NVIDIA unveils NemoClaw tooling
NVIDIA published a new NemoClaw stack for open agent development — part of the company’s broader GTC push to standardize agent tooling and deployable stacks announced. NemoClaw joins Rubin/Vera as NVIDIA’s play to own the agent + inference developer workflow.
NemoClaw can install NVIDIA’s Nemotron models and the new OpenShell runtime with a single command, according to NVIDIA’s GTC press release. (nvidianews.nvidia.com) The stack ships a “privacy router” that Nvidia says lets local agents call frontier cloud models while enforcing data and security guardrails. (zdnet.com) NVIDIA’s announcement lists supported endpoints including RTX PCs, DGX Station and DGX Spark for always‑on agents, while the OpenShell runtime is described as enabling hybrid cloud and on‑prem deployments. (nvidianews.nvidia.com) OpenClaw’s rapid adoption — its main repository shows hundreds of thousands of GitHub stars — has driven a fast‑growing ecosystem of forks and managed offerings. (github.com) Commercial and cloud players are responding: Baidu rolled out a managed DuClaw service and several security‑first forks like NanoClaw have emerged, underscoring why Nvidia emphasized enterprise guardrails in NemoClaw. (msn.com) NemoClaw is linked into NVIDIA’s model and inference stack — NeMo/Nemotron models and NIM microservices — positioning the release as an end‑to‑end developer path from model to deployable agent. (nvidianews.nvidia.com)
