Building a Customer Support Chatbot

What You Are Building

A customer support chatbot that holds multi-turn conversations, remembers what was discussed earlier in a session, and can look up answers from a knowledge base. You will use the n8n Chat Trigger, an AI Agent with Window Buffer Memory, and a search or FAQ tool — all without writing frontend code. This is a complete, production-capable pattern.

Step 1: Chat Trigger and System Prompt

Start with an n8n Chat Trigger node. This provides a ready-made chat UI at a webhook URL you can share directly with users. Add an AI Agent node after it. Write a system prompt that defines the support persona and constraints:

"You are a support assistant for Acme Software. Answer questions about our products clearly and helpfully. If you cannot find the answer, say so honestly and direct the customer to support@acmesoftware.com. Do not invent information or guess at product details."

The system prompt sets the tone, limits scope, and prevents the agent from hallucinating answers when it is uncertain.

Step 2: Connect the Chat Model

Add a Chat Model sub-node. For a support chatbot, a mid-tier model like gpt-4o-mini or claude-haiku balances cost and quality well. Set temperature to 0.2 for consistent, factual responses. Avoid high temperature settings in support contexts — they produce varied output where users expect predictability.

Step 3: Add Window Buffer Memory

Add a Window Buffer Memory sub-node and set the Context Window Length to 10. This means the agent holds the last 10 conversation turns in memory — enough context for most support interactions. Users can ask follow-up questions, reference earlier issues, and receive coherent multi-message responses. If you run n8n in queue mode, use Postgres Chat Memory instead.

Step 4: Add a Knowledge Base Tool

The agent needs access to your support documentation. Two approaches:

• HTTP Request Tool — configured to fetch your FAQ page or documentation URL. Practical for simple setups where your docs are publicly accessible.
• Workflow Tool pointing to a RAG sub-workflow — the proper approach for larger knowledge bases. Covered in lessons 8–10 of this track.

For this lesson, add an HTTP Request Tool configured to fetch your FAQ. Write a description: "Fetch answers from the Acme Software FAQ. Use this when a customer asks a product question."

Step 5: Test and Calibrate

Activate the workflow and open the Chat Trigger URL. Test with three scenarios:

• A question clearly in the knowledge base — the agent should find and cite the answer.
• A follow-up question referencing a previous message — the agent should maintain context.
• A question outside your knowledge base — the agent should acknowledge the gap, not guess.

Open the execution log after each test to read the agent's reasoning trace and adjust the system prompt based on what you observe.

Adding a Human Escalation Path

Add a conditional after the AI Agent: if the response contains phrases like "I cannot find" or "please contact support", route to a Gmail or Slack node to notify your team. This turns the chatbot into a triage layer — it handles what it can, and immediately escalates what it cannot. Most production support chatbots use exactly this pattern.