NVIDIA shipped NemoClaw this week, an open-source security and deployment platform built for OpenClaw agents. Released under the Apache 2.0 license, it’s available now. NVIDIA’s newsroom confirmed the announcement ahead of GTC 2026, with Wired reporting on it in the lead-up to the annual developer conference.
The platform centers on two components. According to NVIDIA, OpenShell routes inference requests between local GPUs and cloud models based on user-defined policies. NVIDIA also describes a privacy router that directs each request to either on-device or external model paths based on configured rules. NVIDIA says agents can run across local and cloud models through a single-command setup. These are vendor claims. No independent technical evaluation has been published.
What NemoClaw actually solves is the gap between “can run an agent” and “can run an agent responsibly in production.” Developers who want to keep certain requests on-device, for compliance, latency, or data sensitivity reasons – previously had to build that routing logic themselves. NemoClaw packages it.
This isn’t a standalone release. It follows NVIDIA’s five-model agent release from earlier in March. MarkTechPost’s coverage of the Apache 2.0 licensing confirms open-source availability. Together, the two announcements sketch a platform: models to run agents, infrastructure to run them safely. Enterprise teams evaluating OpenClaw should treat NemoClaw as the deployment layer, and should note that vendor claims about its security and routing behavior have not yet been independently tested.
Related: NVIDIA’s five-model agent release earlier this month established the model layer for OpenClaw adoption → Full coverage on the Technology page