Sunglasses is a filter that sits ahead of your NemoClaw agent. Always ON. NemoClaw runs OpenClaw inside NVIDIA's OpenShell sandbox — Sunglasses wires in at the OpenClaw layer, which NVIDIA explicitly delegates all LLM-layer security to.

Who this page is for

Enterprise and research teams running NVIDIA NemoClaw (build.nvidia.com/nemoclaw) — the hardened NVIDIA Agent Toolkit reference stack that runs OpenClaw inside a policy-enforced container. NemoClaw adds infrastructure security (sandbox isolation, network egress policy, credential isolation, SSRF protection). Sunglasses adds the runtime LLM-layer filter NemoClaw intentionally does not ship itself.

The command

# Inside the NemoClaw sandbox (after nemoclaw onboard and connect): nemoclaw connect # Then run OpenClaw's MCP registration from inside the sandbox: openclaw mcp set sunglasses '{"command":"python","args":["-m","sunglasses.mcp"]}'

Wire Sunglasses into NemoClaw via OpenClaw MCP. NemoClaw delegates all application-layer security to OpenClaw. Connect to the NemoClaw sandbox (nemoclaw connect), then register Sunglasses using OpenClaw's MCP command. The filter runs inside the sandbox alongside the agent — NVIDIA's network policy may require approval for outbound MCP connections.

Benefit: NemoClaw adds infrastructure security. Sunglasses adds LLM-layer security. Defense-in-depth: sandbox isolation catches code-level attacks, the filter catches prompt-level attacks. Together they cover both surfaces.

Install Sunglasses first

If you haven't installed Sunglasses on your machine yet, do this first in a fresh terminal:

python3 -m venv sunglasses-env
source sunglasses-env/bin/activate
pip install --upgrade sunglasses

On modern macOS and many Linux systems, your system Python may block direct installs. A small virtual environment keeps the install clean. Windows: replace source sunglasses-env/bin/activate with sunglasses-env\Scripts\activate.

Full walkthrough in progress. The core command and identity are ready. Detailed step-by-step screenshots, real-world attack examples, and troubleshooting are being added.

FAQ

What is NemoClaw?

NVIDIA NemoClaw is an open-source reference stack that runs OpenClaw agents inside NVIDIA OpenShell — a hardened sandbox with network egress policy, credential isolation, SSRF protection, and filesystem restrictions. Alpha release since March 16, 2026. Not production-ready per NVIDIA's own docs.

How does Sunglasses integrate with NemoClaw?

NemoClaw delegates LLM-layer security to OpenClaw. Sunglasses registers as an OpenClaw MCP server inside the NemoClaw sandbox using the standard openclaw mcp set command. NVIDIA's network policy may require approval for outbound MCP connections.

Why does NemoClaw not ship prompt injection detection itself?

NVIDIA explicitly architected NemoClaw as infrastructure-layer security (sandbox, network, credentials). They delegate application-layer security — including prompt injection detection — to OpenClaw. Sunglasses is the canonical filter for that layer.

Does NemoClaw require NVIDIA hardware?

No. NemoClaw runs on Linux (Docker), macOS Apple Silicon (Colima or Docker Desktop), Windows WSL2, and DGX hardware. GPU is not required for the base setup. It's a Node.js CLI wrapping Docker.

Is NemoClaw production-ready?

No. NVIDIA's docs explicitly state 'Do not use in production environments.' It's an alpha reference stack — v0.1.0 on npm as of April 2026. Use for research, evaluation, and hardened dev setups, not live workloads.

Where this wiring fits

Sunglasses is one filter with many wiring options. This page covers NemoClaw (NVIDIA). Other wiring paths:

Same filter underneath. Different wiring based on your stack.