Local-first · macOS · works with the logins you already have
Your AI agents have a calendar now.
Stop babysitting terminals and cron scripts. Book agent work on a real calendar — Clockwork runs it in an isolated sandbox at the hour you chose and files a report you can actually read.
Free during beta · No API key · Uses your Claude Code / Codex / OpenCode / Hermes login
The default view: every agent job on one month grid.
Act I — the problem
Your agents are everywhere. Their work is nowhere.
You have powerful AI agents. But the work they do lives in terminal scrollback,
cron tables, CI logs and chat threads — invisible, unrepeatable, unaccountable.
Act II — your agents' workday
Human time + agent time, on one grid.
Clockwork overlays your meetings with scheduled agent work, so you always
know what ran while you were heads-down — and what needs you now.
Why this exists
Agents are powerful. Recurring work has no home.
Right now recurring agent chores live in crontabs, shell scripts, CI pipelines, and sticky notes. Clockwork gives that work a home on your calendar — with budgets, approvals, isolation, and searchable history built in.
Book a job in under a minute — prompt, repo, budget, schedule.Pick the engine per task: Claude Code, Codex, OpenCode, or Hermes Agent.Reports worth reading: what it did, what it skipped, what it cost.⌘K everything: navigate, switch themes, create tasks.
The loop
Book → Run → Review → Repeat
Book
Choose an agent profile, a repo, a budget, and a time. One-off or recurring.
Run
Your own CLI engine executes unattended inside an OS-sandboxed git worktree.
Review
A human-readable report lands in your inbox — branch, diffstat, cost, transcript.
Repeat
Make it weekly. Search your run history. Trust grows with evidence.
Differentiators
Built like a product, not a demo
🗓 A real calendar
Month grid, recurrence, missed-run policies, per-repo mutex — scheduling is first-class, not an afterthought.
🔀 Provider freedom
Claude Code today, Codex / OpenCode / Hermes when you want. Switch per task without rebuilding anything.
🛡 Human-in-the-loop
Risky actions pause the run and ask you. Unanswered asks fail safe — never silently approved.
📦 Real isolation
Every run gets its own worktree and Seatbelt sandbox. Writes locked to scope. SSH keys unreadable.
🔎 Searchable history
Full-text search across your retained reports and transcripts. Your agent's track record becomes auditable.
⌨️ Keyboard-native
⌘K palette, global shortcuts, instant inbox search. Built by terminal people, for terminal people.
Agent library
Ready-to-run specialists, not blank prompts
Every profile ships with mission, constraints, safety rails, output contract, and success criteria — so "update my deps" doesn't turn into "rewrite my repo".
Dependency Surgeon
Groups upgrades, reads breaking changes, runs tests, opens PRs. Never force-pushes.
macOS Seatbelt profiles per run. File writes restricted to the run's worktree.
Credential hygiene
Sanitized environment. SSH agents and credential helpers unreadable from inside a run.
Hard budget bounds
USD soft caps, turn limits, wall-clock timeouts — enforced by the supervisor, not the model's good intentions.
Audit journal
Every approval, denial, and state change recorded. Nothing happens off the record.
🔍 Don't take our word for it. The Seatbelt profile, the credential
deny list and the run-environment allowlist are
published under Apache-2.0
with the tests that prove them — the suite writes a fake secret into ~/.ssh and ~/.aws,
then runs a real cat inside a real sandbox and asserts it fails. Read it, run it on your own Mac, and
see what it does not prove. Clockwork itself stays proprietary; the security boundary is open.
⏾ The honest constraint: jobs run when your machine is awake. Clockwork keeps a plugged-in laptop awake through booked windows and tells you loudly when sleep caused a miss. For true overnight runs, use an always-on machine. We'd rather tell you now than disappoint you at 2am.
Providers
Bring the engines you already pay for
Claude Codesubscription login · default
Codex CLIOpenAI
OpenCodemulti-model
Hermes AgentNous Research
Detection is automatic — if the CLI is installed, it appears.
Get started
Give your agents a workday.
Download Clockwork for Mac, connect once, book your first chore tonight.
Clockwork is not notarised by Apple yet.
An Apple Developer certificate costs $99/year and this is an early build, so the app is
unsigned. macOS will refuse to open it until you clear the download quarantine flag. We would
rather tell you that plainly than have Gatekeeper call it “damaged”.
brew trust is required because Homebrew refuses casks from third-party taps by
default — you are explicitly saying you trust this one.
Homebrew verifies the SHA-256 before installing, then you clear the quarantine flag
yourself — Homebrew 6 removed the --no-quarantine option, so there is no
way to skip it. You are still clearing quarantine on a binary whose hash was checked for
you, which is why this beats a raw download.
Or install the DMG by hand
# 1. verify what you downloaded
shasum -a 256 ~/Downloads/Clockwork_0.4.0_aarch64.dmg
# 2. open the DMG, drag Clockwork to Applications, then:
xattr -dr com.apple.quarantine /Applications/Clockwork.app
# 3. launch it
open -a Clockwork
Step 1 should print the hash shown in
checksums-sha256.txt.
If it does not match, do not install it — tell us instead.
Step 2 removes the quarantine flag macOS adds to downloaded files; you are telling your Mac
you trust this specific app. Only run it on a file whose checksum you just verified.
Pricing
Free today. BYOK, always.
Bring your own API keys — Clockwork never marks up provider charges.
Whatever Clockwork costs later, it will never be a markup on your provider
usage — those stay separate bills.
Personal
Free during beta
Local agents in sandboxed worktrees
Unlimited recurring schedules
All BYOK providers and CLI engines
Human approval gates
Budget guardrails
Run history and reports (30 days)
Approvals and budget caps are never paywalled — they are the brakes on unattended code.
Pro
Not yet priced
Everything in Free
Run history kept for a year
Policy engine — restrict engines, models and spend
Audit log (30 days)
Up to 50 event-triggered agents
These are the pieces teams ask for. They are built, but not for sale — I would rather find out what is worth paying for than guess at a number. Beta users keep their history either way.
Teams & Enterprise
Talk to us
Unlimited audit log with export
Extended retention (2 years)
Policy engine across the team
SSO / SCIM (planned)
Clockwork is proprietary software — previously FSL-1.1, now under the LICENSE in the repository. No payment is collected on this site yet.
Get in touch
Questions, bugs, or want to try it on your repo?
Clockwork is an early build and I read every message. Tell me what broke, what you expected,
or what would make it worth paying for.
