Suits developers comfortable with a terminal who want to run agents with more isolation; check the prerequisites page first, since supported platforms are limited.

What can you build with NemoClaw?

Run an OpenClaw agent in a sandbox rather than directly on your laptop. Try Hermes or LangChain Deep Agents Code with network access limited by policy. Compare inference providers for a sandboxed agent by switching the provider and model during onboarding. Study a reference design for isolating agents, using the architecture and security docs. Install it with help from a coding agent such as Claude Code or Cursor using the starter prompt.

NVIDIA /

NemoClaw

Reference stack for running AI agents inside NVIDIA OpenShell sandboxes

22.8k stars
169 stars since 3 Oct
TypeScript
Updated today
Some experience
Infrastructure & security

Open on GitHub (opens in a new tab) Website (opens in a new tab)

AI agents can run commands, call APIs and touch files, which is risky on your own machine. NemoClaw runs the agent inside an NVIDIA OpenShell sandbox instead, so you control what it can access.

A command-line tool walks you through onboarding: you choose an agent, a sandbox name, an inference provider and a model. NemoClaw then handles network policy, managed inference, integrations, saved state and the sandbox lifecycle.

It sits on top of OpenShell and the agents themselves. The docs have a page on when to use NemoClaw versus OpenShell alone. The README calls it an alpha project, with maintainers responding on a best-effort basis.

What it does

Three supported agents
Runs OpenClaw (the default), Hermes or LangChain Deep Agents Code, each with its own quickstart guide.
Guided onboarding
An installer asks about agent, sandbox name, inference provider and model. On supported DGX or WSL hosts, an express mode uses recommended presets.
Network policy
Baseline rules, an operator approval flow and egress control decide what the sandboxed agent may connect to. Policies can be changed statically or dynamically.
Managed inference
Inference is routed through configured providers, with validation for supported provider paths.
State and lifecycle
The CLI handles state persistence backed by OpenShell and the lifecycle operations of the sandbox.

What you could build

Run an OpenClaw agent in a sandbox rather than directly on your laptop.
Try Hermes or LangChain Deep Agents Code with network access limited by policy.
Compare inference providers for a sandboxed agent by switching the provider and model during onboarding.
Study a reference design for isolating agents, using the architecture and security docs.
Install it with help from a coding agent such as Claude Code or Cursor using the starter prompt.

Get started

Read the prerequisites in the docs, then run the installer and follow its prompts. On a supported DGX or WSL host you can press Enter to accept express install, which sets up OpenClaw. Otherwise answer n to pick Hermes or Deep Agents Code, a provider and a model. The README does not list the install command itself, so take it from the quickstart page.

Commands are copied from the project's README. Read it on GitHub for the full setup.