A spend ceiling is a hard cap on what automated AI work is allowed to cost, enforced before the work runs. Budget is set aside ahead of each paid call and settled to the real cost afterwards, so the cap holds even when calls overlap.
Why it matters
Automated systems fail expensively. A retry loop, a runaway job, or one badly shaped request can turn a rounding error into a number you first see at the end of the month.
Checking a budget afterwards does not help. By the time the figure is visible the money is gone, and a cap that is only ever reported is not a cap.
The check also has to happen as one step. If two calls can each read the remaining budget before either one claims it, both go through, and the ceiling was never real to begin with.
How Jinn treats it
Before any paid AI call runs, Jinn checks it against daily and monthly caps and sets the budget aside, then settles to the real cost or releases the hold afterwards. The check and the reservation happen as one step, so two calls racing for the last of the budget cannot both slip through.
If the meter that tracks spending cannot be reached, the call is refused. A broken meter stops spending; it never allows spending that is not tracked. Spend is reserved and then settled rather than returned to the cent the instant a call ends, which keeps the ceiling conservative by design.
Every surface that spends your money passes one gateway, with caps checked before the call. See it
Related terms
The surprise bill is the failure mode this exists to prevent.
Caps checked before the call, budget reserved then settled, and a broken meter that refuses. See how spend is guarded.