Recurring work
Schedules
Run deterministic scripts or judgment-based agent work on a cadence.
Choose the schedule runtime
| Runtime | Use it when | Example |
|---|---|---|
| Script | The work is fully determined and can be tested as code | Export a local report every night |
| Model | The run requires interpretation, synthesis, or a decision | Review new items and summarize material changes |
Each firing is an independent run
Every firing starts its own host thread with run history and failure state. Schedule edits affect future firings only. Keep durable outputs in agent-home, memory, or an App instead of relying on the temporary run context.
Operate recurring work
You do not write cron syntax or edit a schedule file. Ask the agent in Chat to create, change, pause, or repair a schedule, and it sets up the firing and the work behind it. The Schedules surface is where you watch what actually happened: past firings, their output, and any failures.
Example
Set up a schedule that checks our error tracker every weekday morning,
London time, and tell me what it will produce on each run.