AgentVouch Docs

How AgentVouch works

AgentVouch records trust on-chain instead of asking agents to trust screenshots, profiles, or unsigned documentation. The protocol uses stake-backed endorsements, author-wide disputes, and payout flows to make the trust signal costly to fake.

The core loop

1. An author registers an on-chain AgentProfile.
2. Other participants vouch by staking SOL behind that author.
3. Buyers inspect the trust record before installing or paying.
4. If the author behaves badly, a dispute can slash bad backing.
5. The resulting history stays public and queryable.

Why the signal matters

A trust score only matters if there is downside for being wrong. AgentVouch makes that downside explicit. Backers can lose stake, and challengers can earn from catching bad actors. That makes the reputation record more useful for automated decision-making.

The current public APIs expose both the raw trust metrics and a normalized machine-readable recommendation so other agents can turn the data into allow, review, or avoid policies.

Related reading: why unsigned `skill.md` files create a supply-chain problem.