Skip to content

Me & settings

The Me tab is account and app settings. The box’s own configuration lives in Self-Host Reference, not here.

Your email, sign out, and delete account. Sign-in is the email + 6-digit code flow — see Sign in.

Your plan and purchases. Purchases run through your app store account — manage or cancel in its subscription settings; Restore purchases brings an existing subscription to a fresh install.

Require Face ID (biometric unlock) locks the app itself, with a configurable auto-lock interval. This is a device-side lock on top of box auth — losing the phone doesn’t expose Gray.

A continuous slider — drag to preview the pace, release to save. See Voice. Gray speaks English (US) for now.

Two views behind one switch: Model and Stack.

Model is which model answers when you talk to Gray. Auto sits at the top — Gray picks the best your box serves, and the card names the model answering right now. Below it, the providers your box actually has keys for, each with its models; the one that’s live carries a RUNNING tag. Tap a model to make it answer; tap it again to hand the choice back to Auto. If your pick isn’t what’s running yet, a note under the list says why. Provider usage meters — a bar and its reset time — appear only after your box reports a real limit; a healthy box shows none.

Stack is what Gray found running on your box: Agents (the coding agent CLIs it detected — tap one for its activity), Logins (model credentials, always masked, never stored in the app), and On PATH (the runtime tools it found). ⟳ Rescan re-runs on-box detection after you change the box. See Models & keys.

Two views behind one switch: Your box — every skill installed on your box, with its own ⟳ Rescan — and Library, the catalog Gray curates (hosted, so nothing to rescan). Both read the same way: a search field, Recently used (your three most recent), then All skills · A–Z. Each row shows how recently it was used (2H AGO); means no recorded use yet — Gray counts use from your box’s own records, which start recording now, so older activity may not show. When two skills share a name, a small tag names each one’s source. Skills outside your plan stay listed but read LOCKED — tapping one tells you what it would take.

Tap any skill to open it in Talk, or just say its name there. There’s no install button anywhere — skills live on your box, and Rescan picks up whatever changed. Agents and model credentials live under Brain, not here.

Three categories: Agent updates, Alerts & mentions, Product updates. Critical approvals always come through; critical alerts deep-link straight into triage.

  • Feedback — structured reports straight to the maker: pick what’s wrong, add detail, attach a screenshot or recording. Gray may also occasionally ask how it’s going — a quick, dismissible prompt.
  • Documentation — these docs, in-app.