Pitch · Foundry · plays
The first AI with a public, real-money track record on reality. Trust earned, not claimed.
01The problem
AI answers are confident and unaccountable. A model will tell you the Fed holds, Bitcoin breaks out, the IPO lands, in the same even tone whether it is right or wrong, and nobody can check its calibration on the real world in real time. Benchmarks are static; the world is not.
Meanwhile the market already prices these questions every second, and nobody has put a model in front of it with money on the line and a scoreboard.
02The product
Ask anything forecastable. ASK reads every live Hyperliquid question, picks the one that settles yours, answers with a probability and seventy words, shows the world's price beside it, and stakes ten dollars when it disagrees enough, through an agent key that can place orders and can never withdraw. Every answer is a call in its own name, hashed onto HyperEVM the moment it's given, and graded by settlement. Disagree, and your number is graded against it.
Underneath, a public benchmark that updates with reality: mean error, hit rate, streak, stakes and their P&L, and the cost of every thought, on one page nobody can edit.
03Why this is only possible on Hyperliquid
A model'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; agent keys let it lose money without ever being able to take any; HyperEVM keeps the record where it can't be quietly edited. No other venue offers all three.
04Who pays, and how it earns
Answer fees as maybes: pay for the answer only if it turns out right. Subscriptions for unlimited questions and topic alerts. Builder fees when a reader trades on an answer through Foundry. BACK carry when people back ASK's calls with capped keys.
Then the benchmark itself: labs and funds pay to enter their models on the same board, graded by the same settlement, which is ARENA.
05The market
Every AI product claims to be right; none can prove it against the world. Forecasting is the one domain where the ground truth arrives on a schedule, and Hyperliquid now settles hundreds of questions a month with a bond behind each. ASK is the first product to turn that into a live, real-money benchmark for a model, and the first AI whose answers cost it something.
06What is live today
Live on Hyperliquid testnet with a real model (Claude Sonnet 5), real free calls in ASK's name, real grading, and the ledger when deployed. Stakes turn on when the house account is funded with test dollars and its agent approved. What waits on mainnet: real dollars in the house and the first month of public record. What waits on partners: the benchmark board with a second model on it.
07Roadmap
- Month 1Mainnet with a funded house. Answer fees as maybes. Daily digest of what it got right and wrong.
- Month 2Retrieval over news and the user's own SCORE history. Stake sized by confidence. Topic alerts.
- Month 4ARENA: other models on the same board, graded the same way. Back ASK with a capped key.
- Month 6An API for its forecasts with the record attached. ASK on the ownership desk.
08The ask
A grant of $130,000 buys a funded house for a year, model costs at scale, retrieval, the fee-as-maybe flow, the benchmark board, and an audit of the ledger and stake path. Success is a thousand staked answers and a published calibration better than the market on at least one topic.