NVIDIA's enterprise-hardened OpenClaw wrapper
NemoClaw is NVIDIA's enterprise-hardened wrapper for OpenClaw, announced by Jensen Huang at GTC 2026 (March 16). It adds three critical security layers that OpenClaw lacks: OpenShell sandboxed execution, a PII-stripping privacy router, and intent verification before actions execute. Powered by Nemotron models via NVIDIA Cloud inference. Currently in alpha — not production-ready yet. Positioning: "OpenClaw was built for individuals. NemoClaw is built for companies." Pricing undisclosed but expected to follow NVIDIA AI Enterprise licensing.
All agent actions execute in an isolated sandbox — prevents malicious skills from accessing host systems, files, or network resources
Automatically detects and strips personally identifiable information before data reaches LLM backends — enterprise compliance built-in
Every agent action requires verification against declared intent before execution — blocks unauthorized operations and prompt injection attacks
Powered by NVIDIA Nemotron models via NVIDIA Cloud inference — enterprise-grade with SLA guarantees
NemoClaw v0.0.79 introduces OpenRouter as a first-class onboarding inference provider and hardens sandboxes with process and file-descriptor limits.
$0
Alpha — not production-ready
TBD
Expected via NVIDIA AI Enterprise license
Runs existing OpenClaw skills and integrations with added security layers — drop-in enterprise upgrade
Announced March 16, 2026 by Jensen Huang. Currently in alpha. Enterprise GA expected later in 2026
NemoClaw v0.0.93 updates DGX Station and DGX Spark onboarding guidance, adds resumable vLLM downloads, and prevents downloads on unsupported Intel macOS systems.
NemoClaw v0.0.94 enhances sandbox snapshot restore behavior, adds machine-readable JSONL onboarding progress streaming, and optimizes Hermes image build times.
Custom
Contact NVIDIA