What is LangGraph?
LangGraph is a state-machine runtime for building agents with branching, cycles, persistence, and human-in-the-loop. This article helps you decide whether LangGraph fits your use case, explains the core model, and maps the 6-chapter learning path.
Quick Reference
- →LangGraph models agent logic as a state machine: nodes do work, edges route, state carries context
- →Two APIs share one runtime: StateGraph (declarative graph) and Functional API (@entrypoint/@task)
- →compile() validates the graph and returns a CompiledGraph — invoke() runs it, stream() streams it
- →recursion_limit (default 25) caps super-steps — set it explicitly in production
- →MessageGraph is deprecated since v1.0.0 — use StateGraph with a messages key instead
- →LangGraph adds real overhead — skip it for linear prompt-to-response tasks
The One-Sentence Definition
LangGraph is a state-machine runtime. Your code defines nodes (functions that do work) and edges (rules that route between them). LangGraph handles the execution loop, state persistence, streaming, and interrupts.
If you've used Temporal, AWS Step Functions, or any workflow engine, the mental model is similar. The difference: LangGraph is designed specifically for agents, where the number of steps isn't known upfront and the same node can fire many times in one run.
LangGraph reached v1.0 in October 2025 and is now at v1.1.6. The API is stable. This is a production-grade tool, not an experiment.