https://derekwang2002.github.io/ai-intelligence/en/projects/nvidia-openshell-sentry/ Material changes in AI models, agents, engineering and infrastructure, with primary sources and recommendations. en Mon, 02 Nov 2026 16:56:00 GMT https://derekwang2002.github.io/ai-intelligence/en/events/ev-20261021-04/ ev-20261021-04-published Wed, 21 Oct 2026 15:58:23 GMT New TRIAL <p>NVIDIA published a reference architecture around the Apache-2.0 OpenShell agent runtime. OpenShell separates gateway, supervisor, sandbox and policy-prover controls from the agent workload. NVIDIA also previewed Sentry, a BlueField-4 hardware-isolated layer for traffic inspection, memory introspection and containment; Sentry is planned for early 2027, while OpenShell code is available now.</p><p><strong>Why read: </strong>OpenShell makes external policy boundaries testable now, while Sentry is clearly bounded as unavailable hardware isolation.</p><p>Recommendation: TRIAL</p><p>Sources: <a href="https://developer.nvidia.com/blog/how-nvidia-openshell-reference-architecture-secures-and-governs-ai-agent-workloads-at-scale/">NVIDIA OpenShell reference architecture</a> · <a href="https://developer.nvidia.com/blog/inside-nvidia-sentry-secure-ai-agents-at-runtime/">Inside NVIDIA Sentry</a> · <a href="https://github.com/NVIDIA/OpenShell">NVIDIA OpenShell repository</a></p>