/ ABOUTProtocol foundations

KYC, but for the software placing the order.

KnowYourAgent (KYA) resolves an AI agent to the business running it, validates the signed authorization its principal issued, and keeps the record that settles the argument later. Two API calls at your checkout edge, plus a side-channel for when something goes wrong.

Line illustration of operator identity and KnowYourAgent trust foundationsKYCKYA/ foundationoperator trustKYA
/ 01Our mission

Who authorized this purchase?

An agent arrives at your checkout with a card and an order. Nothing in your stack can tell you who told it to buy, what they capped it at, or who answers for it when the order is wrong.

KYA answers all three. The operator registers under KYB and issues a signed trace carrying the principal hash and spend scope. The merchant resolves the agent and validates that trace at checkout. Both sides keep a record they can pull into a pre-dispute case. Banks did this for people and called it KYC. Agents need the same thing, minus the assumption that there is a person on the other end of the session.

  • AID
    Operator-linked identity

    Every agent ID resolves to a KYB-registered business, not an anonymous session.

  • TXG
    Decision-ready checks

    /api/v1/verify returns accept, review, or decline. /api/v1/checkout/sessions fails out-of-scope orders with a named reason code.

  • LOG
    Traceable records

    Verification IDs, trace IDs, and checkout-session records stay retrievable months later, when someone finally asks.

/ 02The problem

The trust gap.

Three things break when the buyer is software. None of them are fraud problems, which is why fraud tools miss them.

/ 01

No operator on file

An agent presents a session, not a company. Nothing in the request tells you which business runs it, or whether that business exists.

/ 02

No shared history

An agent that burned three merchants last week arrives at the fourth looking brand new. There is no standard record of how it has behaved.

/ 03

Review queues that never clear

Every unrecognized agent order becomes a manual review. That works at ten orders a day and collapses at ten thousand.

/ 03Core values

What we optimize for.

How we settle it internally when two designs both work.

/ 01

Transparency

Every decline names its reason. An operator with a low score can see which inputs produced it and go fix them.

/ 02

Privacy

The authorizing principal travels as a hash, never in cleartext. Verification sees what it needs to decide and nothing else.

/ 03

Neutrality

No model vendor, gateway, or commerce platform gets a better answer than any other. Same contract for everyone.

/ 04

Performance

Two HTTP calls on the authorization path. KYA sits beside your fraud stack, not in front of it.

FAQ

About KYA

What is KnowYourAgent (KYA)?

KnowYourAgent is a verification protocol for AI agents in commerce — the equivalent of Know Your Customer (KYC), but for autonomous software agents. Each agent registers against a KYB-verified operator and carries a signed, revocable ID. Merchants resolve that ID at checkout and get back a recommendation, the reasons behind it, and a record they can cite later.

Why do merchants need to verify AI shopping agents?

AI agents don't carry the signals fraud tools were built to read — no cookie history, no browser fingerprint, no velocity profile that looks human. So those tools block legitimate agent purchases and still miss the bad ones. KYA answers the two questions the fraud stack can't: which business runs this agent, and what did its principal authorize it to spend?

What problem does KnowYourAgent solve?

Merchants have no reliable way to tell a legitimate AI shopping agent from a scraper or a credential stuffer — the traffic looks identical. KYA links each agent request to a registered operator, a trust score with visible inputs, and the signed trace the principal issued, so the checkout decision runs on evidence instead of a guess.

Get started

Start with one agent.

Register an operator profile, add your first agent, and issue a signed trace against it. The API reference has the request shapes. We onboard by hand while the platform is in its design-partner phase, so expect a real person on the other end.