Privacy.

Your writing never leaves your Mac unless you train.

Connecting a source (iMessage, Apple Notes, Apple Mail, WhatsApp, documents) reads the local data on your machine. Nothing is sent to our servers until you press Train.

When you do train, we send the curated examples (de-duplicated, PII-flagged, with your writing context) to our training infrastructure. The resulting LoRA adapter and a manifest of what was used to make it are stored against your account.

We never train base models on your data. We never share it. We never read it ourselves — the pipeline is automated end to end.

You can delete a source, delete a training run, or delete everything at any time. Deletion is immediate and your model retrains from what remains. The audit log shows every byte that touched our systems.

A formal policy will live here before launch.