Chat router for company, market and social media research

Chat router for company, market and social media research

An n8n chat workflow that decides which kind of research a question needs and hands it on. You type a question in n8n's
chat. A GPT-4.1 agent reads it and calls one of three tools: company research, market research or social media research.
Each tool is another n8n workflow of your own, and its answer comes back to the chat with headings and Markdown. The agent
remembers the last few turns of the conversation.

The three research workflows are not part of this file. It is the front door only: you connect it to three workflows you
already have, or build them first (see "The three research workflows" below).

What is inside

• workflow.json: the workflow as n8n exports it, seven nodes with no sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• Company research, Market research and Social media research pointed at three workflows by their IDs in the author's
  own n8n. Those IDs mean nothing in yours, so each tool's workflow choice is now empty for you to fill.
• Nothing else changed. The original copy needed no backslash repair.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key that can use gpt-4.1.
• Three research workflows in the same n8n, or the time to build them.

The three research workflows

Each tool sends the agent's request as one text field named Query, so each research workflow needs:

1. A When Executed by Another Workflow trigger with one input, Query, as a string.
2. Whatever research you want behind it: a search API, a scraper, an agent of its own.
3. A last node whose output is the answer you want the chat to show.

The router's prompt describes what each one should cover. Company research reads a company's site and review sites such as
Trustpilot, G2 and Capterra. Market research looks at trends, competitors and openings. Social media research finds or sums
up LinkedIn and Instagram profiles.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential and select it in OpenAI Chat Model.
3. Open Company research and pick your company research workflow from the list. Do the same in Market research and
   Social media research. Each list shows the workflows in your n8n.
4. In each of the three, check that Query appears under the workflow inputs. If it does not, the chosen workflow's trigger
   lacks a Query input.
5. Give each tool a short description of what it does. The agent reads it when choosing.
6. Save, press the Chat button at the bottom of the editor and ask a question of each kind.
7. Press Publish when the answers look right.

How a run goes

1. When chat message received takes your message from n8n's chat.
2. AI Agent reads it with Simple Memory, which keeps the last five exchanges of that chat.
3. The agent follows the routing rules in its system message: reviews, ratings and comparisons go to company research,
   trends and competitors to market research, and social media only when you name a platform. If your question is unclear,
   it asks you one question first.
4. The chosen tool runs your workflow with the request as Query and hands its output back.
5. The agent answers in the chat with that tool's result, laid out with headings, Markdown and some emojis.

Change the look

• Routing: the rules and examples in the system message of AI Agent. Add a fourth tool by copying a tool node,
  connecting it to the agent and adding a line for it to the rules.
• Tone and layout: the last line of the system message asks for headings, Markdown and emojis.
• Memory: Context Window Length in Simple Memory, five exchanges by default. A smaller window costs fewer tokens.
• Who can chat: by default only you, from the editor. Make Chat Publicly Available in When chat message received gives the
  chat its own web address.

Good to know

• Tool names: the system message calls the tools CompanyResearch, MarketResearch and SocialMedia, while the nodes name
  them Company_research, Market_research and Social_media. The model usually links the two. If it picks the wrong one,
  make the names match.
• One tool a question: the prompt tells the agent to use one tool unless you ask for more. A question that needs two kinds of
  research gets one unless you say so.
• Billing: OpenAI bills each turn of the agent, plus whatever your research workflows spend.
• GPT-4.1: OpenAI's deprecations page, read on 2 November 2026, removes gpt-4.1-2025-04-14, the model behind gpt-4.1, on
  14 January 2027 and recommends gpt-6-luna. Pick a current model in OpenAI Chat Model before then.
• A public chat: once it is public, anyone with the address can run your research workflows on your keys.
• Your data: your questions and the research results go to OpenAI. What else leaves your n8n depends on the research
  workflows you connect.