LangGraph

LangGraph is a Python-based library for building custom AI agents with graph-based state machine orchestration. It is described as a way to build reliable, multi-step agent workflows, especially where traditional linear chains are not enough because the task requires loops, branching, or manual approval.

The library models agents as state graphs. Each node is a computation step, and edges define control flow between steps. LangGraph supports cyclic graphs, conditional routing, persistent state across steps, and built-in checkpointing for pause-and-resume behavior.

A LangGraph application starts with a typed state dictionary that flows through the graph. The example state includes input, messages, output, and steps, and the text says this state carries user input, conversation history, final output, and a step counter for loop control. Nodes are Python functions that update the state, and routing functions can inspect the state to choose the next node. The guide also mentions external API calls such as search, calculator, and database actions, along with debugging through saved checkpoints.

LangGraph is installed with pip alongside langchain and langchain-openai, and the page states that Python 3.10+ is required. It runs entirely locally and does not need an external server. The guide recommends MemorySaver for checkpointing in examples and notes that SQLiteSaver is used for production to survive restarts.