MailTag

AI-powered email classification and organization using on-device inference.

What is MailTag?

MailTag is a Python-based email automation tool that classifies and organizes emails into a 19-category business taxonomy with MLX-powered local inference on Apple Silicon. It supports both IMAP and Gmail providers.

Key Features

• On-device AI via MLX on Apple Silicon -- no cloud API required
• Rules first, models second: validated and learned senders, then domain rules, then embeddings and a local LLM
• Action folders (1-A traiter, 2-A payer, 3-A lire, 4-Pour info, Promotions, 5-A revoir) and PARA category folders
• Learning from your decisions: filing a mail out of 5-A revoir teaches MailTag its sender
• Batch operations for efficient email organization
• Webhook API for external integrations
• Automatic backups of the rules with rotation

Quick Start

git clone https://github.com/fjacquet/mailtag.git
cd mailtag
uv sync -U --all-extras
python src/main.py run --provider imap --validate  # Read-only test

Then read the User Guide for day-to-day use.

Classification Chain

Step	Source	Speed
1. Validated sender	Senders you confirmed	Instant
2. Learned sender	Sender after 2 nomic/Gemma agreements	Instant
3. Domain rules	Validated domains, then computed commercial domains	Instant
4. Nomic centroids	Embedding similarity, score at least nomic_threshold	Fast
5. Nomic and Gemma agree	Local Gemma 4 E4B answers by category number	~1-2s
6. 5-A revoir	Nothing decided	Instant

The first match wins, so the fastest path is always tried first. See Classification Strategy.