Appearance
Recurring tasks
Recurring work is modeled as a parent task that holds the schedule and durable instructions for repeated work.
The parent is the long-lived record. Each scheduled or manual run creates a child instance task where the agent does that run's work.
Parent task
A recurring parent task must be a top-level task. It stores the schedule, per-run limit, and durable instructions such as inputs, output format, success criteria, and reporting expectations.
The parent does not execute the repeated work itself. It stays open while recurrence is active and keeps the durable history across runs.
Schedule and controls
Recurring tasks can run on a daily UTC schedule or on a fixed interval. New recurring parent tasks start paused so the instructions can be reviewed before unattended runs begin.
From the task details, a user can edit the recurrence, pause or resume it, disable it, and see the next run time. If a scheduler attempt fails, the task shows the latest scheduler error.
Each run has a recurrence limit in USD. When no explicit limit is set, the task uses the room default.
Instance tasks
Each scheduled run creates an instance task with its own conversation and execution rules. The instance is the workspace for that single run: it records progress, decisions, changed files or artifacts, and the final run summary.
Important durable results should be summarized back to the parent task before the instance closes.
Manual run
Users and agents can run one recurrence instance immediately with Run now. This is useful for a first checked run after setup, or for an extra run outside the normal schedule.