Skip to content

Automation

Automation is ClusterCode handing a task to an AI agent that spins up a DevBox, does the work, and tears down — either on a recurring Schedule or as an on-demand Run. Every execution happens on infrastructure you own and control: your own workers and cloud workers, never a black-box cloud.

SchedulesRuns
TriggerA cron cadence (e.g. nightly)You start it, once
HumanAutonomous — no one watchingInteractive — you can answer it
Can it ask you a question?No (no human in the loop)Yes — it pauses and asks
Best forRecurring maintenance, periodic checksExploratory work, anything ambiguous

Both run the same way under the hood: a coordinator plans the job and delegates the actual coding to an in-DevBox engine (Claude Code, Codex, or Copilot).

A Schedule is always autonomous. A Run can be either interactive (the default — it can pause and ask you) or autonomous (fire-and-forget, no prompts — a Schedule’s body fired once, on demand). So “can it ask you?” really tracks the mode, not whether it’s a Run.

A few recipes to make it concrete:

  • Nightly dependency audit — a Schedule that checks for vulnerable packages every night and opens a PR bumping them. → Schedules
  • Flaky-test sweep — a Schedule that re-runs the e2e suite three times, finds flaky specs, and files issues with traces.
  • “Refactor this and open a PR” — a Run you start from a ticket; it works in a DevBox, runs the tests, and opens the PR. → Runs
  • A brainstorm that asks you questions — a Run where the agent pauses mid-task to confirm a decision before continuing. → Runs

Automation doesn’t ship your code to someone else’s servers. Each Schedule or Run executes on a worker you registered or a cloud worker you provisioned — machines you control, behind your own network. You pick the host (or let ClusterCode choose from the ones you have), and the AI work bills against your own engine credentials.