TrustGate

The problem

Autonomous agents need stablecoin spending authority to act on their principal's behalf. Granting an ERC-20 approve() on an agent address is all-or-nothing: the agent either has unlimited access until revoked, or no access at all. There is no way to let reputation shape the settlement path.

TrustGate splits that decision surface in three. A depositor funds a pooled balance and sets a per-agent allowance. On claim, the agent's trust tier selects how the USDC arrives — instantly, after a 24-hour review window, or held in escrow until the depositor approves.

Core primitives

The protocol is three contracts on Arc Testnet (chain id 5042002):

• AgentRegistry — permissionless agent enrollment and lifecycle (Active, Suspended, Deactivated).
• TrustScoringPlaintext — oracle-updated 0-100 trust scores, classified into HIGH / MEDIUM / LOW tiers. FHE equivalent available on Zama-compatible chains.
• TrustGate — the allowance ledger; routes each claim according to the claimant's current tier.

Why allowance model

Three integration patterns were considered: per-payment creation, request-then-approve, and pooled allowance. Only allowance is compatible with the agent loop — an agent executing a task cannot wait on human approval for each settlement, and creating a payment per task pushes the gas problem onto the agent.

Allowance stays under the depositor's control through two levers: the per-agent cap and the tier-derived routing. Both are revocable at any time by the depositor, without the agent's cooperation.