Run it as a server

Everything so far ran in a script. To put a graph behind a web page, something has to keep it running and answer requests, and LangGraph ships that something.

Put the graph in a file

Nothing about the graph changes. It only has to be reachable by name, so the last line assigns the compiled graph to a variable the server can find.

# assistant.py
builder = StateGraph(State)
builder.add_node("retrieve", retrieve)
builder.add_node("answer", answer)
builder.add_node("say_no", say_no)
builder.add_edge(START, "retrieve")
builder.add_conditional_edges("retrieve", did_we_find_anything, ["answer", "say_no"])

graph = builder.compile()

Tell the server where to look

One small file beside it, listing your graphs by name and where each one lives.

{
  "dependencies": ["."],
  "graphs": {
    "assistant": "./assistant.py:graph"
  },
  "env": ".env"
}

The value is the path to the file, a colon, and the name of the variable. That is why the last line of assistant.py had to be a plain assignment. Call this file langgraph.json.

Start it

pip install "langgraph-cli[inmem]"
langgraph dev

That gives you a local server with your graph behind it. You get a URL to call from your own code, and a browser view where you can run the graph, watch each node as it goes, and read the state at every step.

What you get for free

Three things arrive without any work, and each one is something you built by hand earlier in this course.

This is the payoff for learning those as ideas rather than as features. The server is not doing anything you have not already done. It is doing it over HTTP.

Calling it

From another program, the client library talks to that URL. The shape will look familiar, because it is the same invoke and the same config.

from langgraph_sdk import get_sync_client

client = get_sync_client(url="http://localhost:2024")
thread = client.threads.create()

for chunk in client.runs.stream(thread["thread_id"], "assistant",
                                input={"question": "How do I reset my password?"}):
    print(chunk.data)