n8n template · AI Inbox Pack

Sort your emails automatically with AI in n8n (IMAP)

This n8n workflow plugs a language model straight into your IMAP mailbox: every incoming email is read, classified into one of four categories and routed accordingly — Slack alert for urgent ones, Supabase logging for everything.

$ n8n import:workflow --input=tri-emails-ia-imap.json
✔ Import successful — 9 nodes, valid connections

The problem it solves

A professional mailbox receives 50 to 150 emails a day, and only a minority deserves immediate reading. Classic filtering rules (sender, keywords) are quickly outmatched: they understand neither context nor intent, and let the essential through while archiving the trivial.

An LLM reads the email the way you would: it tells a customer complaint from an alarmist newsletter, and an invoice reminder from inbound prospecting. This template puts that reading into production in n8n, no code required, for a few cents a day.

How the workflow works, node by node

Node names are in French, exactly as they appear in the imported workflow — the logic is language-agnostic.

1. 01

   Email entrant (IMAP)

   Email (IMAP)

   Connects to your mailbox over IMAP (Gmail, Outlook, OVH…) and triggers the workflow for every new email in INBOX. Works with any provider, unlike proprietary triggers.

2. 02

   Préparer l'email

   Set (fields)

   Keeps only the useful fields (subject, sender, body truncated to 4,000 characters) to control token cost and avoid sending attachments to the model.

3. 03

   Classifier l'email

   LLM chain

   The LLM chain applies a classification prompt: four categories (urgent, client, admin, newsletter), a one-sentence summary and an important-sender flag.

4. 04

   Modèle OpenAI

   OpenAI model

   Model sub-node wired to the chain via ai_languageModel. Set to gpt-4o-mini, more than enough for classification and very cheap.

5. 05

   Format de sortie

   Structured output

   Forces the model to answer in a strict JSON schema (category enum, summary, boolean flag): the rest of the workflow can rely on it without fragile parsing.

6. 06

   Router par catégorie

   Switch (routing)

   Routes each email to the output matching its category. This is where you add a fifth category or change a branch's destination.

7. 07

   Alerte Slack urgent

   Slack

   For the 'urgent' branch only: posts an alert to your channel with sender, subject and summary, for real-time reaction.

8. 08

   Fusion

   No-op

   Convergence point of the four branches: whatever path was taken, the email continues to logging.

9. 09

   Journaliser dans Supabase

   Supabase

   Inserts one row per processed email (subject, sender, category, summary, timestamp) into the emails_tries table: a queryable history and free flow statistics.

What you need to run it

• ✔n8n ≥ 1.60 (cloud or self-hosted), LangChain nodes included
• ✔IMAP access to your mailbox (port 993, app password recommended)
• ✔An OpenAI API key
• ✔A Slack bot with the chat:write scope
• ✔A free Supabase project with the emails_tries table (SQL provided in the pack guide)

Customization ideas

• →Replace the four categories with your own: the prompt, output schema and Switch must stay aligned
• →Add an 'invoice' branch pushing accounting documents to Google Drive or your management tool
• →Swap the model for Claude or a local model (Ollama) by replacing the model sub-node