Security model
How secrets, access and the agent are protected.
- Row-level security isolates every organization at the database.
- Secrets and provider keys are encrypted with AES-256-GCM.
- The agent is outbound-only and authenticates with a revocable token.
- A strict nonce-based Content Security Policy protects the console.
There is no SOC 2 report or independent penetration test yet. For exactly what the agent sends and can run, see How it works and the trust model. Report vulnerabilities through the address in /.well-known/security.txt.