We started by implementing the CSA's "Agentic Trust Framework" (ATF), but realized that "Behavioral Monitoring" wasn't enough for autonomous agents—we needed cryptographic Proof of Execution.
So we built VERA (Verifiable Enforcement for Runtime Agents).
It's a reference architecture with 5 hardened pillars: 1. Identity (DID:web + VCs) 2. Proof of Execution (Ed25519 signatures anchored to Solana) 3. Data Sovereignty (Sub-20ms ONNX firewall at the edge) 4. Segmentation (Policy-as-code) 5. Containment (Circuit breakers)
Everything is MIT. You can clone the repo and run `npm test` — 25 contract validation tests.
Repo: https://github.com/berlinailabs/vera-reference-implementatio... Whitepaper: https://berlinailabs.de/vera-paper.html Interactive demo: https://berlinailabs.de/vera-demo.html
Happy to answer architecture questions.