Veri
Veri gives every AI agent a human-verified identity and a scoped, revocable set of permissions.
Demo
About
Veri is a permission and reputation layer for AI agents backed by a verified human. A person proves personhood once with World ID, then issues named agents under an ENSv2 subname registry they control ,each agent gets a scoped, revocable set of rights (read, transact, update its own endpoint, spin up sub-agents) via ENSv2 Enhanced Access Control, not a single binary trusted" flag. Every registration, grant, and revocation is indexed by a live subgraph, so any third-party service or agent can check, before transacting, exactly what another agent is allowed to do and whether a real human stands behind it.
How it's made
Contracts (Foundry, Solidity, Sepolia): VeriRegistrar wraps ENSv2's IPermissionedRegistry and IPermissionedResolver rather than reimplementing name management ,it issues agent subnames (<label>.veri.eth), enforces a 5-agents-per-human cap, and maps capability grants directly onto ENSv2's Enhanced Access Control role bits on each subname's resolver. One non-obvious decision: agents are keyed by labelHash, not token ID — ENSv2 token IDs can mutate across renewals, so token ID isn't a stable identifier for an agent's identity the way it would be in ERC-721/1155; labelHash is. The Graph subgraph: AssemblyScript mappings listen to VeriRegistrar and the underlying ENSv2 UserRegistry events and build a queryable graph of humans, agents, capabilities, and their full grant/revoke history ,including revoked capabilities, so an agent's past permissions stay visible even after a right is pulled. Backend (Express): the only component holding private keys. It verifies a World ID proof, derives humanId = keccak256(abi.encode(nullifier)) so the raw nullifier never touches the chain or the subgraph, enforces nullifier replay protection itself (World's API only checks proof validity, not reuse), and signs an EIP-712 attestation binding humanId to userAddress that the register contract checks on-chain, bridging an off-chain identity proof into an on-chain trust primitive without ever exposing the signing key to the client. Frontend (TanStack Start + wagmi): a directory of agents filterable by capability, a three-step registration flow (verify , choose label/capabilities , sign), and a per-agent page showing live capability status and history pulled straight from the subgraph via the backend.