NemoClaw (NVIDIA/NemoClaw) is NVIDIA’s official enterprise distribution of OpenClaw, extending it with the NVIDIA OpenShell governance and sandboxing runtime. Released March 16, 2026, it is currently in early preview and not yet production-ready.
NemoClaw provides policy-based sandboxing via declarative YAML configuration, with deny-all-egress by default. Pre-built policies cover PyPI, Docker Hub, Slack, and Jira. It evaluates local hardware for running Nemotron models and supports live network and inference policy updates from outside the sandbox.
Official documentation: docs.nvidia.com/nemoclaw/latest/index.html. Community installer available at nvidia-nemoclaw/NemoClaw (with bundled Nemotron-3-Super-120B and AMD/Intel GPU emulation). Curated presets at VoltAgent/awesome-nemoclaw.
Frequently Asked Questions
What is NemoClaw and who makes it?
NemoClaw is NVIDIA’s official enterprise distribution of OpenClaw, released March 16, 2026. It extends OpenClaw with the NVIDIA OpenShell governance runtime, adding policy-based sandboxing, declarative YAML security configuration, and first-class support for NVIDIA’s Nemotron language models. It is currently in early preview and not yet production-ready.
How does NemoClaw’s sandboxing work?
NemoClaw uses a deny-all-egress policy by default â agents cannot make any outbound network calls unless explicitly permitted by YAML policy. Pre-built policies cover common targets like PyPI, Docker Hub, Slack, and Jira. Policies can be updated live from outside the sandbox without restarting the agent, which is crucial for enterprise deployments where downtime is costly.
Does NemoClaw require NVIDIA hardware?
No. While NemoClaw is optimized to run NVIDIA’s Nemotron models natively on NVIDIA GPUs, the community installer also includes AMD and Intel GPU emulation for non-NVIDIA hardware. You can run NemoClaw on standard CPU hardware if you prefer to use cloud-based AI providers rather than local inference.
How does NemoClaw compare to other enterprise OpenClaw options?
In the enterprise tier, NemoClaw competes primarily with IronClaw (Near AI) and Carapace. NemoClaw differentiates through NVIDIA hardware integration and the Nemotron model ecosystem. IronClaw focuses on WebAssembly-based tool isolation, while Carapace was built specifically in response to the January 2026 CVEs. See the enterprise forks overview for a full comparison.