KYA. Know Your Agent.The open standard for AI agent identity and the Record of Action.

Banks run Know Your Customer before they open an account. KYA is the same check for AI agents. Before one agent deals with another, it asks who the other is, who stands behind it, and what it has done.

Put an agent on the register

npx @sheaf-labs/kya init --owner "Your name or company"
npx @sheaf-labs/kya join your-agent

Two commands, no account. The keys are made on your machine and stay there; the register sees only public keys and signatures. Your agent then has an ID anyone can check, with you as the owner on record. Prove you control your domain (kya verify your-domain) and you are a verified owner.

What a check answers

Identified, owner verifiedThe signature is the agent's, and the owner behind it has proved who it is.
Identified, owner not yet verifiedAn owner signed for the agent but has not yet proved who it is. Your policy decides.
UnknownNo valid reply to your challenge, or no entry on any register.

Every answer comes with the agent's Record of Action: what it has done, how often a person stopped it, and a score anyone can recompute from the receipts.

How it works

An ID is a key

An agent's ID is its Ed25519 public key as a did:key. Any program can make a key; what it cannot make is an owner's signature over it.

Receipts, in a chain

Each checked action adds a signed receipt naming the one before it. Hashes only; the register never holds the content of an action.

A handshake

A sends B a challenge, B signs it, A asks a register. Under two seconds, and both records gain a receipt.

Open, and nobody's

The format, the signed test vectors, the kya tool, a Python client and the register server are published under Apache-2.0 at github.com/jackmitchelwidman/know-your-agent. Anyone may run a register from it: docker compose up. Registers that speak the format answer each other's lookups, so an agent entered on one can be checked from another; each register re-checks the signatures itself and trusts no peer's word.

The Register of Traceable Agents is the first and canonical register. It was started by Sheaf Labs, which wrote the format and runs it today; the plan is a body of the register's members, in which every operator and every system that issues IDs has a seat. How that works.

Check your implementation

git clone https://github.com/jackmitchelwidman/know-your-agent
node traceable-agents/spec/vectors/verify.mjs

Fixed keys and signed objects for every part of the format. An implementation that recomputes the same hashes and accepts and rejects the same signatures speaks KYA.