AI that answers the same way twice.

A model’s answer, posted on Solana with a bond. Anyone can run it again.

Latest receiptReading the network

An answer is posted as a receipt:
one model, one input, one output, and a bond that is lost if a second run gives other bits.

Anyone can be that second run. A browser tab is enough.

Why every machine gets the same bits

Reading the network

Answered on the network

Receipts

Not live

Reading the network

A dispute ends in one step

When two runs differ, the sides take turns keeping the part they disagree on. What is left is a single step, and the program on Solana executes it.

Reading the network

What stands behind an answer

0.00SOL

The bond, locked with the answer

Behind every model answer. A run that proves it wrong takes half, and the other half goes to the protocol.

00

The window, then final

How long anyone has to run the model again and challenge. It belongs to the model and never changes.

1step of

A challenge comes down to one step

It is all the chain ever computes. Its inputs and proofs fit one 4,096-byte transaction, and the program executes it itself.

What a final receipt does

A receipt nobody knocked down in its window is final, and any Solana program reads it.

  • Resolve a market

    A market program reads the judge’s final receipt and pays out on its output.

  • Gate a payment

    A payment releases only against a final receipt for the work it bought.

  • Audit an agent

    An agent’s model call is a receipt anyone can run again.

From a client
const receipt = await laplace.ask('judge', { claim, evidence })
const answer = await laplace.final(receipt)
From a program
let answer = laplace_receipt::read(&ctx.accounts.receipt)?;

Run it again.

Open a receipt, press once, and this browser runs the model.