-level agent abstraction, it provides the underlying infrastructure to define how state moves through an application, drawing on ideas from Pregel and Apache Beam with a graph-style interface inspired by NetworkX.
Its core capabilities target the hard parts of durable agent systems: durable execution that persists through failures and resumes from where it left off, human-in-the-loop points where agent state can be inspected and modified mid-run, and both short-term working memory and long-term persistent memory across sessions. It can be used standalone or paired with the wider LangChain ecosystem — LangSmith for debugging/observability and deployment, and higher-level packages like Deep Agents (planning, subagents, file systems) built on top of it. Python and JavaScript/TypeScript implementations are available.
