LangGraph launch intelligence

Loop through web, news, and shopping evidence in a stateful graph.

What you’ll build

A stateful LangGraph research loop that alternates between an analyst model and specialized search nodes until it can produce a product-launch intelligence brief grounded in facts, reporting, and live marketplace signals.

Default brief: Investigate compact AI voice recorders. Reconcile official product facts, launches and reviews, current prices, disagreements, and a recommendation.

How the agent works

  1. Route from shared state. The analyst chooses a search tool from the current message state; the graph routes the call to a tool node.
  2. Collect three signal types. Web, news, and shopping searches return complementary product, coverage, and price evidence.
  3. Loop until sufficient. Tool results return to the analyst, which either searches again or writes the final intelligence memo.

Core agent setup

This excerpt highlights LangGraph’s explicit analyst-to-tools cycle. The full script also creates the model, validates keys, supplies the prompt, and saves the final message.

tools = [
    web_search(provider="langgraph", allowed_engines=["google_light", "bing"]),
    news_search(provider="langgraph"),
    shopping_search(provider="langgraph"),
]
model = ChatOpenAI(model=MODEL, api_key=api_key, temperature=0).bind_tools(tools)

def call_model(state: MessagesState) -> dict[str, list[Any]]:
    return {"messages": [model.invoke(state["messages"])]}

graph_builder = StateGraph(MessagesState)
graph_builder.add_node("research", call_model)
graph_builder.add_node("tools", ToolNode(tools))
graph_builder.add_edge(START, "research")
graph_builder.add_conditional_edges("research", tools_condition)
graph_builder.add_edge("tools", "research")
agent = graph_builder.compile()

View the complete main.py →

Run the recipe

Set SERPAPI_API_KEY and OPENAI_API_KEY in the repository-root .env, then run:

uv run --isolated --no-project --with 'serpapi-search-tools[langgraph]' --with python-dotenv --with langchain-openai \
  cookbook/langgraph/main.py

Optional controls: OPENAI_MODEL, COOKBOOK_PROMPT, and COOKBOOK_OUTPUT_DIR.

Visible control flow: The graph has explicit research and tools nodes. Use it when you want the tool-routing loop to be inspectable rather than hidden inside a high-level agent.

Inspect the result

The recipe writes cookbook-output/langgraph-launch-intelligence.md. Expect product facts, recent launch evidence, live price signals, disagreements between sources, a recommendation, and direct URLs.

The final assistant message is both printed and saved as the artifact.

SerpApi tools used

web_search finds official product facts, news_search tracks launches and reviews, and shopping_search returns current price and availability signals. These distinct tools give the graph clear routes for each open question before the evidence returns to the analyst node.

Framework pattern: The explicit state and tool loop follows LangGraph’s Agentic RAG notebook and routes across three SerpApi search tools.