Agent Pager
Agent Pager
When your coding agent — Claude Code, or GitHub Copilot in the CLI and VS Code — needs a permission, has a question, or finishes a long run, your iPhone, Apple Watch, or Vision Pro gets the message. Allow or deny from the lock screen. End-to-end encrypted, no accounts, nothing to configure twice.
The real command on your lock screen, with Allow and Deny under your thumb. Allow requires Face ID.
Claude's multiple-choice questions arrive with real option buttons — answer from the watch, or dictate a custom reply.
Long run finished, stalled, or waiting on you? Your phone knows — but only when you're actually away from the desk.
Everything is encrypted between your computer and your phone. The relay in the middle can't read a single byte of it.
http://127.0.0.1:37123/pair), scan the QR with the app, and confirm on the desktop. Done: the next prompt your agent parks will reach your phone.Pair as many computers as you like — work machine and home machine prompts arrive side by side, each labeled with its machine name.
No. On Mac, the free Agent Pager Helper is all you need — a small menu-bar app with no Elgato software involved. On Windows, the standalone helper is available too (see Setup above — unsigned for now, so expect a one-time SmartScreen "More info → Run anyway"). If you own both, they coexist politely — the plugin takes charge and the helper stands by automatically.
In desk mode, about 30 seconds — miss it and the agent's normal in-app prompt appears, so nothing is ever lost. Flip on Away mode (the moon button in the app) and prompts wait nearly five minutes, tuned for a phone in your pocket. The countdown on each prompt always shows the real deadline.
Nothing bad. Dismissing is never treated as a deny — the prompt stays live in the app until it expires, the dismissal is logged in your activity feed, and if it eventually expires unanswered the app pages you again with a "stalled" alert.
Yes — notifications mirror to the watch automatically with the same Allow/Deny buttons, and question options appear as tappable choices. No separate watch app needed. Custom "Other" answers support dictation from the wrist.
Almost nothing: a random pairing identifier, a push token, and encrypted blobs it forwards. Message contents — commands, file paths, questions, answers — are sealed end-to-end between your computer and your phone with keys that never leave your devices. The relay is open source and self-hostable if you'd rather run your own. Details on the privacy page.
Three things, in order: the desktop side is running on the computer — the Agent Pager Helper (menu-bar icon on Mac, system-tray icon on Windows) or, for Stream Deck owners, the Stream Deck software (the plugin lives inside it); the computer is online; and the app is paired (Settings → your desktop should show as confirmed). Re-pairing takes thirty seconds and always fixes a broken state: scan the QR again and confirm on the desktop.
The prompt expired before your answer arrived (or someone answered from the desk first — first answer wins, by design). Late answers are harmless: they're simply ignored, and the outcome in your Recent list tells you who or what settled each prompt.
Yes. Hooks fire wherever Claude runs — VM, SSH session, headless box — and they can point at the listener on another machine over your network with a shared token. Prompts from every wired-up machine reach the same phone. Setup steps are in the desktop-side FAQ; the one rule to remember: re-copy remote hooks after updating, since they don't refresh themselves.
No. Agent Pager is an independent product by Eudaimonia Software, not affiliated with Anthropic or GitHub. Claude and GitHub Copilot are trademarks of their respective owners; this app simply works alongside the developer tools you already run.
Stuck on something the FAQ doesn't cover, or found a bug? Email support@eudaimoniasoftware.com and include what you were doing and what you expected — replies usually within a day or two.