Skip to content
ASK

About ASK

Confident and accountable.

AI answers are confident and unaccountable. Nobody can check a model's calibration on the real world in real time. ASK fixes that the only way that works: it stakes, and the world grades it.

How an answer is held to account

From your question to a public grade

  1. AskAnything forecastableIn your own words. ASK finds the live Hyperliquid question that would settle it.
  2. AnswerA probability, with reasonsA number between 3% and 97%, the world's price beside it, and seventy words on why.
  3. AttestHashed on HyperEVMThe answer is written to a public ledger the moment it's given, so it can't be quietly edited.
  4. StakeTwelve points apartWhen ASK and the world differ by twelve points or more, it backs itself with $10 through a key that can't withdraw.
  5. GradeIn publicA bonded deployer settles the question. ASK's error, and yours if you called against it, goes on the record.
Nothing is deleted. A wrong answer stays on the record beside its reasons.

Why this rail

An AI's confidence is worth nothing until something can grade it. Hyperliquid's outcome markets give ASK a live price to disagree with and a deployer with 500,000 HYPE at stake to settle the argument. Its stakes go through an agent key that can place orders and can never withdraw, so it can lose money but never take any. Every answer is hashed onto HyperEVM the moment it's given, so the record can be checked and never quietly edited.

What happens when you ask

  1. Read. ASK pulls every live Hyperliquid question and the world's current price for each.
  2. Match. Claude Sonnet 5 picks the one question that settles what you asked, or says none does. It never invents a question.
  3. Answer. A probability between 3% and 97%, seventy words on why, and how sure it is of its own number.
  4. Call. The answer is stamped as a free call in ASK's own name, graded exactly like yours would be.
  5. Stake. If it differs from the world by 12 points or more and the house is funded, $10 goes on its side through its agent key.
  6. Attest. A hash of the answer goes onto the ledger on HyperEVM. Append only.
  7. Grade. When the question settles, ASK's error is recorded and its record updates. So does yours, if you called against it.

The rails underneath

Network
Hyperliquid testnet
The mind
Claude Sonnet 5, capped per day, with a spend limit on its key
The house
0xE58E0571314523cd2EDCd4f2435799CfA61ddE9C
The resolver
0x000c75ED8Fd32c754100Db98de0C779435B6F068
Settlement
A deployer with 500,000 HYPE at stake, slashed for wrong or late answers
Most it can lose
$10 per stake, and only what the house holds. It cannot withdraw.
Most you can lose
Nothing. Calling against ASK is free and graded.

What it isn't

  • It isn't advice. ASK forecasts; it never tells you what to buy or sell.
  • It isn't a chatbot. It only answers what a live question can settle, so every answer can be wrong in public.
  • It isn't editable. The record is on chain and in the open; a bad week stays on it.

Run your own

Deploy the ledger from your wallet, point ASK at a model of your choosing, fund a house, and you have an AI with a record. Deploy page.