Clockwork
Put your coding agents on a schedule — Clockwork runs them sandboxed on your Mac and files readable reports.
Last verified:
What is Clockwork?
Clockwork is a macOS application that schedules and runs AI agents (Claude Code, Codex, OpenCode, Hermes) on automated calendars in isolated sandboxes. It's designed for developers who want to automate repetitive repository tasks like dependency updates, testing, code review, and documentation with cost controls and safety guardrails.
Clockwork pricing
Pricing model: Freemium
Clockwork pros
- 13 built-in specialist agent profiles covering engineering, operations, and documentation tasks
- Sandboxed execution with strict security: isolated worktrees, credential deny-lists, Seatbelt profile isolation, and automatic credential masking
- Real calendar-based scheduling with timezone support, recurrence rules (RRULE/cron), and overlap policies
- No cloud dependency—runs locally on your Mac with full transparency; no account, telemetry, or external server needed
- Built-in cost controls: dollar caps, turn caps, wall-clock timeouts, and detailed cost reports per run
Clockwork cons
- macOS only—currently limited to Apple Silicon and Intel Macs
- Early stage (v0.4.0 beta)—reproducible builds not yet supported, limited engine choices
- Requires Mac to be awake for scheduled jobs; for true overnight automation, an always-on machine is needed
- Pricing model not clearly specified in documentation beyond per-run costs shown in examples
Frequently asked questions about Clockwork
What can Clockwork agents access?
Agents can access their own isolated worktree and branch (cut from your base), your CLI login credentials, and a dollar/turn budget. They cannot access your main checkout, SSH keys, cloud credentials, .gnupg files, or shell history. Writes are confined to the agent's worktree.
How does the approval system work?
When an agent performs a risky action, it pauses and asks for approval. If you don't respond in ~2 minutes, the action is denied by default (fail-closed). All approvals and denials are logged in a safety journal.
Can I create custom agents?
Yes. You can write your own agents in the app by selecting skills, a permission mode, budget defaults, and a system prompt.
How are reports delivered?
Every run files a searchable report into your inbox with a summary of what it did, the branch and diffstat, costs against your caps, all approvals and denials, and the full transcript.