Can agents find each other, prove identity, and request services over the open web?

Hypothesis

If autonomous agents are going to work together across organisations, they need a way to rendezvous on the open web: discover each other, prove who they are, and negotiate services — without a human introducing them. The CARP protocol proposes exactly that. Can we stand up a live endpoint and complete a real agent-to-agent exchange?

What we built

Thrivbe's live CARP interface, running on one of our own servers. It exposes a DID (decentralised identifier), an ADILOS challenge/response flow for proving identity, and a service menu other agents can browse and call. With it we ran our first no-money service-exchange experiments with ClawFace — two agents from different operators finding each other, authenticating, and trading services.

Learnings

• The hard part isn't the protocol — it's deciding what an agent is allowed to offer and accept on your behalf. The service menu is a policy document as much as an API.
• Identity-first design changes the conversation: once the challenge/response works, "who am I talking to?" stops being a trust guess and becomes a verifiable fact.
• Agent-to-agent exchange without money is a useful forcing function — it keeps the experiments about capability and trust rather than billing plumbing.

Log

• 2026-07-07 — First no-money service exchange with ClawFace completed.
• 2026-07-01 — CARP reference implementation deployed on Thrivbe infrastructure; DID + challenge/response live.