Build AI agents with n8n: practical tutorial for SMBs

Years ago you needed a developer to build an AI agent. Not anymore in 2026: with low-code tools like n8n you can build working agents in a day. This tutorial gives you a working example: an agent that reads incoming emails, retrieves customer context from HubSpot, and writes a draft response.

What is n8n?

n8n is an open-source workflow automation tool. Comparable to Zapier or Make, but:

• Self-hosted option (data stays in your own environment)
• More powerful logic and code steps
• Hundreds of built-in integrations
• AI nodes for OpenAI, Anthropic, etc.

Ideal for SMBs who want more control than Zapier, but can't maintain a developer team.

What we're going to build

An agent that:

1. Trigger: new email in inbox info@company.nl
2. Reads subject and content
3. Searches sender in HubSpot CRM
4. Retrieves open deals and recent notes
5. Generates a draft response via OpenAI/Claude
6. Creates a draft in Outlook/Gmail for human sending

Step 1: Set up n8n

Option A: n8n cloud

• Go to n8n.cloud
• Create an account (€20-€50/month for SMB use)

Option B: Self-hosted

• Install Docker
• `docker run -it --rm -p 5678:5678 n8nio/n8n`
• Access via http://localhost:5678
• For production: deploy to a server (DigitalOcean, Hetzner) with SSL

Step 2: Add email trigger

In n8n:

1. Create new workflow
2. Trigger: "Gmail Trigger" or "IMAP Email Trigger"
3. Connect your email account (OAuth for Gmail, IMAP credentials for others)
4. Filter: for example only emails to info@

Step 3: HubSpot lookup

1. Add node: "HubSpot - Contact - Get Many"
2. Authenticate with API key from HubSpot
3. Filter by sender email address:

• Field: email
• Value: `{{ $json.from.value[0].address }}`

Step 4: Retrieve deals

1. Add node: "HubSpot - Deal - Get Many"
2. Filter by contact ID from previous step
3. Select relevant properties: deal name, deal stage, amount

Step 5: AI node for draft

1. Add node: "OpenAI" or "Anthropic Claude"
2. Model: GPT-4o-mini or Claude Haiku (cost/speed)
3. System prompt:

1. User prompt:

Step 6: Save draft in inbox

1. Add node: "Gmail - Draft - Create" (or Outlook variant)
2. To: `{{ $('Email').item.json.from.value[0].address }}`
3. Subject: "RE: {{ $('Email').item.json.subject }}"
4. Body: `{{ $('OpenAI').item.json.message.content }}`

Step 7: Test and activate

1. Run "Test workflow" on a test email
2. Check if the draft looks correct
3. Activate the workflow (toggle in top right)

What does this cost?

• n8n cloud: €20-€50/month
• OpenAI API: approximately €0.01-€0.05 per email (depending on model)
• HubSpot: existing license

Budget €60-€150/month for a setup that handles 500-2000 emails per month.

Next steps

First version working? Then improve it:

1. Add knowledge base: a database with FAQ answers the agent can draw from
2. Sentiment routing: angry emails go straight to humans, neutral ones via AI draft
3. Multi-language: detect language and respond in the same language
4. Categorization: automatically tag emails (sales, support, complaint)
5. Auto-send: for high confidence and simple questions, send directly instead of draft

Three pitfalls

1. Let AI send directly without review: always start with "create draft", only upgrade to auto-send once quality is proven
2. No logging: log every step, otherwise debugging is impossible
3. Forget privacy: add GDPR disclaimer to your privacy statement for automated processing

Conclusion

With n8n you build a working AI agent in a day. For SMBs, it's the golden middle ground between "building a simple Zapier" and "hiring a developer". Start with one agent, measure hard, and scale. The ROI is proven in dozens of SME implementations.