Nvidia plans open-source agent platform that runs off-chip

Nvidia is developing NemoClaw, an open-source platform intended to help enterprise software companies deploy AI agents that perform tasks for employees and learn over time. The platform will include security and privacy features aimed at enterprise use.

NemoClaw is being designed to run even when customer products do not use Nvidia chips. Nvidia plans to publish the project under an open-source license and may offer early access to select partners in exchange for contributions to the platform’s development. The company says it will include controls to support secure enterprise deployments.

The platform falls into a category of locally running, open-source AI tools often called “claws.” These tools execute a series of steps on behalf of users and can adapt as they operate. A previous project in this category, OpenClaw (earlier known as Clawdbot and Moltbot), attracted attention for its ability to run autonomously on personal computers; that project was later acquired by OpenAI, which hired its creator.

Nvidia has held talks with several large technology firms, including Salesforce, Cisco, Google, Adobe and cyber security firm CrowdStrike, to explore possible collaboration on NemoClaw. Those conversations are intended to shape features for enterprise customers, and no formal partnerships have been announced.

The effort fits within Nvidia’s broader push to support enterprise adoption of AI agents by supplying infrastructure, security layers and developer tools. Nvidia is also working on a new processor and a different approach to inference computing, a form of processing that runs AI models to respond to queries, to help service providers and model builders run AI workloads more efficiently.

Nvidia plans to present NemoClaw at its GTC developer conference in San Jose next month. The company has said the platform will include integration with a processor developed by the startup Groq as part of its runtime and infrastructure offering.