Nyx

It provides a framework for adversarial testing and runtime security of AI agents, allowing developers to evaluate how their models behave under hostile inputs and conditions. The system offers on‑demand access to a virtual team of AI security engineers, automating the creation, execution, and analysis of adversarial scenarios without requiring a dedicated in‑house security staff. Designed for developers building or deploying AI agents, it aims to surface vulnerabilities early and monitor agents in production, giving actionable insights into potential exploits or unsafe behaviors. The experimental nature of the project means it is intended for users comfortable with evolving tools and willing to contribute feedback as the verification capabilities mature.