AI agents are starting to act for companies and people: paying invoices, moving money, sending mail, talking to other agents. The register exists so that none of that is anonymous. Behind every agent on it there is an owner who signed for it and a record of what it has done.
The register was started by Sheaf Labs in 2026. Sheaf Labs built the first verifier, wrote the format and runs the site today. The plan is for the register to be governed by its members: every verifier that issues IDs and writes receipts gets a say, and the founding company gets no special answer from the lookup. Its own agents go through the same door as everyone else's.
Who an agent is and whether it is in standing is always public: its ID, its owner's ID, the owner's signature, the record's counts and the receipt chain. The owner's name and the agents' names appear only when the owner opts in. Until then the register shows "an owner on record" and the agent's fingerprint, and the owner's page shows counts without a list.
The content of an action. Not the invoice, not the email, not the message. The register holds a hash of it, which lets the owner prove later what the action was without the register ever seeing it. A register that held the content would be a watching system, and good actors would refuse it along with the bad ones.
The lookup gives one of three answers: identified with a verified owner, identified with an owner on record, or unknown. What an agent may do with each answer is your policy. The register says who stands behind an agent and what its record holds. It does not say the agent is good. That judgment stays with you.
Any program can make a key pair. What it cannot make is an owner's signature over it. A key with no signature is unknown to the register, which is the state we want unidentified agents to be in. And because the real key is held by the runtime, a copy of the agent with a new key starts with an empty record and no owner.
An agent's life is the span over which an owner vouches for it. It begins when the owner signs the ID and ends when the owner retires it, the register revokes it for cause, or the owner stops renewing it. IDs are renewed yearly, so an abandoned agent with a vanished owner lapses on its own. A retired ID keeps its record forever and is never reissued to a different agent.
If you own agents: use a verifier that is on the register. The first is Sheaf. Every agent you run through Sheaf gets an ID under your key, and every checked action adds a receipt. Your agents' IDs appear in your Sheaf console.
If you run a system that checks agents: you can issue IDs and write receipts in the published format and submit them to the register. Write to register@traceableagents.org. Verifiers that join have a seat in how the register is governed.
If you are a bank, insurer or regulator: the lookup is free and needs no account. The lookup API returns the same answer a person sees on this site, with the signatures to check it yourself.