NVIDIA Ecosystem
OpenClaw
The autonomous AI assistant that NemoClaw secures. OpenClaw provides agentic capabilities (code generation, task planning, tool use); NemoClaw provides the security envelope. NemoClaw is purpose-built for OpenClaw's architecture.
OpenShell
The general-purpose agent security runtime underneath NemoClaw. OpenShell provides sandboxing, policy enforcement, and privacy routing as reusable primitives. Also works with Claude Code, Cursor, and Codex. Use OpenShell directly for non-OpenClaw agents.
NVIDIA Agent Toolkit
The broader software suite including OpenShell, inference management tools, and deployment utilities. NemoClaw is one application built on Agent Toolkit components.
NVIDIA Nemotron
Family of open-source language models from NVIDIA. NemoClaw's default configuration uses nvidia/nemotron-3-super-120b-a12b, a 120B-parameter model optimized for agent tasks. Can run locally on NVIDIA GPUs.
NeMo Guardrails
Separate NVIDIA project providing programmable guardrails for LLM conversations. Operates at the prompt/response level (what the LLM says), while NemoClaw operates at infrastructure level (what the agent can do). Can be combined for defense in depth.
Inference Providers
| Provider | Type | Privacy | Status |
|---|---|---|---|
| NVIDIA Endpoints | Cloud | Moderate | Supported |
| OpenAI | Cloud | Low | Supported |
| Anthropic | Cloud | Low | Supported |
| Google Gemini | Cloud | Low | Supported |
| Ollama (local) | Local | High | Experimental |
| vLLM (local) | Local | High | Experimental |
Messaging Integrations
Host-side messaging bridges connect the sandboxed agent to external platforms. All bridges run on the host (not inside the sandbox), keeping platform credentials out of the agent's reach.
Telegram, Discord & Slack
Configure via environment variables (TELEGRAM_BOT_TOKEN, DISCORD_BOT_TOKEN, Slack app credentials). Forward messages to/from the sandbox. Subject to the same network policies as other connections.
Community
NemoClaw is open source under Apache 2.0 with an active community: 18,000+ GitHub stars, 2,100+ forks. Support channels include GitHub Discussions, GitHub Issues, and the NVIDIA developer Discord. Security issues can be reported via the NVIDIA Vulnerability Disclosure Program or GitHub private advisories.