Routines

Automate recurring work with scheduled agent runs.

Routines turn your best agent prompts into standing checks that run themselves. Save an Agent prompt once and have it run automatically on a recurring schedule, so recurring investigations, health checks and reports happen in the background — no need to trigger them manually every time.

Routines list, showing scheduled agent prompts

Creating a routine

From the Routines page, click Create routine and fill in:

• Routine name — e.g. "Scan Airflow DAGs daily".

• Routine prompt — the exact instruction sent to the agent on every run, e.g. "Scan Airflow DAGs daily. Surface runs that failed or took longer than usual."

• Attached files (optional) — drag and drop or browse for files (images or other attachments, up to 2MB) to include with every run.

• Routine schedule — choose a cadence:

  • Hourly (with a minute of the hour)

  • Daily (with a time, in UTC)

  • Weekly (with a day of the week and a time, in UTC)

  • Monthly (with a day of the month and a time, in UTC)

• Slack notifications — toggle notifications on run completion and choose a Slack channel to post to.

Routines can be toggled Active or Inactive at any time without deleting them.

Routines generation catalog

Routines list

Routines are shown as cards, each displaying the routine's name, a truncated preview of its prompt, and a cadence badge (e.g. "Daily", "Weekly"). Selecting a card opens its details on the right.

Routine details

The detail panel shows the routine's prompt, attached files, schedule and Slack notification settings, along with:

• Run now — manually triggers the routine immediately, without waiting for its schedule.

• Edit (pencil icon) — update the routine's prompt, schedule or notification settings.

• Delete (trash icon) — removes the routine, after a confirmation dialog. This can't be undone.

• An Active/Inactive toggle.

Routine detail panel with schedule, notifications and run history

Runs

The Runs section lists every past run of the routine, labeled with its trigger timestamp. Clicking a run opens it in the Agent page, showing the full conversation, tool calls, generated code or SQL, catalog lookups, and any query results or charts — exactly as with a manually started agent conversation. If a run failed to start, its entry is shown disabled.

If the routine has never run, this section shows "Never triggered."

If a scheduled run comes due while a previous run of the same routine is still in progress, it is skipped rather than started concurrently.