Sigvara gives an AI agent a score built from real payments, and a bond it loses if caught cheating, on top of ERC-8004. The code, the tests and the simulation are public. What the project cannot supply for itself is independence: nobody outside the team checks a score before working with an agent yet, and the same person runs both oracle operators. Here are four ways to change that.
For an MCP server, an x402 endpoint, an agent marketplace, or anything else that takes requests or money from agents it does not know.
It is one read, isTrusted(agent, minScore, minBond), with no account and no API key. Tell us what you run and we will write the integration with you. The first few integrators get as much of our time as they need.
For someone who runs servers and would rather check the scores than take them on trust.
The checker recomputes every score from public chain data and records each time it disagrees with the primary. Today the same person runs both, so a disagreement catches a bug but not a dishonest operator. An operator outside the team is what fixes that. The checker runs as one Docker container, and on testnet its bond is test SVR, which we will help you post.
See the operators today and who checks the checker, or open an issue.
For anyone who likes breaking things.
Our own measurements say a ring of self-funded wallets tops out near 55, and that going past about 60 takes standing in ERC-8004 that an attacker does not control. Prove either claim wrong on testnet, or make a score say something untrue another way, and tell us how. Whatever holds up goes into the whitepaper's known weaknesses, with credit if you want it.
Start with what a fake score costs and the known weaknesses, then open an issue.
A bug in the contracts, the oracle or this site goes to security@sigvara.xyz instead. See the security policy.
For anyone building an agent.
Register it on testnet, post a test bond and let it build a record from real payments and signed outcomes. It costs nothing, and one agent from outside the team tells us more than another of our own.
Open an issue on GitHub, or message @SigvaraProtocol on X. Say what you run and what you would want from a score. A few lines is enough.
What is done and what is still open is in the launch checklist.