An AI agent that reads your email, replies on WhatsApp and runs commands can do anything on your machine that you can, including things you never asked for. NemoClaw locks it inside an NVIDIA OpenShell sandbox, controls which internet destinations it can reach (you approve or deny each new one) and lets you switch models, cloud or local, without rebuilding everything.
Who this is NOT for
If you do not already have an agent like OpenClaw or Hermes running, or do not know what you would do with one, this is not your starting point: it is not an assistant, it is the cage for one you already have. It is also not for anyone who needs something stable for the business by tomorrow. Its own README calls it an alpha project with best-effort support and no response timelines, and the repo has 777 open issues and PRs. If you do not use the terminal or know what WSL or Docker are, do not install it on your own.
Almost all agent talk is about what they can do. This NVIDIA project is about what they should not be able to do, and that is why it is worth a look. The obvious alternative is installing OpenClaw directly on your computer and trusting its internal permissions. The mistake almost everyone makes is believing that an agent you have "restricted" is already contained. If it runs on the same machine as your files and passwords, that restriction can be pure decoration. What separates things here is a real sandbox with a network policy, and you pay for it in complexity: it is a full stack, not a one-click installer. It is very active (created March 2026, commits today) but alpha: use it to learn how to contain agents before you hand them real data.
Your first step
Open its Prerequisites page (docs.nvidia.com/nemoclaw/latest/get-started/prerequisites.html) and confirm your machine is a supported platform before installing anything. If it is, paste its official starter prompt into Claude Code. That prompt tells the agent to ask you one question at a time and to run nothing without your approval.
The NVIDIA NemoClaw implementation guide
The project belongs to NVIDIA Corporation (NVIDIA) and it is free: the link beside asks for nothing. The guide to set it up without getting lost is ours: step by step, common mistakes and a checklist. Leave your email and we will send it.
No spam. Unsubscribe anytime.