Skip to content

Automations

An automation is a saved agent run — a prompt, an agent, a target — executed on a schedule. Nightly dependency audits, morning triage of new issues, a Friday changelog draft: anything you’d start by hand at a regular time.

  • You define the automation in the app: what to run, where, and a cron schedule.
  • The server’s dispatcher fires it at the scheduled time (it also survives server restarts and multiple replicas — exactly-once per tick).
  • Each firing creates a normal session on the target machine. It appears in your session list like any other — open it, read the run, keep prompting if you want to take it further.
  • You can also trigger an automation manually at any time, and review its run history.
  • Write prompts that terminate. Include a clear deliverable (“write the summary to TRIAGE.md and stop”) so runs don’t idle.
  • Cap spend. Scheduled + unattended is exactly where a spending cap (Pricing) earns its keep.
  • Prefer read-only prompts unattended. Let scheduled runs investigate and report; keep destructive actions for sessions you’re watching.

For agent-driving-agents workflows — one session coordinating others, spawning worktree runs, managing project cards — see Project Manager mode under Agents.