Nvidia’s NemoClaw: enterprise OpenClaw
Nvidia unveiled NemoClaw as an enterprise‑grade, secure take on the OpenClaw agent stack — adding OpenShell isolation, network guardrails, privacy routers and policy engines for auditable agent execution. (youtube.com) Industry commentary claims OpenClaw’s adoption exploded ‘in weeks’ (compared to Linux’s decades), and Nvidia pairs NemoClaw with specialized models like Nemotron, Cosmos, GROOT and BioNemo to target regulated environments. (youtube.com)
NemoClaw was unveiled during NVIDIA’s GTC keynote on March 16, 2026, where the company rolled several agent- and model-related announcements. (developer.nvidia.com) NVIDIA labels NemoClaw “alpha” and its developer guide warns the stack is an early preview released March 16, 2026 and “not to be used in production,” while providing a one-line installer (curl | bash) in the quickstart. (docs.nvidia.com) NemoClaw’s GitHub repo shows active development with roughly 15.7k stars and about 1.6k forks, plus continuous commits and recent security hardening changes in the repository. (github.com) OpenClaw’s own GitHub presence has surged—its repo shows roughly 330k stars and about 64.2k forks—figures NVIDIA’s CEO used when arguing OpenClaw outpaced Linux’s historical adoption curve in weeks, a claim he made on-camera at GTC. (github.com) NVIDIA pairs NemoClaw with new open model families announced the same day—Nemotron 3 (Nemotron 3 Ultra billed for high-throughput NVFP4 performance), Cosmos and Isaac GR00T variants—and the company listed adopters including CodeRabbit, CrowdStrike, ServiceNow and Perplexity in its March 16 press materials. (investor.nvidia.com) NVIDIA’s technical blog documents OpenShell as the runtime that sits between an agent and infrastructure, distributed under Apache-2.0, and even shows a sample command (openshell sandbox create --remote spark --from openclaw) that runs claws unmodified inside a managed sandbox with remote inference routing. (developer.nvidia.com) The company’s model announcement also tied BioNeMo work to a new Proteina-Complexa model and an open dataset co-developed with Google DeepMind, EMBL-EBI and Seoul National University for AI-driven protein complex predictions. (investor.nvidia.com)
