Skip to main content
Use LangGraph to orchestrate multi-step, stateful agent workflows. This example shows a minimal graph that takes an input message, adds an automated response node, and returns the result.

​
Prerequisites

  • Python 3.10+
  • Install dependencies:
pip install langgraph

​
Minimal example (Python)

main.py
from typing import TypedDict, List
from langgraph.graph import StateGraph, END

# Define the graph state structure
class ConversationState(TypedDict):
    messages: List[str]

# Node: add an automated response

def respond(state: ConversationState) -> ConversationState:
    user_last_message = state["messages"][-1] if state["messages"] else ""
    assistant_reply = f"You said: {user_last_message}. This is a response from the graph."
    return {"messages": state["messages"] + [assistant_reply]}

# Build the graph
builder = StateGraph(ConversationState)

builder.add_node("respond", respond)

# Start at the respond node and then end
builder.set_entry_point("respond")
builder.add_edge("respond", END)

# Compile into an executable app
app = builder.compile()

if __name__ == "__main__":
    initial_state: ConversationState = {"messages": ["Hello LangGraph"]}
    # Stream returns intermediate steps; iterate to observe progression
    for update in app.stream(initial_state):
        print(update)

​
Expected output

{'respond': {'messages': ['Hello LangGraph', 'You said: Hello LangGraph. This is a response from the graph.']}}

​
How it works

  • State: ConversationState holds a list of messages that persists through the graph.
  • Node: respond reads the latest message and appends an automated reply.
  • Edges: The flow enters respond and then immediately terminates (END).

​
Next steps

  • Add more nodes for tools, retrieval, or function calling.
  • Branch conditionally based on state values.
  • Explore persistence and memory patterns.
For advanced patterns, see the official LangGraph documentation: Build with LangGraph.