LangGraph
Stateful agent orchestration with graphs
What it is
LangChain extension that models agent workflows as directed graphs. Each node is a function, each edge a conditional transition. Enables loops, branching, and persistent state across calls — essential for agents that need to think multiple steps, retry, and maintain memory across sessions. Used in production by major AI companies.
How to use it, step by step
- 1
Install: pip install langgraph (requires LangChain).
- 2
Define a 'State' (usually a typed dict) representing the data flowing through the graph.
- 3
Create a StateGraph and add nodes: each node is a function that takes the state and returns an updated version.
- 4
Connect nodes with add_edge and add_conditional_edges to define branching and loops; set the entry point and END.
- 5
Compile the graph with .compile() and invoke it with an initial state; add a checkpointer for persistent memory.
💡Practical tips
- →Keep state immutable: each node returns a new dict, never mutates the existing one.
- →Use conditional edges to let the agent decide its next step, including looping back.
- →The checkpointer lets you pause and resume an agent: handy for human approvals mid-flow.
💰Pricing
Open-source and free. LangGraph Platform (managed hosting) has optional paid plans.
The official site opens in a new tab.