Mailaya local service (Laya)
============================

Mailaya's judgments (importance, needs-reply, "ask") come from a Laya model running
on your own computer at http://127.0.0.1:8765. Mail never leaves your machine.

Install (macOS, Apple Silicon recommended):

    cd mailaya-server
    ./setup.sh

It installs uv, a Python 3.12 environment with Laya, downloads the models (~3 GB,
once), and registers a login service that restarts on crash. Takes ~5 minutes the
first time, then ~40 s to load after every login.

Check:      curl -s http://127.0.0.1:8765/health      → "status": "ready"
Log:        ~/Library/Logs/laya-server.log
Uninstall:  ./setup.sh uninstall      (models stay in ~/.cache/huggingface; delete to free 3 GB)

Not a Mac: create a venv, `pip install "laya @ git+https://github.com/NandhaKishorM/laya@42626c3"`,
then run `python server/server.py`. Without an Apple GPU it runs on CPU, noticeably slower.

Keep this folder where it is: the service runs from it.
