Open-source memory poisoning defense for AI agents — detect, sanitize, quarantine, and block unsafe memory before it influences an agent.