Run a weekly short video series on autopilot with a Sume schedule

To publish a short-video series every week without a person starting it, create a Sume schedule with a five-field cron expression such as 0 9 * * 1, an IANA timezone and a per-run spend cap. Each firing is one agent run that you can poll or receive by webhook. The Developer API cannot create schedules, so you author it once in the dashboard and then call it from code.

Set it up once, in the dashboard

Open the Scheduled page and use Create. Write the instructions as you would a prompt, because they persist across runs, then pick a model, a cadence and a cap. Empty instructions cannot run.

Two choices are fixed at creation. trigger_type cannot change afterwards, so an API-only schedule never gets a cadence, and a cron schedule never becomes API-only. A cron schedule can also enable the API trigger and accept both, which is the useful setting for a series: the clock covers the normal week and your service covers an extra episode.

Pick the timezone deliberately. A cron slot with an IANA timezone follows that zone's clock, so a Monday 09:00 slot stays at 09:00 local through daylight saving changes, while a UTC slot would drift against your audience. Set the next run's expectations from cron.next_run_at on the schedule read, and compare it with what you see on the Scheduled page.

The fields that matter for a series

Keep episodes consistent

A series depends on the same look every week. Put the stable rules in the schedule instructions, or have the instructions call a saved Format so the recipe lives in one place. Use previous_run_id on a Format run when an episode should continue the previous one as another turn of the same thread.

Bind an output schema with a primary_output_key such as episode. A run whose schema is satisfied but whose primary output is empty ends as failed with primary_output_missing, so a missing episode does not look like a success.

Handle overlap and skips

A schedule defaults to skipping a firing when its previous run is still active. The skipped run is recorded with skip_reason: previous_run_active and next_action: retry_later. For a weekly job that is rarely hit, but a long video run that overruns can cause it, so alert on skipped instead of assuming the week was published.

Start an extra episode from code

With the API trigger enabled and a key holding actions:read and actions:write, send POST /v1/actions/{action_id}/runs with an Idempotency-Key. A body can lower the cap for that run with generation_spend_cap_usd, but never raise it above the schedule's own. Store the opaque aut_ id rather than the vanity path, because the vanity path changes if you rename the handle or slug.

Service-account keys cannot create schedule runs, and schedules owned by a team workspace are not available over the public API yet. Use a personal key for the call.