Governed continual-learning memory for AI agents — rejects poisoned facts, derives unstored relationships. Reproducible head-to-head benchmarks.