LangGraph

LangGraph is a library for building stateful, multi-actor applications with LLMs, used to create agent and multi-agent workflows. It extends the LangChain ecosystem with graph-based orchestration primitives.

Overview

LangGraph models agent workflows as directed graphs where:

Nodes represent computation steps (LLM calls, tool invocations, or custom logic)
Edges represent transitions between steps (conditional or unconditional)
State is passed between nodes and persisted across interactions

This graph-based approach gives you fine-grained control over agent behavior, including cycles, branching, and human-in-the-loop patterns.

Key Features

Stateful Graphs: First-class support for persistent state across graph executions
Cycles and Branching: Model complex agent reasoning with loops and conditional edges
Human-in-the-Loop: Pause execution for human review and resume from checkpoints
Streaming: Stream node outputs, LLM tokens, and intermediate state updates
Persistence: Built-in checkpointing for fault tolerance and conversation memory
Multi-Agent: Compose multiple agents as subgraphs within a parent graph

Installation

pip install langgraph

Quick Start

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

class AgentState(TypedDict):
    messages: list
    next_step: str

def call_model(state: AgentState) -> AgentState:
    # LLM call logic
    return {"messages": state["messages"] + [response]}

def should_continue(state: AgentState) -> str:
    if state["next_step"] == "end":
        return END
    return "call_model"

# Build the graph
graph = StateGraph(AgentState)
graph.add_node("call_model", call_model)
graph.add_conditional_edges("call_model", should_continue)
graph.set_entry_point("call_model")

app = graph.compile()
result = app.invoke({"messages": ["Hello"], "next_step": "continue"})

Architecture Patterns

ReAct Agent

The most common pattern: the agent reasons about what tool to call, executes it, and loops until done.

Plan-and-Execute

The agent first creates a plan, then executes each step, revising the plan as needed.

Multi-Agent Collaboration

Multiple specialized agents coordinate through a supervisor agent that routes tasks.

Documentation

License

MIT License