Plans are priced on model spend, because that is what actually varies — a text run costs a fraction of a cent and an image run costs many times that. The enforced cap, the kill-switch and the full decision record are not upsells; they are on the free tier too. Paying buys room to run, not permission to be safe.
USD, billed monthly, cancel any time. No team plan yet — Prompt Tornado is single-account today. If you need seats with separate budgets per person or per client, tell me and it moves up the list.
Most tools alert you after the spend. This blocks the call before it happens.
Every model call — text, image, audio, video — goes through one dispatch path that reserves budget first. If the reservation fails, the call is never made.
Reservations are atomic in the database. Simultaneous runs starting at the same instant cannot each read "budget remaining", each see room, and both proceed.
Image, audio and video steps are estimated per generation rather than as text, so the amount held against your cap matches what the run actually costs.
Written down so you can decide with the real picture. If any of these is a dealbreaker, don't pay yet — tell me and I'll say when it lands.
There is no multi-user mode, so no per-person or per-client budgets. The spend cap covers the whole account. This is the most-requested gap and the most likely next build.
Sign-in is email or social, and there is no audit report to hand a security team. The architecture is public and the run record is exportable, which is not the same thing.
PII and secret detection runs on outbound prompts and records what it found. Masking before send exists behind a flag and is off in production.
Built by one person, in public, with a public changelog you can use to judge the pace. Support is direct email and usually same-day — the upside of the same fact.
Point it at a workflow that already costs you money and read the trace.