LangGraph Documentation

What LangGraph is

LangGraph describes itself as a low-level orchestration framework and runtime for building, managing and deploying long-running, stateful agents. You describe an agent as a graph: nodes are functions that read and update a shared state, and edges decide what runs next, including conditional edges that loop back to tools while the model keeps calling them.

The docs, for both Python and JavaScript, cover:

• Quickstart: a calculator agent built with the Graph API, from state definition to compile() and invoke().
• Graph API and Functional API: two ways to express the same workflows.
• Persistence: checkpointers that save state after every step under a thread id.
• Interrupts: interrupt() pauses a graph for human input, and a resume command continues it later.
• Memory: short-term memory in the thread and long-term memory in a store.
• Streaming, subgraphs, time travel and durable execution.
• Deployment and observability, with LangSmith for tracing.

Is it worth learning, and when?

Learn it after you have written a tool loop by hand. LangGraph earns its place when control flow matters: approval steps, retries, branches, long-running tasks and runs that must survive restarts. The docs are clear that you do not need LangChain to use LangGraph, and that LangChain's own agent abstractions are built on top of it. They also note users including Klarna, Uber and J.P. Morgan.

If your agent is a short loop with a few tools and no pauses, a lighter SDK may be simpler. If you need a workflow a reviewer can read as a diagram, with state you can inspect at every step, LangGraph is a strong choice.

Prerequisites and time

Solid Python or TypeScript, a working knowledge of tool calling, and ideally the free Introduction to LangGraph course from LangChain Academy. Expect a week or two of building to feel productive, and more to design good state and memory.

Free or paid?

The library is free and open source under the MIT licence. LangSmith, the companion tracing and evaluation platform, has a free tier and paid plans; you can use LangGraph without it.

What you can do after it

Build agents with explicit state, persistence, human approval and streaming, explain the control flow as a graph, and resume paused or crashed runs from a checkpoint.

Common questions

Many tutorials look different. Why? LangGraph reached version 1 in 2025 and some imports and prebuilt helpers moved. Check the version on any tutorial and prefer the official docs.

Python or JavaScript? Both are maintained and follow the same concepts; the Python library is older and has more examples.