NVIDIA donates OpenShell and previews non-production NVKE and NVKR
First merged summary.
NVIDIA has donated OpenShell to the CNCF Sandbox and released NVKE and NVKR as non-production public previews.
At KubeCon, NVIDIA said it is contributing OpenShell, an open-source agent runtime, to the CNCF Sandbox and contributing to Google-led Agent Substrate. It also announced NVIDIA Kubernetes Engine, NVIDIA Kubernetes Recipes, and Snapshot for the DSX AI Factory Platform. NVKE and NVKR are Apache 2.0 public previews; MarkTechPost reports NVIDIA says they are not for production because the API uses plain HTTP with authentication off and lacks upgrade and backup procedures. Crusoe and Saturn Cloud reported much shorter inference startup with Snapshot on NVIDIA HGX B300 systems. Firmus and SK Telecom are integrating both projects into AI factories, according to MarkTechPost.
- OpenShell applies deterministic controls as agents use tools, inference, and workloads.
- NVIDIA said hundreds of organizations use OpenShell after tens of thousands of contributions.
- After install, the NVKE API listens on port 30080 without authentication.
- Conformance was tested on Kubernetes 1.36; a local eval needs 6 CPUs and 12 GiB RAM.
- Saturn Cloud reported vLLM startup falling from 2–8.5 minutes to 6–34 seconds.
- Crusoe reported Snapshot cut GLM-5.3 readiness from about eight minutes to 31 seconds.
What each outlet adds
beyond the shared facts- NVIDIA BlogOpenShell CNCF donation, Agent Substrate, and Snapshot startup figures.
- MarkTechPostApache 2.0 license, auth-off API, Kubernetes 1.36, and integrator names.
Coverage
oldest first · one row per article · UTCNVIDIA donates OpenShell to CNCF and adds Kubernetes AI tools
A Crusoe test of Snapshot cut GLM-5.3 readiness from about eight minutes to 31 seconds.
NVIDIA open-sources NVKE and NVKR for GPU Kubernetes
Both Apache 2.0 projects are public previews; the NVKE API starts with authentication off.