Nvidia Introduces NemoClaw, an AI Stack to Make OpenClaw Agents More Secure

Nvidia introduced NemoClaw, an artificial intelligence (AI)-powered stack for the OpenClaw agents, on Monday. The announcement was made at the company's annual GPU Technology Conference (GTC) during the keynote session. NemoClaw essentially adds privacy and security guardrails to OpenClaw's AI agents, making them deployable and scalable for enterprises. The Santa Clara, California-based tech giant highlighted that the AI stack can be installed with a single command and is platform and agent agnostic. NemoClaw is currently available for developers and enterprises in preview.

Nvidia Introduces NemoClaw

In a newsroom post, the tech giant introduced and detailed NemoClaw. The new offering is an interesting one, given that it was specifically developed for OpenClaw (although it can work with other agents as well). The agentic automation company was recently acquired by OpenAI after it gained popularity among developers for its general-purpose agents, also known as claws.

Unlike typical AI agents, which are built for specific use cases and have limited access and understanding of external tools, OpenClaw's agents can perform a wide range of tasks in a virtual environment, making use of a large variety of tools. However, its proficiency is also its biggest vice.

Since OpenClaw is able to access the Internet, various databases, and the device it is running on, it becomes an easy target for cyber criminals. The open-source tool's lack of standardised security and privacy safeguards has also resulted in minimal adoption from enterprises. This is the problem Nvidia hopes to solve with NemoClaw.

NemoClaw uses Nvidia's Agent Toolkit software to optimise OpenClaw for enterprises and specific developer environments. The company claims it works with a single command by installing the open-source OpenShell runtime. The runtime creates an isolated sandboxed environment and open models, via which security and privacy guardrails can be added to autonomous agents.

Nvidia said NemoClaw can use any coding agent to tap into AI models, including the recently released Nemotron 3 Super, and run locally on the user's dedicated system. It uses a privacy router to let users run frontier models in the cloud. Currently, it is available in preview and can be downloaded from the company's website or GitHub listing. Enterprises can also deploy and run AI agents via popular cloud service providers.