A first agent should be small. Here is a safe pattern: an assistant that answers questions from a price list.
Step 1: Define the job
Write one sentence: “Answer customer questions about product prices using our sheet, and say when unsure.”
Step 2: Add a trigger
Use the Chat Trigger to test in the editor. Later you can connect messaging apps.
Step 3: Add the AI Agent node
Connect a chat model through the model input. Set credentials for your chosen provider (see using AI models in n8n).
Step 4: Write a clear system prompt
State the role, the tone, what it may and may not do, and what to say when unsure.
Step 5: Give it one tool
Add a tool that reads your sheet (agent tools). Describe the tool clearly so the agent knows when to use it.
Step 6: Test with hard questions
Try missing items, odd spellings and off-topic questions. Check the execution log to see which tool the agent used.
Step 7: Add limits
Limit tools, set a maximum number of steps and keep a person in the loop for risky actions.
Next
Add memory and cost control.
Quick answers
Which nodes do I need?
A trigger, the AI Agent node, a chat model and one or more tools.
How do I test an agent?
Ask hard and off-topic questions and read the execution log.
Can it run without a paid model?
You need a model provider; some offer free or low-cost tiers.
Related reading
- What Is an AI Agent in n8n and How Is It Different from a Workflow?
- n8n AI Agent Memory Explained: How an Agent Remembers a Conversation
- n8n AI Agent Tools: How to Give an Agent Safe Abilities
Want to learn n8n with a trainer, not just videos?
Our 15-day live n8n course takes you from the editor basics to APIs, messaging, AI agents and a final project, in Hindi and English. Fees start at ₹4,999 (final pricing may vary by format).
See the 15-day n8n training | Find your city | Book free counselling