Trinity Guardian
Human-authorized AI agent payments with on-chain permissions, spending policies, and risk checks.
Demo
About
Trinity Guardian is a security and authorization layer for autonomous AI agent payments. It lets humans safely delegate spending authority to AI agents without giving them unrestricted wallet access. Each agent receives an ENS-based on-chain identity and programmable spending mandate, including per-payment limits, daily limits, and destination allowlists. Before a payment is executed, Guardian validates the agent's ENS permissions and passes the transaction through a risk checkpoint. Sensitive transactions can require explicit proof-of-human approval through World ID before funds move. The result is a layered authorization pipeline where autonomous agents can make routine payments while owners retain control over high-risk or out-of-policy actions. read more details: https://ethglobaltokyo2026.github.io/trinity-guardian-presentation
How it's made
Trinity Guardian is built with Next.js, TypeScript, Solidity, viem, Foundry, and SQLite, with smart contracts deployed on Ethereum Sepolia. ENSv2 provides the on-chain identity and authorization layer: each AI agent receives a hierarchical ENS subname whose permissions and resolver records define its spending mandate. Guardian reads these records before authorizing a payment. World ID is integrated through IDKit as a proof-of-human approval layer for actions that require owner intervention. Agent payments use x402, while Intercepta is designed as the pre-payment risk checkpoint. Because we had not received Intercepta API credentials by submission time, the demo uses simulated risk responses behind the same integration interface to demonstrate allow/block behavior. This creates a layered ENS → Intercepta → World ID → payment pipeline where each checkpoint can stop a transaction before funds move.