Nvidia Launches Autonomous AI Agent Suite and Nemotron 3 Ultra to Revolutionize Enterprise Software Engineering Healthcare and Global Industrial Workflows
Nvidia has released an integrated software suite, and family of open source models to enable the deployment of autonomous AI agents in every industry across the world. The new tools are designed to accelerate engineering, health care and enterprise software applications with infrastructure to create a secure digital co worker. These tools are being utilized by leading software platforms to convert standard AI models into agents that can complete multi step, complex tasks according to official Nvidia developers.
To build an autonomous agent, a software component known as a harness needs to wrap around a foundation model in order to provide a tool interface, a reasoning engine, context awareness and memory. The Nvidia Agent Toolkit provides this: an enterprise ready software suite including NemoClaw to plug in popular harness designs and OpenShell, a secure runtime that adds strong privacy and safety barriers at the local level. Agents use these frameworks to take full advantage of existing GPU accelerated libraries as agent skills.
"The leaders in software worldwide are making AI agents part of every system where work gets done to demonstrate how AI coworkers help people work faster and tackle big problems by performing more complex, multi step workflows and calculations," said Jensen Huang, founder and CEO of NVIDIA.
"Through NVIDIA NemoClaw, enterprise software developers can build secure, long lived AI coworkers that complement human capabilities to revolutionize work."
In industrial engineering and semiconductor design simulation and verification are long, compute intensive jobs that can run for weeks or months manually. With NemoClaw, engineering firms Cadence, Dassault Systmes, Siemens, and Synopsys are building autonomous AI engineers to handle multi step, complicated workflows that may compress engineering verification by several weeks down to a few hours.
- Cadence is using the OpenShell to secure its ChipStack AI Super Agent.
- Nvidia is using the OpenShell to verify its microchip designs internally.
- Siemens is plugging NemoClaw into Fuse EDA AI Agent toorchestrate multi tool workflows for printed circuit board designs.
Outside of engineering, Foxconn is piloting the software to bring Nurabot and CoDoctor clinical platform tools, which aid in documentation and medical reasoning. The company is also leveraging these models to power its Nurabot and CoDoctor AI assistants for healthcare, and the MoMClaw factory agent which connects machines to real time sensor data to execute tasks. These agents use OpenShell to protect privacy and provide safety mechanisms on the factory floor.
Nvidia announced the Nemotron 3 Ultra- a 550 billion parameter mixture of experts model that can efficiently handle code, research, and enterprise workflows to power long running agents. The new model achieves speeds that are up to 5x faster, and runs operations that are up to 30% cheaper than similar open source models, Nvidia said. Nemotron 3 Ultra is designed to work with the leading deployment tools including LangChain Deep Agents, OpenClaw, and Hermes Agent.
As agents gain more autonomy with writing code and access to local data, security became a critical concern. Nvidia is partnering with Microsoft to integrate its OpenShell into Windows, ensuring that agents run under user control, and are able to utilize local models based on corporate privacy rules to obscure private information before sending data to cloud networks. Furthermore, Red Hat and Canonical are integrating OpenShell into Linux distributions, SAP is utilizing it for its Joule Studio runtime and ServiceNow is using it for its Project Arc desktop agent.
Nvidia has made its propriety CUDA X libraries available as direct skills that autonomous agents can use to solve challenging mathematical and logistical issues. These include:
- The cuDF library can process large structured data sets at very high speeds.
- The cuOpt library can optimize routing, scheduling, and supply chain logistics in real time.
- PhysicsNeMo library to create high fidelity physics models and benchmarking.
The NemoClaw software is available immediately while the OpenShell security runtime is in early preview. Nemotron 3 Ultra will be available in the coming months as a microservice on popular hosting services like Hugging Face, ModelScope and the Nvidia model portal. Verified agent skills are available through the Claude Code plug in marketplace and the Hermes Skills Hub.
