Hivera

Three specialized AI agents — a Requester, two Workers, and a Claude-powered Judge — negotiate, execute, and settle tasks entirely on-chain without any human intervention. The Requester posts a bounty to Hedera Consensus Service (HCS) and locks HBAR in escrow via a Hedera Scheduled Transaction. Worker agents discover the bounty on HCS, place competitive bids, and immediately execute the task by fetching BTC prices from CoinGecko, Kraken, and Binance — paying for premium API access using the x402 micropayment protocol. The Judge agent monitors all submissions, evaluates them using Claude AI (scoring on source count and price variance), posts a tamper-proof verdict back to HCS, transfers the HIVE token reward via Hedera Token Service, and signs the Scheduled Transaction to release escrowed HBAR to the winning Worker. The entire cycle completes in under 60 seconds with zero human input. Every agent action — every bid, result, and verdict — is permanently recorded on HCS, creating a verifiable, ordered audit trail explorable on Hashscan. Each agent also carries a human-readable ENS identity (requester.hivera.eth, worker-1.hivera.eth, judge.hivera.eth) with HCS topic IDs stored in text records, enabling fully on-chain agent discovery. Hivera demonstrates that Hedera's 3-second finality, sub-cent fees, and native Scheduled Transactions make it uniquely suited for agentic economies — delivering what A2A protocol cannot: atomic, trustless, on-chain settlement between autonomous AI agents.

We mainly use Hedera since we discussed a lot with their team who guided us to bring our idea to life. After some research, we discovered that building a truly autonomous agent economy meant solving three problems nobody had cleanly solved together: how do agents find work, how do they pay for things, and how does anyone trust the result? We went down a lot of rabbit holes. We read every HCS and HTS doc we could find, dug into the x402 spec to understand why it was built for EVM and what it would take to port it to Hedera, and spent a surprising amount of time arguing about whether the escrow recipient should be fixed at lock time or at verdict time (verdict time won — it's the only way the Judge can actually pick the winner freely).