LangChain, Inc.

LangGraph

Build stateful, controllable agents as graphs.

LangGraph models agent workflows as a directed graph of nodes (functions) and edges (transitions) over a shared, typed state. It adds durable execution, time-travel, human-in-the-loop interrupts, and streaming — the runtime behind most production LangChain agents.

Install

bash

pip install langgraph

bash

npm install @langchain/langgraph

Quickstart

A minimal example to verify your setup.

python

from typing import TypedDict
from langgraph.graph import StateGraph, START, END

class State(TypedDict):
    input: str
    output: str

def respond(state: State) -> State:
    return {"output": f"Echo: {state['input']}"}

graph = StateGraph(State)
graph.add_node("respond", respond)
graph.add_edge(START, "respond")
graph.add_edge("respond", END)

app = graph.compile()
print(app.invoke({"input": "hello"}))

Core concepts

State graphs

Define a typed state schema, then declare nodes that read and update it. Conditional edges branch on state to express loops, retries, and routing.

Checkpointers

Plug in SQLite, Postgres, or Redis to persist state across steps. Resume crashed runs, fork from past steps, and audit every transition.

Human-in-the-loop

Interrupt the graph before any node to wait for approval, edits, or tool confirmation, then resume with updated state.

Streaming

Stream tokens, node outputs, and intermediate state updates to drive responsive UIs over WebSocket or SSE.

Common use cases

›Multi-agent collaboration
›Long-running research and coding agents
›Approval workflows with human review
›Reliable tool-using assistants

Resources

← All frameworks Next: LangSmith