A remote Activity Monitor for the agentic era

Your Mac’s Activity Monitor, in your pocket.

Agents, dev servers, builds and exports on your Mac, and the CPU, memory and heat they cost, on your iPhone. It stays quiet until something needs you or breaks.

For Mac (macOS 14 or later) and iPhone.

Farglass on an iPhone: a Mac mini with Claude working in shop at 62% CPU, and a MacBook Pro warning it has only 8 GB free.
The Farglass menu on a MacBook Pro: Claude Code in storefront is waiting for your approval to use Bash; another Claude Code session in api is working. CPU 18%, memory 21.4 of 48 GB, 52 degrees.

How it works

  1. Install Farglass on your Mac. It lives in the menu bar and watches what’s running.
  2. Use the same iCloud account on your iPhone. That’s the pairing. No codes, no sign-up.
  3. Open the iPhone app. Your Mac is already there, with a widget and Live Activity when an agent needs you.

What it watches

Agents
Claude Code with its exact status from its own hooks: working, waiting for your approval, done. Codex from its own done signal.
Dev servers and ports
Next.js on :3000, Storybook on :6006, by project and folder. Not node, node, node.
CI
GitHub Actions runners on your Mac: which job, which pull request, how the last one went. Read from the runner’s own log, no token.
Exports
Media Encoder says when a file is done. Final Cut Pro gets the facts: when it stopped keeping the Mac awake, and when the file was last written.
Resources
CPU, memory, swap, disk and heat, and what’s using them most.

Facts, not guesses. States come from the agents’ own signals. Everything else is shown as a measurement, like “teaser.mov last written 14:01, 4.2 GB”, never “probably done”.

The Farglass menu listing Claude Code, Codex and Next.js by folder, then five running things: Vite on :5173, Next.js on :3000, Storybook on :6006, and two Actions runners, one on CI Verify for #1903, the other's last Deploy failed.

Only when it matters

Busy isn’t news. Broken is. Farglass interrupts you for four things:

  1. An agent is blocked, waiting on you.
  2. Your work is about to stop: the battery guard lets go, or macOS cuts out for heat.
  3. It’s hot enough that macOS is slowing the chip down.
  4. The work is done: an export its app reports finished, or once, when everything has finished.

Everything else, a full CPU, a warm chip, a server you left up, waits quietly in the app.

FarglassClaude is waiting for approvalstorefront
FarglassExport finishedteaser.mov
The Farglass menu after an export: Final Cut Pro stopped preventing sleep at 14:02, teaser.mov last written 14:01, 4.2 GB; Adobe Media Encoder finished in 12 minutes.

Act from your phone

A dev server you forgot is still eating memory? Stop it, and everything it started, with one tap. Keep the Mac awake while an agent runs, or let it sleep.

Your phone can only act on what your Mac listed: keep awake, sleep, stop or pause a listed server, clean up a finished worktree. No shell, no arbitrary commands.

Your data stays in your iCloud

Where it goes
Your own iCloud, in its private database. Your Mac writes, your iPhone reads. There is no Farglass account.
What it sends
The summary you see: the names it works out, never the command lines it read them from.
Our one server
A small relay used only to refresh your home-screen widget. It passes Apple a “content changed” nudge and never sees a reading. It keeps your phone’s push address and a fingerprint of your Macs’ key, nothing else.
Never
What’s in your files, what you type, or your screen.

Questions

Why not just Activity Monitor?
It stays on the Mac you walked away from, and it shows you node. Farglass tells you it’s Next.js on :3000 in storefront, knows when Claude is waiting on you, and puts that on your phone.
Does it need an account?
No. Sign in to the same iCloud account on your Mac and iPhone and they find each other.
One Mac or many?
One Mac is the main case. Add more, like the mini in the corner running CI, and they show side by side.
Which agents?
Claude Code and Codex, with exact status from their own signals: Claude Code’s hooks, and Codex’s notice when a turn is done. Other agents show as measured facts: how long they’ve run, CPU and memory.
How fresh is it?
Each Mac reports every 15 seconds.

Watch your Mac from anywhere.

The beta is open to developers running agents, and to video editors with long exports.