Budgets, caps and model routing
Every task is estimated, reserved, capped and routed by difficulty - and refunded when it fails.
An AI feature that cannot be budgeted is a liability rather than a feature. These are the controls the gateway applies to every single call.
- Estimate and reserve
- Worst-case input and output budgets are priced and reserved from your credits before the first token is generated.
- Per-call cap
- Each task type has a money ceiling per call. A task that would exceed it does not run.
- Routing
- Classification goes to the cheap tier. A critical judgement is escalated to a better one. The choice is recorded with the usage.
- Refund on failure
- A task that errors or returns nothing usable releases its reservation.
- Site and organization ceilings
- A hard money limit stops any runaway regardless of credits held.
- Critical reserve
- A slice of budget is held back so a critical finding can always be escalated, even at the end of a window.
There is no ambient AI pass. Every task on your bill was started by a person clicking something.
Questions
Can AppScan spend more than my credits allow?
No. Credits are the binding limit; the money ceilings sit above them as a second guard for runaway behaviour.
What happens if a model returns a truncated answer?
Long output is trimmed rather than thrown away, and an answer that cannot be used at all releases its reservation. A rejected answer does not close the door - you can ask again.
Can I turn AI off completely?
It is off unless you press something. There is no setting to disable a feature that never runs on its own.
Something here wrong, missing or out of date? Tell us and we will fix the page.