NVIDIA NemoClaw

NVIDIA NemoClaw is a self-hosted stack for people who want to run AI agents with controls over network access and credentials. It supports OpenClaw, Hermes and LangChain Deep Agents Code inside NVIDIA OpenShell sandboxes.

• Agent isolation: Run supported agents in sandboxes with network policies that control which external services they can reach.
• Local or cloud inference: Connect to an existing vLLM server, use Ollama with OpenClaw or Hermes on supported platforms, or choose hosted providers such as NVIDIA Endpoints, OpenAI, Anthropic, Google Gemini and OpenRouter.
• Credential handling: OpenShell holds credentials for managed inference and host-configured integrations, keeping their custody separate from the agent runtime.
• Messaging integrations: Connect OpenClaw or Hermes to services such as Telegram, Discord and Slack so you can interact with the agent through a messaging channel.
• Persistent agent state: Preserve sandbox state through OpenShell and manage agent lifecycles through the same stack.

NemoClaw is open source under Apache 2.0. It runs on supported macOS and Linux systems, with platform-specific paths for Windows WSL and NVIDIA DGX hardware. Local runtime availability depends on the agent and hardware; managed vLLM requires an eligible platform. Hosted inference sends model requests to the selected external provider and requires its credentials. Installer and supported CLI operations attempt to send operation telemetry to NVIDIA by default; you can disable it.