Chat research report from SerpAPI, Jina pages and Wikipedia

Chat research report from SerpAPI, Jina pages and Wikipedia

An n8n workflow that turns a chat question into a research report. A model proposes up to four search queries, SerpAPI runs
them on Google, and Jina's Reader opens the result pages. A second model keeps what matters, and a third writes the report in
Markdown with a Wikipedia tool at hand.

What is inside

• workflow.json: the workflow as n8n exports it, seventeen nodes: fourteen that do the work and three sticky notes. It
  holds no credential.
• README.md: these notes.

What we changed from the original copy

• The credential links are gone: OpenRouter on the model node and a Header Auth credential on the Jina request. Select your
  own.
• The model was google/gemini-2.0-flash-001. OpenRouter's public model list, read on 12 November 2026, no longer shows it
  and does list google/gemini-3.8-flash, which the node uses now.
• The SerpAPI request put the key in a query parameter whose value read $credentials.SerpAPI.key. It now selects n8n's
  SerpAPI credential type, which adds the key as api_key on its own.
• The report node was called Generate Comprehensive Research Report. It is Generate Research Report now.
• The three setup notes are rewritten as short notes without their links.
• Nothing else.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenRouter API key.
• A SerpAPI API key.
• A Jina API key for a Header Auth credential. Jina's Reader page and its repository, read on 12 November 2026, say a key is
  optional and gives a higher rate limit.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenRouter credential and select it in LLM Response Provider (OpenRouter).
3. Add a SerpAPI credential and select it in Perform SerpAPI Search Request.
4. Add a Header Auth credential with your Jina key. Use the header name from Jina's own documentation. Select the credential in
   Perform Jina AI Analysis Request.
5. Open the chat from Chat Message Trigger and ask a research question.

How a run goes

1. Chat Message Trigger takes the question.
2. Generate Search Queries using LLM asks the model for up to four search queries as a JSON list.
3. Parse and Chunk JSON Data removes any code fence, parses the list and splits it into four chunks.
4. Split Data for SerpAPI Batching runs Perform SerpAPI Search Request once for each chunk.
5. Format SerpAPI Organic Results turns each organic result into a title, an address and a source.
6. Split Data for Jina AI Batching sends each address to Perform Jina AI Analysis Request, which reads the page.
7. Extract Relevant Context via LLM keeps what bears on the queries.
8. Generate Research Report writes a Markdown report. It can call Fetch Wikipedia Information, and two
   Window Buffer Memory nodes keep the last twenty messages for the two agents.

Change the look

• The model for all three steps: LLM Response Provider (OpenRouter).
• How many queries: the system message of Generate Search Queries using LLM.
• The shape of the report: the system message of Generate Research Report.
• The search settings: the query parameters of Perform SerpAPI Search Request.

Good to know

• Format SerpAPI Organic Results reads $input.first(), so only the first search's results are formatted, and the results of the
  other queries go unused.
• The report agent's system message speaks of the original question, and its input holds only the extracted contexts.
• The report's input is labelled merged, and no node merges the contexts. Each agent runs once for each item it receives.
• Both memory nodes use the session key my_test_session, so two questions in a row share a history.
• A run calls the model, SerpAPI and Jina several times, so a long answer takes a while.
• Read each site's rules before you fetch its pages in bulk.
• The question goes to OpenRouter and its model provider, the queries go to SerpAPI and the page addresses go to Jina.
• The JSON holds no credential.