Scheduled maintenance runs
A scheduled maintenance run is a recurring job that runs an agent on a cadence to keep its work — or its data — current. Examples: refreshing a summary every morning, re-checking a data source weekly, or producing a regular report. Scheduled runs are managed as tasks under /automation.
What a scheduled run needs
| Setting | Meaning |
|---|---|
| Agent | Which agent executes the run. Requires the Scheduled Tasks feature on the agent. |
| Instructions | The prompt the agent runs each time the job fires. |
| Schedule | A one-time date/time, or a recurring frequency. |
| Timezone | The timezone the schedule is interpreted in (the system stores times in UTC internally). |
| End condition | When a recurring job stops: never, on a date, or after a number of runs. |
Recurring options
Recurring runs support these frequencies:
- Minutely / hourly / daily / weekly / monthly with an interval (for example, every 2 hours, or every 1 week).
- Weekly with specific days of the week.
- Custom with a standard 5-field cron expression.
A recurring job keeps running until its end condition is met. Each run records its result so you can review history.
Set one up
- Enable the feature
Open the agent in the Agent Builder and turn on Scheduled Tasks so the agent can be assigned to scheduled jobs.
- Create the task
Go to Automation and create a task. Assign the agent and write the instructions it should run each time.
- Set the schedule
Choose a one-time date and time, or a recurring frequency. Set the timezone, and for recurring jobs choose an end condition (never, a date, or a number of occurrences).
- Confirm and monitor
Save the task. You can run it immediately to verify behavior, then watch its run history. Pause, resume, or delete it from the task list as needed.
Tips for maintenance jobs
- Keep instructions self-contained. A scheduled run has no person in the loop, so the prompt must include everything the agent needs.
- Pick the longest cadence that still works. Hourly runs cost more than daily ones; don’t run more often than the data changes.
- Set an end condition for jobs that shouldn’t run forever, so stale tasks don’t pile up.
- Review history periodically and pause runs that no longer add value.