Nvidia donates agent-security runtime OpenShell to Cloud Native Computing Foundation

The latest:

Nvidia is handing OpenShell, its runtime for securing and governing autonomous AI agents, to the Cloud Native Computing Foundation as a Sandbox project, the company said at KubeCon on November 10, 2026. Alongside it, Nvidia unveiled three new pieces of its DSX AI Factory Platform and joined a Google-led effort to run long-lived agent workloads on Kubernetes.

Details:

• What OpenShell does: Nvidia said the runtime applies deterministic controls to agent activity, letting teams enforce policy as agents call tools, reach inference services and execute workloads. Under CNCF Sandbox status it will be governed by the organizations embedding it in their products rather than by Nvidia alone.
• Adoption to date: According to Nvidia, hundreds of organizations have deployed OpenShell and the project has drawn tens of thousands of contributions in under a year. The company did not name users or give a timeline for graduating beyond Sandbox status.
• The Linux Foundation view: Jim Zemlin, CEO of the Linux Foundation, framed secure operating environments for agents as a new infrastructure problem for the cloud-native community, and said the contribution gives developers “a shared foundation to build on” and a role in shaping how agent security evolves.
• The Google tie-in: Nvidia is also contributing to Agent Substrate, a Google-originated CNCF Sandbox framework for long-running agent workloads. Nvidia said it is making the project work with its Dynamo inference stack and KAI Scheduler, and is building a joint OpenShell–Agent Substrate integration for cloud and on-premises deployments.
• The compute argument: Agent Substrate preserves an agent’s environment between tasks while releasing compute when the agent sits idle. Nvidia said that lets teams scale agent workloads without holding dedicated hardware open for every session — the cost problem behind persistent agents.
• New Kubernetes tooling: Nvidia Kubernetes Engine and Nvidia Kubernetes Recipes entered public preview as open source building blocks for managed Kubernetes on GPU infrastructure, automating cluster lifecycle management and shipping signed, version-pinned images. Nvidia did not say when either reaches general availability.
• Launch partners: The two tools launched with ecosystem support from Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud and vCluster, together with Firmus and SK Telecom.
• The Snapshot numbers: Snapshot restores checkpointed replicas in seconds without model reload or warmup. Design partner Crusoe, testing GLM-5.3 on Crusoe Cloud with HGX B300 systems, cut time-to-ready from roughly eight minutes to 31 seconds, Nvidia said.
• Second benchmark: Saturn Cloud, which Nvidia described as a token factory platform, benchmarked Snapshot for scale-from-zero inference and reduced vLLM startup times from a range of two to 8.5 minutes down to six to 34 seconds.

Background:

The CNCF is a nonprofit under the Linux Foundation and the host of KubeCon. Its Sandbox tier is the entry stage for early projects, placing technical direction with a multi-vendor community rather than a single corporate owner.

Between the lines:

The two contributions sit on either side of the same bottleneck. Agent Substrate frees idle compute and Snapshot shortens cold starts to seconds, which together make it cheaper to run agents that persist between tasks — and OpenShell is the control layer those always-available agents would need. Nvidia’s stated work wiring Agent Substrate to Dynamo and KAI Scheduler keeps its own inference and scheduling stack at the center of the open standard.

What’s next

Watch whether OpenShell and Agent Substrate advance past CNCF Sandbox status, when Nvidia Kubernetes Engine and Recipes exit public preview, and whether the joint Nvidia–Google OpenShell integration ships.