Regent Protocol Start building
Platform · AgentID

Every action starts with who.

AgentID gives an agent a verifiable identity connected to a responsible party. Without it, an API key is acting — and an API key answers to no one.

The failure it prevents

A leaked key spends like the agent it impersonates. When something goes wrong, the log shows a credential — not an accountable party. Identity has to be cryptographic and bound to a human before the first action, not reconstructed after the incident.

Identity graph

What an AgentID connects.

Responsible partyKYC-verified human/entity AgentIDagent_… · DID · KMS-signedanchored on Solana (devnet) Mandateswhat it may do with money Audit historyevery decision, anchored binds + can revoke
Revocation flows down the same edges: the responsible party's kill switch suspends the mandates and closes the history.
Mechanics

Issuance, in plain terms.

Technical issuance

Registration mints the agent_… identifier and DID, signs the identity payload with Regent's KMS root key, and queues the Solana anchor. Included in every plan.

Human binding

The responsible party is the KYC-verified account that registered the agent — never a client-supplied value. Optional human-KYC verification through a licensed provider is a separate, pass-through step; documents stay with the provider.

Revocation

One-way. A revoked AgentID stops authorizing immediately — all its mandates are suspended via the agent.revoked event.

Build with it

Register an agent and read it back over REST or the Python SDK. Live on devnet.

Read the API Next: Mandates →

One agent, three questions

Who is it?

A cryptographic identity, not an API key: requests signed via managed KMS, an AgentID anchored at issuance, checkable by any third party.

Who answers for it?

A KYC-verified person or entity who approved the agent and holds the kill switch. Incidents attribute to someone accountable — not to a credential.

How does it behave?

An evaluation tier, signed into the agent's public passport with an expiry. Trust is a process — expired claims vanish on their own.

The kill switch is held by a human

Verified person → approved the agent → holds the switch. Flip it, and every door closes: the registry answers revoked, verifying services refuse the signature, mandates stop authorizing — in our benchmarks, the first refusal lands in under half a second.

No secrets on the agent

The agent never holds a key it could leak. Requests are signed inside a managed KMS; a stolen prompt, a dumped memory, or a compromised host yields nothing an attacker can replay.