Your notch, now a status island.

One glance is enough

• Working: The agent and its state, how long it has been at it, and the task itself, which rolls through what it said and what it is touching.
• Needs you: A pixel question mark when it is waiting for your answer.
• Done: A green check the moment a task finishes.
• Failed: A red mark for a failed build, test or command.
• Listening: A dev server that is up, with its port.

Answer the agent without leaving your editor.

When Claude Code, Codex and others ask for permission, the prompt opens in the notch with the command in plain sight. Allow it, always allow it, deny it, or send it back to the terminal. When Claude Code or Grok Build ask you a multiple-choice question, answer that in the notch too: pick one, tick several, or type your own. Anything Runnock cannot answer falls back to the agent's own prompt, so nothing is ever stuck.

Finished work drops out of the notch.

Each finished task, failed build or waiting agent separates from the notch like a drop and hangs below it, with a small tail pointing back. It is drawn the same way in the app: no video, no trick.

See what the agent changed.

When a turn ends, a bubble says how many files changed and by how many lines. Open it for the list, then any file for its diff, in a window of its own. Runnock compares your work tree from before the turn with the one after it, so every agent gets this, and your repository's index, branch and stash are never touched.

Know before you hit the limit.

Runnock tells you when Codex, Claude Code, Grok or Devin reaches 50%, 80% and 95% of a limit, when it runs out and when it comes back, and when it is usable again.

Everything that runs on your Mac

• Dev servers: Start one from a project folder. It appears with its port, an Open button and a Kill button. Docker containers and Compose projects are watched too.
• Builds and tests: Xcode builds, test runs and any zsh command that takes a while report back with the error count or the line that failed.

It reads what Claude Code reports.

Claude Code tells Runnock far more than "working" and "done". The island uses it.

• Compacting: A /compact you asked for and one Claude started itself are told apart. Yours ends with a check instead of staying on "working".
• Tasks, model, subagents: Its own task list as 2/5, the model it is on, and how many subagents are running.
• Why it stopped: A rate limit, a sign-in problem or a billing error is named, not just "failed". When the limit resets and Claude carries on by itself, it says that too. A session that went quiet can be ignored until it speaks again.

Know when the pipeline needs you.

With the gh command you already use, Runnock reads your GitHub notifications for pull requests (a review asked of you, changes requested, a mention, an approval or a merge), follows a CI run from start to red or green, and watches the deployment of a repository you chose: Vercel, Cloudflare Pages and Netlify show up while they build, and again when they go live or fail, with the line that broke. Merge conflicts need no login at all. Each one is a line on the island that goes away when it is over.

Push a status from any script.

runnock alert puts a line on the island when tests fail, a deploy finishes or a backup is done. It talks to the app over a local socket, opens no network port, and exits with a code your script can read. Shortcuts can do the same with a runnock://alert link.

It finds your agents and connects them.

On first launch Runnock looks for the coding agents on your Mac and offers to connect them in one click. Hooks are added next to yours, after a backup.

You decide what it watches

Every source is a switch, every alert has its own sound, and nothing runs that you did not turn on.

Pay once. Keep it.

No subscription, no account, no analytics.

57615 for 1 device,

• One-time payment
• Every feature, every agent
• Your license key arrives by email
• Plans for 1, 2 or 3 devices
• Free to download; it unlocks with your key

Questions

Does Runnock see my code?

Not off your Mac. Runnock reads the events your agents report: the tool name, a short summary with secrets masked, and the message you sent. For Claude Code, while Live messages is on, it also reads the end of the conversation file Claude Code keeps on your Mac, so the notch can show what it said, what each tool printed, the diff of a file it edited and pictures you sent. That appears on your screen (mind screen sharing) and is held in memory only: nothing is written down or uploaded, secrets in text are masked, a picture is not. Switch it off in Settings and the file is not opened. There is no analytics. What leaves your Mac is the license check, GitHub requests you turn on (through your own gh login), and, if you keep usage alerts on, a request to Anthropic or xAI for your usage with the sign-in Claude Code or Grok keeps on your Mac. Clipboard context is off until you switch it on, and keeps nothing.

Can my own scripts talk to it?

Yes. runnock alert (or a runnock://alert link from Shortcuts) puts a status on the island. It uses a local socket, opens no network port and keeps nothing on disk. When Runnock is not running, or you switched it off in Settings, the command exits with a code that says so.

What do I need?

A Mac with Apple silicon, on macOS 26 or later. Intel Macs are not supported.

Does it work on a Mac without a notch?

Yes. On a screen without one, Runnock draws a notch at the top of the screen.

How many Macs can I use a key on?

As many as the plan you bought: 1, 2 or 3. Deactivate one in Settings and the key is free for another Mac.

Can I try it first?

There is no trial. Downloading is free, but Runnock stays locked until a license is activated. If it is not for you, see the refund page.