clampd

Most AI security tools work at the model layer - prompt filtering, jailbreak detection, content classification. They check what the agent says. Clampd works at the runtime layer. It checks what the agent actually does - every tool call, before it executes. What makes it different: Self-hosted, Rust, runs on your infra (not a SaaS API) Behavioral kill switch fires under 25ms on attack patterns A2A delegation chain governance with Ed25519 scope tokens Multilingual PII detection.