Understanding the cancellable window

Irreversible autonomous actions wait in a short cancellable window before they run. Where to see them, how to stop one — even from a phone notification — and what happens after.

Not everything the agent does can be undone afterward. Moving money, posting to the ledger, sending an outbound notice — once those land, they have landed. So Aptoria splits autonomous actions by reversibility: low-risk, easily reversible actions simply run, because undoing them later is trivial, while irreversible ones are held in a short cancellable window (about 15 minutes) before they execute.

During the hold, the action is decided but not done. That gap is deliberately yours: it is the difference between supervising an agent and merely reading about what it already did. Nothing about the window slows down reversible routine work — it applies exactly where a mistake would be expensive.

Pending actions show up in two places. On your home-screen queue, the "Happening now" group lists automatic actions still inside their window, each with a live countdown and a Stop control. The agent activity feed shows the same rows and has a "Pending (cancel)" filter so you can see everything currently waiting in one list.

You do not have to be at the app to intercept. The push and SMS alerts about a pending action carry a cancel link; tapping it opens the activity feed and fires the cancel immediately, without you hunting for the row. The link is one-shot — it cancels that specific action and cannot be replayed.

If you do nothing, the action commits when its window ends and shows as committed in the feed. If you cancel, it never executes: the cancellation and its reason are recorded, and the action shows as cancelled rather than quietly disappearing. Either way the record is complete — what was decided, what you did about it, and what actually happened.

Cancel and undo are different tools. Cancel stops a pending action inside its window, before execution. Undo reverses an already-committed action, where the action is reversible. Both live on the activity feed, and actions that genuinely cannot be reversed are labeled as such so you are never misled.

The window is wired into the rest of the safety model. The emergency stop voids everything currently sitting in a window in one motion, on top of pausing all autonomy. And your daily and monthly spend ceilings count actions still inside their window, so a burst of pending actions cannot slip past a limit before any of them has actually run.