Exercises 08 · AI Support Agent

Start from workflow/starter.json (Chat Trigger + lookup_order) or your finished lesson workflow.

Exercise 1: Orders from a real data source

Task: Replace the demo orders object with a Google Sheet (order_id, email, status, items, carrier, tracking, estimated_delivery).

Done when: adding a row to the sheet makes the agent find that new order.

Hint

Use the Google Sheets Tool (the tool version of the node) with a filter on order_id, and let the model fill the value with $fromAI('order_id'). Or call a sub-workflow with the Call n8n Workflow Tool.

Exercise 2: Real tickets

Task: Make create_support_ticket actually save the ticket (Google Sheet row) and notify you (email/Slack/Telegram).

Done when: asking for a human creates a row and sends you a message with the summary.

Hint

Build a sub-workflow (inputs: summary, customer_contact, priority) that appends + notifies + returns the ticket ID, then connect it with Call n8n Workflow Tool. Remember: publish the sub-workflow (Module 04).

Exercise 3: Embed it on a website

Task: Put the chat on a web page.

Done when: a local index.html shows the chat bubble and conversations work.

Hint

Switch the Chat Trigger mode to Embedded Chat, add your page's origin to Allowed Origins (CORS), and use the @n8n/chat package's createChat({ webhookUrl }) snippet.

Stretch: Evaluate it

Write 10 realistic customer messages with expected facts, run them through the chat API in a loop (like test/test.sh), and track how many pass when you change the prompt or model.