Routines · Schedules · Monitors
It runs at 6am. It doesn't cut corners at 6am.
Work that repeats shouldn't need you to repeat it — or to babysit it. Routines are outcomes defined once and run unattended, with a quality floor that never bends for cost and a rule for trouble: escalate once, then park for you.
Three primitives
The work, the clock, and the tripwire.
Aghata separates what most tools blur into 'automation': the definition of the work, when it runs, and what's watching for the moment it should.
Routine
The work, defined once
An outcome with its tools and standards attached. Run it by hand today, on a schedule tomorrow — it's the same definition either way, and it absorbs what other products call custom agents.
Schedule
The clock
Attach a cadence and the routine dispatches itself — each firing a fresh, fully recorded session. Pause the schedule and nothing lingers; there is no daemon to forget about.
Monitor
The tripwire
A condition worth acting on — watched for you. When it trips, work starts as a session with the trigger recorded, instead of a notification you'll read too late.
Unattended, not unaccountable
The rules that hold when nobody's watching.
Automation usually trades quality for convenience the moment you look away. These are the guarantees that stop that trade.
It never goes cheap in the dark
Unattended runs are routed with a hard quality floor — never below standard, no matter what it costs. Cost-saving happens when you're watching and can judge; never when you aren't.
One retry up, then park
A run that fails verification escalates once to a stronger model. Still stuck? It parks with everything it has and notifies you. No confident nonsense at dawn.
Logged in with nobody awake
Grant a vault credential to the routine itself and its runs authenticate unattended — every use still ledgered, high-risk origins still reserved for humans.
Consequences still wait
A scheduled run that reaches an external send or a destructive step pauses for approval like any other session. The calendar doesn't outrank your judgment.
Nothing hums between runs
Schedules dispatch sessions; they don't keep agents warm. Between firings your routine is a definition, not a process.
Every run, inspectable
Each firing lands on your Tasks page with its plan, steps, artifacts, cost, and the trigger that started it. Automated never means invisible.
The standard
Every automatic run is still a session you can open.
Questions
What people ask about unattended work.
What exactly is a routine?+
A repeatable outcome definition: the goal, the tools it may use, and the standards it must meet. Where other products have you build a 'custom agent,' Aghata has you define the work once — then run it by hand, on a schedule, or when a monitor fires.
Does scheduled work get a cheaper model when nobody's watching?+
No — the opposite guarantee. Unattended runs carry a hard quality floor: routing never drops below standard quality for cost. The report you didn't watch being made is held to the same bar as the one you did.
What happens when a 6am run hits a problem?+
It escalates once automatically, and if that doesn't clear the problem, it parks and notifies you — with its progress and evidence intact. You never get a confidently wrong report because nobody was awake to say 'that's off.'
How does it log into things at 6am?+
With routine-scoped credential grants. You grant a vault credential to the routine itself, so its scheduled runs can authenticate with nobody at the keyboard — and every use still lands in the ledger. High-risk sites still require a human.
Can I see what a scheduled run actually did?+
Every run — manual, scheduled, or monitor-fired — is a normal execution session on your Tasks page: same visible plan, same step record, same artifacts in the Library, plus a link back to the routine and trigger that started it.
Chapters of one system
Give it the outcome. Keep the control.
Start free in the product, or talk to us about how your team would use it.