Cost control

Cap AI spend before a workflow runs

Most AI tools tell you what you spent after the invoice arrives. A multi-step workflow makes that worse: one prompt can fan out into research, writing, image and voice calls across several providers. Prompt Tornado checks your spend cap before a run starts, not after.

What happens before every run

Your month-to-date spend is added up from the recorded cost of this month's runs.
The run's cost is estimated from its plan: which steps, which models, and those models' current prices.
The estimate is reserved against your cap in a single step, so two runs started at the same moment can't both slip under the limit.
If the run would take you over a blocking cap, it doesn't start. You see which cap stopped it, and when it resets.
When the run finishes, the reservation is settled at the actual cost, step by step.

Three kinds of cap, and how they combine

Cap	Who sets it	Example
Plan allowance	Your plan	Free: $5 a month. Pro: $10 a month.
Your limit	You, in Settings → Budget	$50 a month, set to warn only
Operator limit	Prompt Tornado, across all accounts	A platform-wide safety ceiling. None is set today.

Caps don't add up. Each applies at its own limit, and the lowest blocking cap is the real ceiling. A warn cap lets the run through and flags that you've passed it. Spend is counted per calendar month, in UTC.

What a run actually costs

Every step records its model and its cost.

Workflow	Steps	Typical cost
Multilingual Product Launch (copy, voiceover, image)	4	about $0.04–0.07
Support Inbox Action Plan	4	about $0.12
Billing Bug Fix & Independent Review	4	about $0.29–0.36
Vendor Decision Briefing (web research + five steps)	6	about $0.44–0.58

Every step records the model that ran it and what it cost, priced from the same model registry the router uses. These are the four example workflows in the app, measured on production runs in September 2026. Their prompts and full outputs are on GitHub.

Where a spend cap can't help

An estimate is an estimate. If a model writes more than expected, the actual cost can be higher than what was reserved. The difference is settled when the run finishes and counts against your next run.
A run with no cost estimate reserves nothing up front. It can't be blocked before it starts, but its real cost still counts against every run after it.
If the budget system itself is unreachable, paid plans proceed rather than being blocked, and the run is recorded as degraded. The Free plan fails the other way: it blocks.

Frequently asked

Can I set a limit lower than my plan's allowance?

Yes. Your own limit can only be stricter than the plan's, never looser.

Does a warn cap stop runs?

No. It flags that you're over and lets the run continue. Only a block cap stops runs.

When does month-to-date spend reset?

On the 1st of each month, in UTC.

Know the cost before it runs.

The Free plan includes 10 runs and $5 of model spend a month.

Start Free Try Live Demo