Contributions to CNCF projects and new Kubernetes tools focus on AI agent security, workload management and GPU infrastructure, as NVIDIA works with ecosystem partners to support deployment and faster scaling across enterprise AI computing environments.
NVIDIA has announced open-source software and infrastructure contributions at KubeCon + CloudNativeCon, including tools for securing autonomous AI agents, managing agent workloads and operating GPU infrastructure on Kubernetes.
The company is contributing NVIDIA OpenShell, an open-source runtime for securing and governing autonomous agents, to the Cloud Native Computing Foundation (CNCF) as a Sandbox project. It is also contributing to Agent Substrate, a Google-led framework for running long-running agent workloads on Kubernetes that has been accepted into the CNCF Sandbox.
Alongside these contributions, NVIDIA introduced Kubernetes Engine, Kubernetes Recipes and Snapshot as new components of its DSX AI Factory Platform. The tools address cluster management, infrastructure deployment and inference scaling.
OpenShell brings agent security into community development
OpenShell applies deterministic controls to agent activity, allowing organisations to enforce policies as agents use tools, access inference services and execute workloads. Its architecture supports integration with existing compute resources, inference providers, credential storage and monitoring systems.
Through CNCF, the project will bring together organisations incorporating OpenShell into their products and security professionals contributing to its development.
“The cloud-native community has a new infrastructure challenge: giving agents secure environments in which to operate,” said Jim Zemlin, CEO of the Linux Foundation. “NVIDIA’s contribution of OpenShell to CNCF gives developers a shared foundation to build on and an opportunity to shape how agent security evolves. Bringing this work into the community enables broader collaboration on infrastructure that will be critical to the next generation of cloud-native applications.”
According to NVIDIA, hundreds of organisations have adopted OpenShell, and the project has received tens of thousands of contributions in less than a year since its launch.
Agent Substrate targets workloads at scale
NVIDIA’s work on Agent Substrate includes ensuring compatibility with NVIDIA Dynamo for model inference and KAI Scheduler for allocating shared cluster resources. The effort connects agent execution, inference serving and resource scheduling on Kubernetes.
Agent Substrate preserves agent environments between tasks while releasing compute resources when agents are idle. Suspending and resuming these environments allows organisations to manage a shared pool of agents without maintaining dedicated compute for every session.
NVIDIA and Google are also collaborating on an integration between OpenShell and Agent Substrate to support agent security and governance across cloud and on-premises environments.
New tools address cluster management and inference startup
NVIDIA Kubernetes Engine (NVKE) and NVIDIA Kubernetes Recipes (NVKR), both available in public preview, provide open-source components for delivering managed Kubernetes services on GPU infrastructure. They automate cluster lifecycle management and provide signed images pinned to specific versions.
Mirantis, Netris, OpenNebula Systems, Rafay, Spectro Cloud and vCluster, alongside Firmus and SK Telecom, have aligned with NVIDIA on plans to integrate NVKE and NVKR capabilities into their offerings.
NVIDIA Snapshot addresses delays associated with loading models and warming up inference workloads. The modular component restores checkpointed replicas without repeating model loading or warmup, supporting startup, recovery and scaling of inference and agentic workloads on Kubernetes.
Design partners Crusoe and Saturn Cloud evaluated Snapshot using NVIDIA HGX B300 systems. According to the announcement, Crusoe tested GLM-5.3 on Crusoe Cloud and reduced workload readiness time from approximately eight minutes to 31 seconds.
Saturn Cloud’s scale-from-zero inference tests reduced vLLM startup times from between two and 8.5 minutes to between six and 34 seconds.
See What’s Next in Tech With the Fast Forward Newsletter
Tweets From @varindiamag
Nothing to see here - yet
When they Tweet, their Tweets will show up here.
