Agent payments, AP2

Let your agent check out for people, inside the mandate they gave it

Under the Agent Payments Protocol, a person signs a mandate that says what the agent may buy, from whom, for how much, and until when. That lets the agent finish a purchase while nobody is watching. VeriProof checks each purchase against the mandate and against your own business rules, then sends anything outside them to the person who can approve it.

Two sets of limits, checked together

A typical setup denies a purchase that breaks the mandate and sends one that fits the mandate but not your rules to a person. The customer has said yes; your business hasn't yet.

When a purchase falls outside

An AI booking agent reserves a hotel for a customer under a mandate capped at $900. The room fits, but the hotel is a merchant your company has never paid, so your rule sends it to the travel desk lead. She's notified in Slack or by email, with a reminder if it sits, and sees the booking, the mandate, and the merchant details in VeriProof. She approves it, the agent completes the payment through your provider, and the provider's transaction reference joins the record.

Months later, when finance asks why the agent booked with a new merchant, the answer is one record: the mandate, the rule version, her approval, and the reported payment.

What stays the same for your developers

Your mandate service, wallet, signer, and payment provider stay in place. The application wraps the payment call in a policy check from the VeriProof core SDK, so a denied or paused purchase never reaches the provider. The SDK names the AP2 fields for you: mandate type and status, whether the person was present, and how deep the delegation chain runs. No separate AP2 package is needed.

Other ways agents pay: x402, pay.sh and MPP, and stablecoin supplier payments.