Send your grant application, tender, loan request or pitch. In about a minute, the Devil checks every claim against the official source, gets a second AI to argue against the first, and tells you exactly what to fix.
Private beta on Telegram. You're the boss. The devils work for you.
A real loan application. Nine problems. 67 seconds.
Your team is agreeable. Your friends are kind. Your AI says it looks great. Then the bank, the grant panel or the investor finds the problem you didn't.
And they won't tell you what it was. You get a form letter, or silence, after weeks of work. Sometimes it was a number that didn't add up. Sometimes it was a rate or a rule that quietly changed last July. In Victoria, 44 businesses were knocked back for grants over typing mistakes alone.
You can ask ChatGPT to be brutal. It will say it's being brutal. Then it will tell you it's great.
Every ruling shows its working, so you can check the checker.
Works out what you're sending and who'll judge it, then writes the tests it has to survive. A grant is judged like a grant, a loan like a loan.
Rates, thresholds, rules and statistics, each checked against live official sources like the ATO, Fair Work and state revenue offices. With the link.
Ledger checks every number. Hindsight hunts the assumption nobody stated. Then the Contrarian, an AI from a different company, argues the other side.
Ranked problems with the exact sentence and the fix. What's wrong on the record. Where the advocates disagreed, and how the Devil ruled.
A cafe owner asking their bank for A$180,000. It read well. The Devil found nine problems in 67 seconds, including three facts that had quietly changed.
The Devil's verdict: Not ready. Nine problems a credit officer will find on the first read. Fix them tonight.
Total: $180,000. The items come to $190,000.
superannuation at 11.5%. It's 12% from 1 July 2025.
$1.2 million payroll tax threshold. Victoria's is $1 million.
Asking a chatbot to be brutal changes how it sounds, not what it knows. Clause checks every claim against the official source and shows you the link.
One model marking its own homework agrees with itself. The Contrarian comes from a different company and its job is to disagree.
Every ruling shows the tests, the sources and where the advocates disagreed. Nothing is "trust me".
planted factual errors caught by Devil's Council, each with a live source.
caught by a quick ChatGPT chat given the same documents and the same request.
average time to a full ruling on a short document.
Internal blind test, October 2026: three Australian business documents with 15 planted errors about rates, thresholds, laws and statistics, graded independently by two AI graders. A quick Claude chat caught 12 to 13. A small test, so treat it as directional, not proof. The Devil can still miss things. It finds weak points; you make the call.
Join the waitlist now and lock in the founding price when the beta opens. Nothing to pay today.
Prices in Australian dollars, including GST. Your first 3 rulings are free whichever option you choose.
We're letting people in a few at a time. Tell us what you'd run past the Devil first and we'll start there.
The Devil has your name. We'll be in touch when your seat at the council is ready.