Build today — use agent-classified nodes, signed vouches and local evidence verification.
An application can recognize an agent whose signed messages come from a persona that the recipient's trust reaches. This is a foundation for “acting for someone you trust”; your application must separately define who may authorize which actions.
1. Bind the agent's key
Arrange early access, including the app origin allowlist, and develop on staging. The bind-start contract accepts nodeKind: 'agent'. Complete the AuthenSee ceremony and applicable device approval flow; selecting the classification does not bypass binding or device authorization.
Does not establish
The agent classification is a label, not an independent check of software, ownership, humanhood or delegated authority.2. Make the owner's vouch explicit
An owner can issue a signed, directed vouch from their own node to the agent's node. Use the standard vouch model. The owner controls their outgoing score and can update or revoke that vouch.
Treat that edge as a trust declaration. If your product needs to say “this owner authorizes this purchase,” define and verify that separate authorization in your application. A path through an owner does not by itself specify permitted actions.
3. Verify the agent's message and your connection
Have the agent carry signed message evidence. The recipient checks the content signature, certificate and fresh device status, then queries from their own node to the agent's persona and verifies the proof locally.
Show which checks succeeded. Use “signature verified” and “connected within your policy” as separate results. A server-side agent cannot query from another user's node using its own key.
4. Keep action permission in your app
Bind sensitive requests to the intended operation and recipient, set an application freshness rule and require whatever confirmation the action needs. Model owner approval and revocation explicitly if you offer delegated actions.