Nvidia is contributing OpenShell, its runtime for controlling autonomous agents, to the Cloud Native Computing Foundation as a Sandbox project.
OpenShell applies deterministic policies when agents call tools, access model services or execute workloads. Organizations can connect their own computing infrastructure, model providers, credential stores and monitoring systems. Moving the project to CNCF places its governance with a wider cloud-native community rather than Nvidia alone, although Sandbox status is an early stage of foundation maturity.
The company is also contributing to Agent Substrate, a Google-originated CNCF project for long-running agent workloads on Kubernetes. It can preserve an agent environment between tasks while releasing idle compute, and Nvidia is working on integration with its Dynamo inference software and KAI Scheduler.
Three additional open-source components target GPU operations. Nvidia Kubernetes Engine and Kubernetes Recipes, both in public preview, automate cluster lifecycle management with version-pinned images. Nvidia Snapshot restores checkpointed inference replicas without a full model reload or warmup. In partner tests cited by Nvidia, startup times fell from minutes to seconds, but those vendor-reported results will need validation across different models and cluster configurations.