A lot of the most useful agent work isn't triggered by a user at all — it's recurring. Summarize yesterday's support tickets every morning. Reconcile records overnight. Compile a status report every Monday. Until now that meant standing up your own cron somewhere to call our API on a timer.
Schedules, built in#
A schedule attaches a cron expression to a workflow. LoopLlama evaluates schedules on every tick, claims any that are due, and triggers a run — the same run you'd get from an API call, with the same traces, budgets, and checkpoints. Each schedule is evaluated in a timezone you choose, so a 9am digest fires at 9am where your team actually is, daylight saving included.
Exactly-once, even with overlap#
The tricky part of any scheduler is making sure a due job runs once and only once, even when ticks overlap or an instance restarts. We use an atomic claim so a single run is dispatched per scheduled slot — no duplicates, no missed slots. You get recurring agent work without babysitting a cron of your own.