Clockwork
The calendar where your AI agents show up for work
111 followers
The calendar where your AI agents show up for work
111 followers
Book AI coding agents on a real calendar. Clockwork runs them unattended in sandboxed git worktrees on your Mac, pauses for approval when something looks risky, and files a report with the actual dollar cost. Bring your own key or ride your existing subscription. Schedule recurring agent jobs on a real calendar, run them unattended in isolated sandboxes, and review honest reports.
This is the 2nd launch from Clockwork. View more
Clockwork
Launching today
Book Claude Code, Codex, OpenCode or Hermes on a calendar. Clockwork runs each job in a sandboxed git worktree, caps the spend, pauses risky steps for your approval (even on your phone), and files a report you can turn into a PR. Pipelines, worker Macs, a CLI.












Free Options
Launch Team
Clockwork
Hi Product Hunt. I'm Vimox, and I build Clockwork alone.
I launched here on 4 September. You asked six questions in the comments, and five of them were about trust:
If it pauses for approval at 2 am, does the slot die or hold until I say yes?
What does a timeout do to an agent in the middle of a rebase?”
At the time, two of my answers were wrong. So I fixed the product first, and then corrected the answers.
Fifteen releases later, this is what it does:
It waits for you. A risky step pauses the run. The approval holds for the run's entire time budget, and Clockwork can find you on Telegram, Slack, email, or through a notification. On Telegram, you can tap Approve or Deny. (This is a Claude Code feature; for the other engines, the sandbox provides the containment.)
It keeps what matters. A run cut off in the middle of a rebase keeps its worktree, and the report says so. Every run gets its own worktree and branch, inside a macOS sandbox that cannot read ~/.ssh or your cloud credentials.
It runs on the Mac that stays awake. Pair a Mac mini as a worker with a signed handshake that you approve. Pin a job to it, and your laptop can sleep. If a worker goes quiet, its run is marked lost. It is never quietly reported as done.
One job can become a pipeline. Fan out to three reviewers, then fan in to one summary that waits for all three.
You say it the way you think it. Weekdays at 9. Clockwork shows how it interpreted the rule and when it will actually fire. If the rule is ambiguous, it asks. Drag a booking to a new slot, and use Undo if you change your mind.
The morning after. The report ends with the next action. One of them is Open PR. There is also a terminal interface: clockwork status, clockwork runs, and clockwork approve.
What it still does not do: the build is not notarised (the install section has the one-line fix), updates are manual, it is macOS-only, and Docker execution is not wired to tasks yet.
The repo is public and source-available. The sandbox profile, credential deny list, and their tests are Apache-2.0, so you can check what a run can touch. The test suite also fails when the docs claim more than the code actually does.
That's why this comment has a limits paragraph.
It's free for personal use.
One question for you: what is the first job you would leave running overnight?
I read every answer.