Repolode
Submit
NVIDIA

NVIDIA/NemoClaw

A reference stack from NVIDIA for running AI agents such as OpenClaw, Hermes and LangChain Deep Agents more safely inside OpenShell sandboxes.

View on GitHubWebsite
Stars
22,774
Forks
3,146
Open issues
765
Last commit
5 hours ago

What is it?

NemoClaw is an open source reference stack for running supported AI agents inside NVIDIA OpenShell sandboxes, which isolate the agent from the host. It supports OpenClaw (the default), Hermes and LangChain Deep Agents Code, and is operated through the NemoClaw CLI and agent-specific aliases.

The project covers the pieces needed to run an agent in a sandbox: guided onboarding, managed inference, network policy with operator approval for egress, managed integrations, state persistence backed by OpenShell, and lifecycle operations. Installation is done from a terminal installer, with an express mode that applies recommended presets on supported DGX and WSL hosts.

The project is labeled alpha, and maintainers handle issues and pull requests on a best-effort basis. Documentation covers the architecture, sandbox hardening, inference provider choices and troubleshooting.

Key features

  • 01Runs OpenClaw, Hermes and LangChain Deep Agents Code inside OpenShell sandboxes.
  • 02Provides managed inference with supported providers and routed configuration.
  • 03Enforces network egress policy, with operator approval for dynamic changes.
  • 04Offers guided onboarding and an express install mode on supported DGX and WSL hosts.
  • 05Persists agent state through OpenShell and manages sandbox lifecycle via the NemoClaw CLI.

Use cases

  • Running autonomous coding or general-purpose agents without giving them unrestricted access to the host machine.
  • Controlling which external network endpoints an agent can reach during a task.
  • Testing agent behavior with a managed inference backend inside an isolated environment.

Quick start

  1. 1

    Review the prerequisites for your platform before installing

  2. 2

    Run the installer in your terminal (the official install script from the README)

  3. 3

    Follow the quickstart guide for your chosen agent (OpenClaw, Hermes or LangChain Deep Agents Code)

Things to know

  • NemoClaw is an alpha project, and maintainers respond on a best-effort basis.
  • It runs agents inside NVIDIA OpenShell sandboxes, so OpenShell is required.

Topics

#ai-agents#sandboxing#inference#network-policy#typescript#agent-security#deep-agents#hermes#nvidia#openclaw

FAQ

What is NemoClaw?
A reference stack from NVIDIA for running AI agents such as OpenClaw, Hermes and LangChain Deep Agents more safely inside OpenShell sandboxes. NemoClaw is an open source reference stack for running supported AI agents inside NVIDIA OpenShell sandboxes, which isolate the agent from the host. It supports OpenClaw (the default), Hermes and LangChain Deep Agents Code, and is operated through the NemoClaw CLI and agent-specific aliases.
Is NemoClaw free to use? What license does it have?
NemoClaw is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is NemoClaw still maintained?
Actively maintained. The last commit to NemoClaw was on Oct 20, 2026.
What can I use NemoClaw for?
Running autonomous coding or general-purpose agents without giving them unrestricted access to the host machine. Controlling which external network endpoints an agent can reach during a task. Testing agent behavior with a managed inference backend inside an isolated environment.
How do I install NemoClaw?
Review the prerequisites for your platform before installing Run the installer in your terminal (the official install script from the README) Follow the quickstart guide for your chosen agent (OpenClaw, Hermes or LangChain Deep Agents Code)
Where can I get NemoClaw?
The source code and installation instructions are on GitHub: https://github.com/NVIDIA/NemoClaw

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects