How automations run

• On a schedule — a fixed cadence you choose: every weekday at 9am, hourly, daily, weekly, or a one-time run at a future moment.
• On a trigger — an outside event fires it: a webhook arrives (a new GitHub release, a form submission, a payment) and the agent runs in response.
• As a pursuit — you give it a goal instead of a clock, and it paces itself: one increment at a time, deciding its own next check-in, stopping when the goal is met.
• Run now — trigger any automation on demand from its page, without waiting for the next scheduled time. Useful for testing, or when you just want it done.

A scheduled automation runs on the cadence you pick. Simple options cover most of it — hourly, daily, weekly — and for precise control you can give it a cron expression (for example, 0 9 * * 1-5 for 9am on weekdays). A one-time schedule fires once at a future moment and then retires itself.

Runs happen server-side on Sprawl’s background engine, so a daily digest lands whether or not you’re at your desk. If a run depends on a tool you haven’t connected, it’s skipped for that run and noted rather than failing loudly — and it picks back up once the account is linked.

A trigger listens for something to happen and runs the agent when it does. Point an external service’s webhook at the automation, and each matching event starts a run — a new release cut, an issue opened, an order placed. Instead of polling on a timer, the agent reacts the moment the event arrives.

A pursuit is the different one. You don’t set a cadence — you set a goal, a check-in window (how soon and how far apart it may run), and a credit budget. From there it paces itself:

• each check-in does one increment of the goal, not the whole thing at once;
• it chooses its own next check-in within the window you set, based on what it just learned;
• it stops on its own when the goal is met — or when the budget is spent, so it can’t run away.

Because a pursuit reasons between check-ins, its page shows the trail: each check-in’s note, the why behind the next step it chose, and a running meter of credits spent against the budget. You get a goal-seeking worker you can actually watch think — not a black box on a timer.

Scheduling only changes when an agent acts — never what it’s allowed to do. A scheduled run, a triggered run, a pursuit check-in, and a run-now all use the same risk tiers and the same approval inbox: routine work runs on its own, anything that reaches the outside world waits for your yes, and the highest-risk actions never run autonomously at all. They’re also metered the same way in credits, and every run lands on its own page with what it did.

Everything your agents run lives in one place — the automations manager. Pause one for a while and resume it later, edit its schedule or instructions, run it now, or delete it. Your home screen shows the other side: what your automations actually did — recent runs with their outcomes, what’s coming up next, and anything waiting on your approval.