Clockwork has launched an early macOS app that treats coding agents more like scheduled workers than interactive chat sessions. Users can book Claude Code, Codex CLI, OpenCode or Hermes jobs on a calendar, assign a repository, budget and time window, and let the task run unattended. The product also supports recurring schedules and chained agent jobs.\n\nThe security design is more explicit than the usual “run an agent in the background” pitch. Clockwork says every run gets a fresh git worktree and branch under a macOS Seatbelt sandbox; writes outside the worktree are denied, sensitive credential locations are blocked, and risky actions pause for approval. A supervisor also enforces dollar, turn and wall-clock limits outside the model. Reports record the branch, diffstat, costs, approval history and transcript.\n\nClockwork is local-first: the current product page says there is no Clockwork account, cloud service or telemetry, with run data stored in local SQLite. It can use existing CLI subscriptions or user-supplied API keys stored in macOS Keychain. The app requires Apple silicon and macOS 14 or newer, and scheduled work only runs while the Mac is awake.\n\nThere are important early-build caveats. Version 0.4.0 is not notarized or signed by Apple, so users must verify the published checksum and explicitly clear macOS quarantine before launching it. The maker also says hard wall-clock timeouts terminate the agent process rather than waiting for a git operation to finish; the worktree is preserved for recovery, but a rebase or commit can still be interrupted mid-operation.\n\nClockwork is currently free for personal, non-commercial use. Commercial use requires a separate license and paid tiers are not yet generally for sale. Product Hunt’s launch discussion also shows the product is still collecting feedback on recurring-run and timeout edge cases. That makes Clockwork interesting as an orchestration pattern today, but still an early product rather than a mature unattended-agent platform.