LangGraph overview

LangGraph is LangChain's open-source framework for building AI agents as graphs: small Python functions that share one state dictionary and decide which step runs next.

Last updated: 29 Sep, 2026 · LangGraph 1.2 · Python 3.10+

It is developed by LangChain and released under the MIT licence. The docs describe it as a low-level orchestration framework for building and running long-running, stateful agents.

What LangGraph is for

LangGraph suits work that mixes fixed steps with decisions a model makes, and that needs pauses, retries or memory. LangChain provides chat models, message types and tools; LangGraph decides the order steps run in and keeps the state between them.

How it works

The LangGraph section of the video starts from the same three pieces: edges, nodes and state. Its example is a workflow that turns a YouTube video into a blog post: one step takes out the transcript, the next writes a title from it, and the last writes the content from both. The crash course builds it with LangGraph's Graph API, the style this course uses too.

A LangGraph program is a set of Python functions, called nodes, that pass one dictionary, called the state, between them. You write the functions and connect them with edges, which say which one runs after which; LangGraph runs them and carries the state along.

What you will build

A support agent that reads a message, decides what to do, looks things up with tools, remembers the conversation, and stops to ask before anything it cannot undo. Each lesson adds a few lines, runs them, and explains what changed. Where a lesson has a video clip, you run the video's example first, then the support agent's version.

Where you use LangGraph

• Multi-step assistants where a model decides the next action, not a fixed script.
• Flows that branch, retry, or run steps in parallel.
• Agents that pause for a human approval and resume later.
• Anything that needs memory across turns or a run you can inspect and replay.

Advantages and limits

• Advantage: the flow is explicit, so branches, loops and retries are visible and debuggable.
• Advantage: state, memory, human approval and streaming are built in.
• Advantage: uses LangChain's models and tools, so any provider LangChain integrates with is available.
• Limit: for a single model call with no branching, a plain function is simpler.
• Limit: you think in graphs, a small step over calling functions directly.

Prerequisites

• Python functions and dictionaries.
• The early lessons are plain Python. The model lessons call a real model, so you need an API key; the setup lesson shows how, starting with a one-line key check.
• Python 3.10 or newer.

Related

• Next: Installation and setup
• Reference: LangGraph documentation