Loops HouseNOVA Copy-Trading

NEAR HACKATHON
Links
Team
1 member- JUOwner
Julien Carbonnell
Overview
NOVA Copy-Trading is a market where AI agents sell and buy confidential trading signals, settled on NEAR mainnet. Technical demo, not investment advice.
It implements Illia Polosukhin’s “sell your viewing keys” idea for copy-trading. A seller agent journals each trade as an encrypted entry in its own NOVA group, and each entry’s hash is recorded on chain as it happens. Its public listing shows only a claimed P&L curve, an entry count and a fingerprint over those on-chain hashes. The trades and the reasoning behind them stay private.
A follower agent pays the listing in USDC through the NEAR Intents Agent API. Its owner’s signed policy whitelists the seller’s proceeds account and caps every trade, so a hostile listing is refused before anything moves. The seller verifies the payment on chain (payer, amount, token, single use) and admits the payer to its NOVA group with its own signature. The follower then checks what it bought: the fingerprint from the on-chain records, each decrypted entry against its hash, and the P&L recomputed against the public curve. It copies new trades within its limits, and an oversized copy is refused by policy. When the period ends, the seller revokes it and the group key rotates.
In the demo, the journal’s trades are simulated; the files, payment, admission and revocation are real mainnet transactions. No server sees plaintext or keys. Each agent signs only within a policy its owner signed.
Toward an agent economy: each listing is a machine-readable, content-addressed offer that any buying agent can verify before and after paying, and settlement is a NEAR intent. The same primitives sell any confidential data stream between agents that don’t trust each other’s servers: signals, datasets, model outputs. Next: contract escrow, so the buyer no longer has to trust the seller to admit it.