# Omni-Ouro Inc.

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Donovan Patterson](https://cerebralvalley.ai/u/OmniDon)
- **GitHub:** https://github.com/Omni-Ouro/Eyd/tree/main
- **Demo video:** https://youtu.be/5X7RR0bsljE
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1315

EyD: Proof of Trust for AI agents

Agents are moving from answering questions to taking action: booking, buying, paying, and operating software on behalf of people and businesses. Businesses are responding by blocking them. Banks, airlines, and retailers now restrict agent traffic, not because agents are not useful, but because there is no accountability. There is no identity a site can verify, no record it can trust, and no way to tell a well-behaved agent from a misbehaving one. EyD is the accountability layer that lets sites open the door.

EyD is a small library the agent framework imports. Every tool call is signed under the agent's own Ed25519 key as an Action Record and reported to the EyD Clearing House before it executes. The counterparty site independently reports its side afterward. The Clearing House pairs the two signed reports, classifies the action (Expected, Unexpected, Gated, Declined, Deviated, or Unverified), and issues a signed Proof Receipt to both parties. An agent's Proof of Trust (PoT) score is counted from these receipts, weighted by how independent the reporting parties are. It is earned, never claimed.

Sites enforce with the EyD verifier SDK (Python and TypeScript, with turnkey FastAPI and Express gates): require EyD, set a minimum PoT score per action, and check AI Passport standing, including interoperability tested against ServiceNow's AI Passport conformance suite (verifier 81/81, caller 74/74). Gate refusals classify as Gated: score-neutral and streak-neutral, with the gate itself (threshold, set by whom, since when) bound into the receipt, a design change that came directly from professional review during the hackathon. A business dashboard lets operators adjust PoT requirements at runtime, query any agent's requests, and retrieve the receipt issued for each one.

Meta's Muse models run in two places: Muse drives the acting agents in the demo, whose real tool calls are what gets judged (including a Bad Agent that routes payments to an email tool and is caught as Deviated), and open-weight Muse Spark pre-registers an expected-outcome sketch for every declared action, bound into the receipt for the site to attest against. Classification and scoring are deterministic and never depend on the model.

Live now: the hosted Clearing House beta and a business playground at eyd.omni-ouro.com, where any EyD agent can test against real gates and any business can see what installing EyD looks like.

## More from Meta Global AI Developer Hackathon

- [FieldProof by 180Climate](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1312)
- [OPIE CHISOM SAMUEL](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1313)
- [Shunfeng](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1314)
- [PAVLENS AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1316)
- [Deep Shah](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1317)
- [ShowMe](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1318)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1315. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt