Automations

Automations let an Amp thread continue working later or on a repeating schedule. Tell Amp what to do and when in the same language you use for any other task. You do not need to write a cron expression or configure a separate workflow.

Amp saves the schedule and a prompt on the current thread. When the schedule fires, Amp wakes the agent with a short message. The agent reads the saved prompt with the get_schedule tool, which you can expand in the thread to see the full prompt, and then carries it out. The agent keeps the thread’s context and history, so it can continue earlier work instead of starting over. In an orb thread, Amp resumes the orb when the task needs the repository or other tools.

A schedule can run once or repeat. Each thread can have one schedule. A repeating schedule runs until you pause or delete it, it reaches its configured end, or Amp clears it after reaching a completion condition you gave it.

Examples

Investigate Slow Queries Every Morning

Every morning at 9:00, find the five slowest database queries from the past 24 hours. Investigate them in this orb using ultra mode, then DM me the list and your findings on Slack.

This schedule runs an investigation every morning and sends the result to Slack.

Follow Up After a Change Ships

Merge this change, then remind me on Slack in two days to remove the feature flag and roll it out to everyone.

Amp finishes the current task and sets a one-time follow-up on the same thread. The reminder keeps the context about the change and its feature flag.

Watch a Long-Running Job

Check this backfill job every ten minutes. Ping me on Slack if it stalls or starts showing errors. Tell me when it finishes, then stop checking it.

The same thread checks the job until it reaches the completion condition. It can report a problem as soon as one appears and clear the schedule when the job finishes.

Triage New Errors Every Hour

Every hour, use the inference-error-triage skill to inspect errors from the past hour and group related ones. For each new group, start a thread to fix the errors and report back in the #bugs Slack channel. Do not report groups that were already handled.

This prompt tells Amp what evidence to inspect, how to avoid duplicate work, when to start other agents, and where to report the result.

Write a Useful Prompt

A scheduled prompt should say:

  • What Amp should inspect or change
  • What result Amp should report, including what to do when nothing changed
  • Where Amp should send the result, if the thread is not enough
  • When a monitoring task is complete and the schedule should stop

Put the timing in your request to Amp. Amp stores it as the schedule and saves the remaining task instructions as the prompt that runs each time.

Manage an Automation

Scheduled threads show an automation indicator. Open it to see the title, schedule, next run, and saved prompt. You can pause or resume the automation with the switch, or delete it with the trash button.

You can also manage it by talking to Amp in the same thread:

Change this schedule to weekdays at 8:30 AM.
Pause this automation until I resume it.
Delete this schedule.

One-time schedules complete after their first run. If a run fails, Amp pauses the automation and shows the error. Fix the problem and resume it when you are ready.

See Right on Schedule for the announcement and a video of an automation at work.