How to Build an AI Inbox Triage Agent with n8n (Step by Step)

The short answer

An AI inbox triage agent is a workflow that watches your inbox, asks an LLM to classify each new email (urgent, needs reply, FYI, newsletter, spam-ish), applies a Gmail label, and drafts a reply when one is needed. In n8n this takes four nodes: a Gmail trigger, an AI Agent or LLM node, a Gmail label node, and a Gmail draft node.

What you need

You need an n8n instance (n8n Cloud or self-hosted), a Gmail account connected with OAuth, and an API key for an LLM provider.

• n8n (Cloud or self-hosted on a VPS)
• Gmail OAuth credential in n8n
• An LLM API key (any provider n8n supports)

Step 1: Trigger on new email

Add a Gmail Trigger node set to poll for new messages. Filter out anything already labeled so the agent never processes the same thread twice.

Step 2: Classify with an LLM

Pass the sender, subject, and the first part of the body to an LLM node. Ask for strict JSON: a category, a priority from 1 to 3, and whether a reply is needed. Strict JSON keeps the next nodes simple and makes failures easy to spot.

Step 3: Label and draft

Use a Switch node on the category to apply the matching Gmail label. For emails that need a reply, generate a short draft and save it with the Gmail 'create draft' operation. Never auto-send in v1; drafts keep a human in the loop.

Step 4: Daily digest

Add a Schedule Trigger that runs each morning, collects yesterday's priority-1 emails and open drafts, and emails you a one-screen summary.

Want the full version?

The full Inbox Triage Agent template adds editable label rules, an offline sample run with tests, honest cost-per-run notes, a Python + AGENTS.md version for Claude Code or Cursor, and a changelog.