A Framework that extends LangChain and orchestrates long-lived, stateful LLM agents. An agent (and the overall system) is represented as a graph of nodes (LLMs, tools, sub-agents) and transitions between them, similar to finite-state machines. This makes it possible to build complex, branching, and cyclic workflows; persist and restore state; run multiple LLM agents with different roles in various interaction patterns (planner–executor, expert–critic, etc.); and reliably deploy agents to production. State persistence allows you to keep conversation or process progress across requests, recover after failures, and scale the system. LangGraph is a strong fit when you need long sessions, multi-step business processes, or complex transition rules.