NVIDIA NemoClaw
NexusAi Summary: NemoClaw is NVIDIA’s open-source security stack for AI agents, combining OpenShell kernel sandboxes, a default‑deny network policy engine, a Privacy Router for local/cloud routing, and Nemotron local inference to keep sensitive data on your GPUs with full auditability.
NexusAi Overview
Agents run inside OpenShell sandboxes with filesystem, process, and syscall policies. Outbound requests are blocked by default; operators approve domains via a TUI or policy files. The Privacy Router classifies data sensitivity, routing protected content to local Nemotron models and non-sensitive calls to cloud APIs. All activity is logged for compliance.
How NemoClaw Works
Platform, security, and MLOps teams adopting agents for customer support, sales operations, SecOps automation, and infrastructure management benefit most. Regulated industries—including finance, healthcare, and public sector—gain SOC 2 and HIPAA‑supportive controls through auditable logs, default‑deny networking, and strict data residency. Startups and enterprises can standardize a security posture from a single RTX workstation to DGX clusters, supporting offline‑only or mixed local/cloud deployments.
- Isolate agents in kernel-level sandboxes with strict filesystem, process, and syscall policies.
- Enforce default‑deny egress and approve external domains in real time via TUI.
- Route sensitive prompts to local Nemotron; send safe, nonsensitive queries to cloud.
- Run Nemotron 3 Super 120B MoE offline on NVIDIA GPUs for privacy.
- Audit every action and network event with immutable logs and operator approvals.
Highlights
What Users Say
Install with a one‑liner on Linux, Windows via WSL2 with Docker Desktop, or macOS using Colima or Docker Desktop. The installer provisions Node.js, the OpenShell runtime, and the NemoClaw CLI, then runs nemoclaw onboard to bootstrap a secure environment. Define egress allowlists and sensitivity rules in policy files, approve new domains in the OpenShell TUI, and monitor audit logs centrally. Local inference requires NVIDIA GPUs; cloud routing works without GPUs. Podman on macOS is not currently supported.
NexusAi: Default‑deny networking with human approvals gave us SOC 2-ready control without rewriting agents.
Getting Started
NemoClaw unifies kernel isolation, default‑deny networking, sensitivity‑aware routing, and local Nemotron models into a deployable stack. One command brings enterprise guardrails to existing OpenClaw agents with minimal changes, while auditability and human approvals enable compliance narratives security teams recognize. Hardware auto‑detection and multi‑platform support streamline rollout from laptops to DGX, cutting spend and keeping data on-prem.
Open the tool and review its core product experience.
Create your account or access your existing workspace.
Use your own task to judge speed, quality, and fit.
Check similar AI tools before making a final decision.
NVIDIA NemoClaw
NVIDIA is a leader in accelerated computing and AI, building GPUs, systems, and software that power datacenters and edge deployments. The NemoClaw project extends NVIDIA’s Nemotron model family with secure agent tooling—OpenShell sandboxes, privacy routing, and network controls—for enterprises deploying AI safely at scale.
NemoClaw is NVIDIA’s open-source security stack for running AI agents safely in production. It combines OpenShell, a kernel-level sandbox runtime, with a default‑deny network policy engine, a Privacy Router that classifies and routes data between local and cloud models, and Nemotron local inference. Sensitive prompts remain on your GPUs, while low‑risk queries can leverage external providers. Operators approve or pre‑approve outbound domains, and every action is logged. The result is enterprise controls—segmentation, audibility, and data residency—without rewriting existing OpenClaw-based agents. Deployable on GeForce RTX workstations through DGX systems, NemoClaw installs with a single command on Linux, Windows via WSL2, and macOS with Colima or Docker Desktop. It auto-installs prerequisites and ships a CLI to onboard agents quickly.
Visit NVIDIA NemoClaw
Comments (0)
No Comments Found