Nvidia has transferred OpenShell, a protection and control system for autonomous AI agents, to the Cloud Native Computing Foundation (CNCF) under the Linux Foundation. Simultaneously, the company announced support for Google’s Agent Substrate project and introduced new open components for the DSX AI Factory Platform. All of these developments are built around Kubernetes, which has effectively become the standard environment for running large-scale AI workloads in enterprise and cloud infrastructures.
OpenShell: Rules for Autonomous Agents
OpenShell applies strict rules to software module actions, allowing enterprises to manage tool usage, access to neural network inference services, and task execution. The solution's architecture is extensible by connecting custom compute resources, credential stores, and monitoring systems. According to the announcement, more than 100 organizations are already using the tool a year after its launch, and the developer community has contributed over 10,000 code commits. Handing the project over to CNCF marks a transition from corporate management to independent industry stewardship, reducing vendor lock-in risks.
Agent Substrate and Compute Optimization
Efficient scaling of systems with autonomous agents requires compute optimization. Agent Substrate preserves AI agent execution environments between tasks and frees up capacity during idle periods. Combined with Nvidia Dynamo for model inference and KAI Scheduler for cluster resource allocation, it allows pausing and resuming agents within a shared pool without allocating dedicated servers for each session. Nvidia and Google are preparing an integrated solution combining OpenShell and Agent Substrate for unified agent control across cloud and on-premises networks.
Open Components of the DSX AI Factory Platform
To accelerate infrastructure deployment, the company announced open components Nvidia Kubernetes Engine and Nvidia Kubernetes Recipes, which automate cluster lifecycle management and supply verified software images. The Nvidia Snapshot tool restores saved replica states in seconds without restarting the neural network. Company partners tested the solution on HGX B300 systems: workload readiness time dropped from 8 minutes to 31 seconds, and request processing service startup accelerated from 2–8.5 minutes to 6–34 seconds.
Partner Ecosystem and Industry Significance
These developments are supported by a partner network of six managed Kubernetes service vendors and two telecommunications companies that will integrate the respective features into their products. Shifting critical AI infrastructure components under independent foundations lowers the barrier to entry for enterprises, eliminates dependency on proprietary solutions, and establishes a standardized environment for secure and cost-effective scaling of autonomous systems. Thus, Nvidia cements Kubernetes as the foundational layer for industrial AI, delegating core component management to industry communities.
