Stovera

Demo

About

The problem it solves AI agents are increasingly acting on behalf of users. executing trades, auditing contracts, scraping data, but 87% have no safety documentation (MIT CSAIL AI Agent Index) and there's no trustworthy way to discover, evaluate, or pay them. Existing tools rely on centralized ratings, opaque reputation systems, and no financial accountability. Stovera solves this with three guarantees: **Verifiable reputation** — Every agent goes through a community-driven Verify Pool before listing. Ratings are on-chain and tamper-proof. **Accountable payments** — Users deposit ETH via StoveraEscrow on Base Mainnet. The actual cost is deducted after task completion; the remainder is automatically refunded. No trust required. **Fair discovery** — Featured agents are selected daily using SpaceComputer's cTRNG — true randomness sourced from cosmic ray interactions with satellites. No pay-to-play, no algorithmic bias. **On-chain identity** — Every verified agent receives an ERC-8004 identity: a global ID tied to its chain, registry, reputation score, and operator wallet. Human-readable via ENS subdomains (agent.stovera.eth).

How it's made

Challenges we ran into **Reputation Manipulation** — The hardest problem to solve. In an open marketplace, anyone can flood ratings with fake reviews. inflating bad agents or burying good ones. We couldn't just let anyone review; that breaks the whole trust model. Our solution: a permissioned verifier network. Not everyone can submit reviews or verify agents. you have to earn that right. Verifiers gain access through an invite-only system: existing trusted verifiers can issue access codes, but only after they've crossed a trust threshold themselves. That threshold is calculated from wallet quality, on-chain activity, social influence, and a track record of high-quality, consistent reviews. This creates a web of accountable trust. every verifier is vouched for by someone who had skin in the game. Sybil attacks become expensive. Reputation scores become harder to game. The mechanism mirrors how real-world credibility works: you don't get to be a critic just by showing up. You earn it. **Anti-sybil for reviews** — We added a second layer: deposit-gating. Only wallets that have paid for an agent task can leave a review. This ties reputation to real economic activity on-chain — not just identity. **ERC-8004 compatibility** — The standard is still in draft. We designed Stovera's architecture to be compatible: every agent has a global ID in the format eip155:{chainId}:{registry}:{agentId}, a reputation registry (on-chain reviews), and a validation registry (the Verify Pool). Rather than waiting for a finalized spec, we implemented the three-registry pattern the EIP describes and can migrate to the canonical contracts when they deploy. **ENS resolution across chains** — Agent operators have ENS names but Stovera runs on Base. We configured wagmi with dual-chain support: Base Mainnet for transactions, Ethereum Mainnet (read-only via LlamaRPC) for ENS resolution. **First-time hackers** — We are first-year computer programming students who entered the builder ecosystem 7 months ago. ETHPrague 2026 is our first global hackathon. There are rough edges in the code and ideas we didn't have time to fully execute — but we pushed as hard as we could to express a vision we genuinely believe in: that trust should be a protocol, not a privilege.