Website support chatbot with FAQ answers and email handoff

Website support chatbot with FAQ answers and email handoff

An n8n workflow. A chat window on your website talks to an AI agent on gpt-4o. It answers questions from your FAQ through
a Pinecone assistant, remembers each conversation in PostgreSQL, and emails your support team through Gmail when a
visitor needs a person.

What is inside

• workflow.json: the workflow as n8n exports it, six nodes. It holds no credential. The original copy left the
  Pinecone address, its headers and the support address empty. Here they are placeholders: the assistant's address,
  YOUR_PINECONE_API_KEY, the content type, Pinecone's API version 2026-07 and YOUR_SUPPORT_EMAIL.
• README.md: these notes.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• A Pinecone account with an assistant, and your FAQ uploaded to it as files.
• A PostgreSQL database n8n can reach, for the chat history.
• A Gmail account to send the handoff emails from, and the address they go to.
• A website where you can add a script, for the chat window.

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. Add a Postgres credential with your database's host, port, database name, user and password, and select it in
   Postgres Chat Memory. It writes to a table named n8n_chat_histories.
4. In Pinecone, create an assistant and upload your FAQ, price list or policies to it.
5. Open Rag Questions. In its URL, put your assistant's name in place of YOUR_ASSISTANT_NAME, and in the Api-Key
   header put your Pinecone API key in place of YOUR_PINECONE_API_KEY.
6. Add a Gmail OAuth2 credential and select it in Send Email. Replace YOUR_SUPPORT_EMAIL with your team's address.
7. Save and press Publish, then copy the production URL of When chat message received.
8. Add n8n's chat window to your site with that URL (below).

The Pinecone assistant

Rag Questions sends the visitor's question to a Pinecone assistant and passes its answer to the agent. The body is
Pinecone's standard chat request: one user message, no streaming, and gpt-4o as the model the assistant answers with.
Pinecone's documentation, read on 2 November 2026, gives the address as
https://prod-1-data.ke.pinecone.io/assistant/chat/<assistant name>, with the Api-Key, Content-Type and
X-Pinecone-Api-Version headers this node sends. If your assistant's page in the Pinecone console shows another host,
use that one.

Any other service that answers a POST with a question in it can stand in. Change the URL, the headers and the body
together.

The chat window

When chat message received is set to Embedded Chat and open to the public, so it takes messages from a chat window on
your site. n8n's own window is the @n8n/chat package. Load it on your page and give it the trigger's production URL
as its webhookUrl, as the package's page on npm shows. Its messages arrive with chatInput and sessionId. The agent
also reads message and the memory also reads chatId, for a widget of your own that sends those names instead.

How a conversation runs

| Node | What it does |
|---|---|
| When chat message received | Takes each message from the chat window |
| AI Agent | Reads the message with a short system message: answer from the FAQ, hand off by email |
| OpenAI Chat Model | gpt-4o, the agent's model |
| Postgres Chat Memory | Keeps the last five exchanges of each session, keyed on the session and chat IDs |
| Rag Questions | Asks the Pinecone assistant a question the agent writes |
| Send Email | Sends your team an email with the subject New Support request and a message the agent writes |

The system message asks the agent to put the visitor's name and email address in the handoff, so it asks for them
first. The email goes to your team only: the address is fixed, and nothing writes to the visitor.

Change the result

• What the bot says and when it hands off: the system message of AI Agent. It is two short rules, so this is the
  first place to add your tone, your opening hours or when to hand off.
• What it knows: the files in your Pinecone assistant.
• How much it remembers: Context Window Length in Postgres Chat Memory, five exchanges unless you set it.
• The handoff email: the subject in Send Email, and the address it goes to.
• Which sites may embed it: Allowed Origins (CORS) in the trigger's options, which allows any site until you set it.

Good to know

• Open chat: anyone who finds the trigger's URL can send it messages, and each one is an OpenAI call on your key.
  Setting Allowed Origins to your own domain stops other sites from embedding it.
• Handoff spam: a visitor can ask for a handoff as often as they like. Every one lands in your team's inbox.
• Two bills: OpenAI bills the agent's gpt-4o calls, and Pinecone bills its assistant for each FAQ question.
• Where the data goes: each message goes to OpenAI, FAQ questions go to Pinecone, and every conversation is stored in
  your PostgreSQL table, with any name or address the visitor gives.
• The email is sent as HTML with no n8n line at the end.