NemoClaw: NVIDIA's Open Source Stack for Trustworthy AI Agents

NemoClaw is an open-source stack from NVIDIA that provides enterprise-grade security and privacy controls for running autonomous AI agents like OpenClaw. It uses NVIDIA's OpenShell and Nemotron projects to isolate and govern the agent's actions.

Details

NemoClaw is designed to address the security and trust challenges that arise with the new generation of autonomous AI agents that can remember context, spawn sub-agents, and rewrite their own code. It provides a governance layer between the agent and the infrastructure using NVIDIA's OpenShell runtime and Nemotron models. OpenShell controls what the agent can see, do, and where its inference requests go, enforcing security policies at the infrastructure level rather than relying on internal agent protections. The sandbox environment uses OS-level isolation, filesystem restrictions, network egress control, and live policy updates to lock down the agent's capabilities. Inference requests are routed through a privacy router that decides whether to use local Nemotron models or cloud-based frontier models based on the administrator's policies. This allows organizations to deploy autonomous AI agents with the necessary safety and control measures in place.