Aegis Pay

AegisPay is a trust layer for autonomous AI agents that need to pay, sign, or execute sensitive actions on-chain.

As AI agents gain the ability to move value autonomously, there's no standard way to control when they should act alone, when they need human approval, and when an action should be blocked entirely. AegisPay solves this with a pre-action risk assessment engine.

Every agent action gets a risk score (0-100) based on target address, amount, and interaction history. A configurable policy applies a verdict — ALLOW, WARN, or BLOCK — and the result is recorded immutably on Hedera Testnet via AssessmentRegistry. If the verdict is WARN, a human operator approves via Ledger hardware wallet with ERC-7730 Clear Signing. If it's BLOCK, the transaction never executes.

The result is auditable, accountable, autonomous AI — with a trust layer that scales.

AegisPay is built on Hedera EVM (Testnet, Chain 296) using three Solidity contracts: AgentRegistry (ERC-8004 pattern for agent identity), PolicyManager (configurable ALLOW/WARN/BLOCK thresholds with Chainlink Price Feed integration), and AssessmentRegistry (immutable on-chain audit trail of every decision).

The frontend is Next.js + TypeScript + wagmi/viem, reading real on-chain data from Hedera via JSON-RPC relay. No mock data — every stat on the dashboard comes from chain.

Chainlink: PolicyManager integrates Chainlink AggregatorV3Interface for real-time HBAR/USD pricing, used in risk scoring. A CRE Workflow page demonstrates decentralized risk assessment running on a Chainlink DON simulation.

Ledger: When a transaction scores in the WARN zone, a human-in-the-loop modal triggers with Ledger Wallet Provider. ERC-7730 Clear Signing JSON files are included for all three contracts so Ledger devices display human-readable transaction details instead of raw hex.

ENS: Agent and operator identities resolve to ENS names in the sidebar and audit trail (mainnet transport for resolution, Hedera for settlement).

The autonomous agent (agent/src/agent.ts) uses Hedera Agent Kit natively for HBAR transfers and runs in dual mode — auto-executing ALLOW verdicts and pausing for human approval on WARN.