LangGraph — Build Stateful AI Agents as Graphs

What it is

LangGraph is the LangChain framework for building AI agents as stateful directed graphs rather than linear chains. Each node represents a computation (LLM call, tool use, decision), and edges define how state flows between them. The graph abstraction supports cycles (retry loops), conditional branching, parallel execution, and human-in-the-loop gates.

LangGraph targets developers building production-grade AI agents that need persistence, observability, and failure recovery. It is the recommended approach for agents within the LangChain ecosystem.

How it saves time or tokens

Linear chain architectures break down when agents need to loop, branch, or wait for human input. LangGraph provides these patterns as first-class primitives. State is automatically checkpointed, so crashed agents resume from the last successful node instead of restarting entirely.

Streaming support lets your UI show progress in real time as the agent works through graph nodes, reducing perceived latency.

How to use

1. Install LangGraph:

pip install langgraph

1. Define state and build a graph:

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

class State(TypedDict):
    messages: List[str]

def chatbot(state: State):
    return {'messages': [llm.invoke(state['messages'])]}

graph = StateGraph(State)
graph.add_node('chatbot', chatbot)
graph.add_edge(START, 'chatbot')
graph.add_edge('chatbot', END)
app = graph.compile()

1. Invoke the agent:

result = app.invoke({'messages': ['Hello']})

Example

# Multi-agent supervisor pattern
from langgraph.graph import StateGraph, START, END

def supervisor(state):
    # Decide which agent to call next
    if state['task'] == 'research':
        return 'researcher'
    return 'coder'

graph = StateGraph(State)
graph.add_node('researcher', research_fn)
graph.add_node('coder', code_fn)
graph.add_node('reviewer', review_fn)
graph.add_conditional_edges(START, supervisor, {
    'researcher': 'researcher',
    'coder': 'coder'
})
graph.add_edge('researcher', 'reviewer')
graph.add_edge('coder', 'reviewer')
graph.add_edge('reviewer', END)
app = graph.compile()

Related on TokRepo

• Multi-Agent Frameworks -- Compare LangGraph with CrewAI, AutoGen, and others
• AI Tools for Agents -- Browse agent development frameworks

Common pitfalls

• Every graph needs a path from START to END. Missing edges cause the graph to hang without error messages in some versions.
• Human-in-the-loop (interrupt_before/after) requires a persistence backend. Without persistence, interrupted state is lost.
• Streaming requires the caller to iterate over the stream object. Forgetting to consume the stream means no output appears.