I build agents that can spend money and finish a job without waiting for a person at every step. The part I care about most is making that autonomy safe: budgets, x402 settlement, and proof of what happened onchain.
Top 4 finalist, Tether Developers Cup
A peer-to-peer football betting app with no bookmaker or central server. Friends post a bet, match each other's USDT stake, then three on-device QVAC models decide the result and release the pot onchain.
Round of 32, 0G Zero Cup
An AI bouncer for NFT allowlists. It interviews applicants inside a TEE, records attestations on 0G, and exports a Merkle root for EVM mint contracts.
Top 100, QVAC Hackathon
A peer-to-peer marketplace where people sell spare on-device inference. Work is metered by token and settled in stablecoins between peers.
Hackathon winner
An AI bounty hunter that finds open-source bounties, works out which ones are worth pursuing, and handles the claim. Payment settles over x402.
A research agent that pays for its own work up front and bills cost plus margin when it delivers. Spend caps and balance checks stop it from running past its budget.
Undercollateralized lending with encrypted credit scoring. Zama fhEVM keeps borrower data private while the protocol makes a credit decision.
I keep coming back to the same question: how much freedom can you give an agent without losing track of its money or decisions? That is why the plumbing matters to me as much as the demo.
TypeScript 路 Python 路 Solidity 路 Next.js 路 x402 路 Solana 路 EVM 路 0G 路 Zama fhEVM 路 TEE attestation 路 Circle Agent Stack