Skip to content

CommandLatch Docs

CommandLatch is a safe remote-action agent for your own Mac. A small menu-bar app runs on the machine; from your phone, a web dashboard, the terminal, Siri, or a Claude Code hook you can tell it to lock, sleep, keep-awake, adjust volume, or pop a notification — and nothing else.

CommandLatch is intentionally focused. It cannot type, click, run shell commands, read your files, or see your screen. Every request is validated against a fixed list of permitted actions — nothing else can run.

  1. Install — download the DMG, drag CommandLatch to Applications, open it. → Install
  2. Pair — open the dashboard, sign in, and type the 6-digit code the app shows you. → Pair
  3. Try it — once the device shows Online, click Lock in the dashboard. Your Mac locks within a couple of seconds. → Dashboard

From there, set up the Claude Code hook or a Siri shortcut if you want hands-free triggers.

  • A Mac running a current version of macOS (Apple Silicon or Intel).
  • An email address — sign-in is a one-tap email link, no password.
  • An internet connection on both the Mac and whatever you trigger actions from. CommandLatch talks to its backend over the network; it does not connect device-to-device.

There is no account to create in advance — signing in with your email the first time you open the dashboard creates your account automatically.

you (phone / browser / terminal / Siri / Claude Code)
│ "lock my Mac"
CommandLatch backend ── checks the request against a fixed list of
│ allowed actions, then queues it
the menu-bar app on your Mac ── polls for queued commands every
couple of seconds and runs them locally

Nothing reaches your Mac directly. A trigger queues an action; the menu-bar app picks it up on its next poll and runs it. Actions remain queued even after you close your browser or terminal. The Mac must be awake and online — CommandLatch cannot wake a sleeping machine.

For the full trust model — what’s stored, what can and can’t run, how to revoke access — see Security & trust model and Privacy, and the accepted trade-offs in Known limitations.