LangChain deep market research

Create a source-backed market brief with web and news search.

What you’ll build

A LangChain agent that produces a source-backed market brief instead of a one-shot answer. It plans the investigation, separates durable sources from recent reporting, compares independent evidence, and identifies unanswered questions.

Default brief: Assess commercial battery recycling in the United States. Finish with market signals, risks, open questions, and source URLs after comparing at least three independent sources.

How the agent works

  1. Plan the questions. The system prompt asks for a short research plan before the first search.
  2. Split durable and current. Web search gathers background evidence; news search tracks recent market developments.
  3. Compare before concluding. The agent reconciles independent sources and marks risks or missing evidence in the final brief.

Core agent setup

This excerpt shows the LangChain agent’s research contract and tool set. The full script also configures the OpenAI-compatible model, environment, and Markdown report.

agent = create_agent(
    model=model,
    tools=[
        web_search(
            provider="langchain",
            allowed_engines=["google_light", "bing"],
        ),
        news_search(provider="langchain"),
    ],
    system_prompt=(
        "Make a short plan before searching, distinguish current reporting "
        "from durable background sources, and never present unsupported claims as fact."
    ),
)
result = agent.invoke({"messages": [{"role": "user", "content": PROMPT}]})

View the complete main.py →

Run the recipe

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

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

Optional controls: XAI_MODEL, XAI_BASE_URL, COOKBOOK_PROMPT, and COOKBOOK_OUTPUT_DIR.

OpenAI-compatible model: The recipe uses LangChain’s ChatOpenAI client with xAI by default. Change the model or base URL through the environment without touching the search tools.

Inspect the result

The recipe writes cookbook-output/langchain-market-research.md. Expect a decision-ready brief with durable context, recent signals, source comparisons, market risks, unanswered questions, and direct URLs.

Set COOKBOOK_PROMPT to reuse the same research discipline for another market.

SerpApi tools used

web_search gathers durable market and company sources, while news_search finds recent developments. Their separate schemas help the agent plan each search against the right evidence type and compare sources before writing the brief.

Framework pattern: The research workflow follows LangChain’s Deep Agents from scratch guide and uses SerpApi web and news tools.