[NVIDIA Launches OpenShell Reference Architecture to Provide Secure Isolation and Hardware-Level Control for Enterprise AI Agents] TechFlow News, October 21 — According to an announcement from NVIDIA, the company has introduced the NVIDIA OpenShell reference architecture, designed for building, running, monitoring, and constraining AI Agents within enterprise infrastructure and edge systems, while enhancing governance, isolation, and security controls. This architecture is based on the OpenShell secure runtime and integrates NVIDIA Vera Rubin, Vera CPU, and BlueField-4 DPU. Vera Rubin handles model inference, Vera CPU is used for running the Agent harness and tool invocation, and BlueField-4 continuously monitors Agent behavior and enforces hardware-level policy control through NVIDIA Sentry. When an Agent exceeds predefined software boundaries, Sentry can isolate and terminate it within milliseconds.