Current AI governance relies on extrinsic proxies (e.g., Open Policy Agent) that block calls at the API boundary. This is a "perimeter defense" model. Flockrush integrates Runtime Formal Verification directly into the cognitive loop.
Actions are checked against Linear Temporal Logic (LTL) specifications. We don't ask "Does this look safe?"; we prove "This transition does not violate the safety invariant." If an agent tries to exfiltrate data, the monitor detects the effect and blocks it mathematically.
Engineers design Genotypes—parameters for replication, mutation, and tool affinity. The Phenotype (the running agent) emerges from the interaction of this DNA with the environment. This replaces brittle "System Prompts" with evolving behavioral primitives.