JevMail

# Jevmail

**Open-source AI email triage for Gmail.** Reach inbox zero by seeing only the mail that needs a reply. Read-only, local, and sorted by [Jev](https://vercel.com/ai-gateway/models/jev), TypeSafe AI's new decision model, through [Vercel AI Gateway](https://vercel.com/ai-gateway).

Every message lands in one of five trays: **Needs reply**, **Updates** (bank alerts, deliveries, receipts, OTPs), **Promos**, **Sales**, or **Spam**, with a 1–5 urgency score. You watch it happen: previews ride the progress bar, cross the Jev checkpoint, and drop into their tray. 1,000 emails sort in about a minute for around 3 cents.

> Jevmail is an independent side project. It is not affiliated with TypeSafe AI, Vercel, or Google.

## What is Jev?

Jev is a "System One" model from [TypeSafe AI](https://typesafe.ai), released in September 2026 by ChatGPT co-creator Diogo Almeida. It is not a large language model and it never generates text. You give it a block of state and a set of typed questions, and it answers all of them at once with calibrated probabilities: a choice from your options, a score on your scale, or a yes/no. TypeSafe reports it runs 40 to 200 times faster than frontier LLMs and costs $0.042 per million input tokens with free output.

For email classification that is exactly the right shape. Jevmail asks three questions per message: which tray, how urgent, and whether a human wrote it to you. There is no prompt to engineer, no JSON to repair, and no way for the model to invent a label that isn't in the list.

## Why not an LLM?

An AI email assistant built on a chat model has to write a prompt, parse the reply, validate the JSON, retry on failures, and still pays for output tokens. Jev skips all of it. On a 1,000-email inbox that is the difference between minutes an