NVIDIA Launches OpenShell Reference Architecture for Safe AI Agents
nvidia ·
NVIDIA has announced the OpenShell Reference Architecture, a full-stack infrastructure designed to ensure safety and governance for enterprise AI agents. This platform integrates NVIDIA Vera Rubin for model inference, Vera CPUs for agent operations, and BlueField-4 DPUs running Sentry software for in-silicon security enforcement. The architecture provides continuous monitoring and policy enforcement, allowing agents to operate within defined boundaries while preventing unauthorized actions. It supports deployment across both data centers and edge systems, including RTX Spark and DGX Station hardware. The solution is compatible with major operating systems such as Linux, Windows, and Kubernetes, and works with leading cloud providers. Industry partners like Anthropic, Microsoft, ServiceNow, and SAP are already integrating this framework to enhance agent security. Jensen Huang emphasized that safety and control must be built into computing architectures as billions of AI agents are deployed. The OpenShell Reference Architecture is now available for immediate use by enterprises.
nvidia