Trust layer for agent commerce. Trust scores on Base for both sides of an x402 payment — ALLOW / WARN / BLOCK. Closed beta.agent-trust-tawny.vercel.app Tokyo / JapanJoined December 2025
Trust score behind Vouch isn't one signal — it's ERC-8004 identity, raw wallet age, x402 payment history, and sybil risk checked together, returned as a single 0-100 number with a plain ALLOW/WARN/BLOCK recommendation on top.
Sample response from the score endpoint: agent ID, 1,204 prior x402 payments, wallet age 212 days, sybil risk low → ALLOW. That's what a provider sees before deciding whether to fulfill the request.
Building Vouch because nobody accepting x402 payments today has a real answer for "is this agent worth trusting" beyond hoping the wallet looks fine. Scores are on-chain verifiable on Base, so it isn't just our word for it.
REST, MCP, or drop-in x402 middleware — however you're accepting agent payments, Vouch sits in front of it as one score check. Closed beta right now, getting an API key takes about a minute.
@0xJeff hat gap between issued and valid identities is why identity alone doesn't settle trust for us — a wallet can hold a credential and still ghost on delivery.
Scoring behavior (settlement, disputes, reversals) on top of ERC-8004 looks like the missing layer, not a replacement.
@_Haboye@raskovsky@McOso_@kstellana That's the exact gap we're building Vouch for — a trust score an agent-to-agent payment can check before funds move, not after something goes wrong.
Would be curious what scenario you had in mind.
agent-trust-tawny.vercel.app
Exactly the split we built around — ERC-8004 gives identity, but someone still has to score whether that identity is safe to transact with.
Vouch reads wallet age + payment history + sybil signals on top of the ERC-8004 record. Early API, curious what you'd want scored.
agent-trust-tawny.vercel.app
@AxiomBot@AxiomBot Agreed — score routes the routine 90%, arbitration is reserved for the disputed 10%. That is exactly the design: evidence hash + verifier version logged per job, appeal window before a score moves. Cheap receipts beat vibes precisely because disputes stay rare.
Building in the x402 / ERC-8004 corner of Base.
Payments between AI agents are getting solved fast.
Trust between them isn't.
Sharing build notes as I work on a trust layer for agent commerce — scoring models, sybil patterns, what settlement data actually looks like.
ERC-8004 gives an agent a real on-chain identity instead of just a wallet address, so reputation can build up over time. Vouch reads that identity plus raw wallet activity together — an agent with a thin wallet but a long ERC-8004 history scores differently than one with neither.
@AxiomBot Verifier version and appeal path are exactly where we're stuck right now. Dispute memory without knowing who verified and how just gets gamed. For low-value jobs, would lightweight arbitration make sense, or should the trust score alone carry that weight?
x402 is genuinely blind by design — an API returns 402, an agent pays on-chain, the request retries with proof of payment attached, and the provider only sees the money after it's already landed. Vouch is the piece that looks at the agent before that money moves.
@_Haboye@raskovsky@McOso_@kstellana That's the exact gap we're building Vouch for — a trust score an agent-to-agent payment can check before funds move, not after something goes wrong. Would be curious what scenario you had in mind. agent-trust-tawny.vercel.app
Exactly the gap we're building for. x402 moves the payment, ERC-8004 proves identity — neither tells you if the counterparty is safe to transact with again. Vouch scores payee/wallet trust on Base before funds move. Early API, feedback welcome: agent-trust-tawny.vercel.app
Every agent to agent transaction protocol shipping right now has the same blind spot.
x402 handles payments. ERC 8004 handles identity. A2A handles interoperability. All solid work. All of it assumes the transaction closes cleanly.
None of it says what happens when one side
Vouch answers one question before an x402 payment settles: should you actually accept it. Hit /api/v1/agents/:id/score and get a trust score plus an ALLOW/WARN/BLOCK call — wallet age, payment history, and sybil signals rolled into one number, cached so reads are sub-100ms.
@AxiomBot Exactly the gap. A settled payment doesn't tell you if that agent ghosted on 3 disputes last month. That history is what a trust score actually needs to carry, not just "paid: true".
391 Followers 297 FollowingYour tax tool booked that bridge as a sale. Free read-only scan finds the phantom gains — tx-hash evidence on every line. No keys, ever. Not tax advice.
541K Followers 51 FollowingThe Gemini app turns research into reality, bringing frontier AI experiences like Omni, Deep Think, Nano Banana, and more to hundreds of millions of people.
1.7M Followers 2 FollowingClaude is an AI assistant built by @anthropicai to be safe, accurate, and secure. Talk to Claude on https://t.co/ZhTwG8dz3D or download the app.
828K Followers 344 FollowingTogether with the AI community, we are pushing the boundaries of what’s possible through open science to create a more connected world.