NVIDIA launched the NVIDIA Agent Toolkit at GTC Taipei on Sunday, introducing an open-source software stack designed for building secure, long-running enterprise AI agents. The company also announced partnerships with major software firms, aiming to reduce engineering time from weeks to hours.
The Agent Toolkit includes key components such as NVIDIA NemoClaw blueprints for agent orchestration, the OpenShell secure runtime for privacy and policy controls, Nemotron open models for inference, and CUDA-X libraries for domain-specific agent skills. NemoClaw is available now, while OpenShell is in early preview. The new Nemotron 3 Ultra model, a 550-billion-parameter mixture-of-experts model promising up to 5x faster inference and 30% lower costs, is set to launch on June 4.
“NVIDIA NemoClaw provides enterprise software developers with the open building blocks to create more secure, long-running AI coworkers that amplify human expertise as they reshape how work gets done,” CEO Jensen Huang stated during his keynote.
Companies such as Cadence, Dassault Systèmes, Siemens, and Synopsys are the first to adopt NemoClaw for building autonomous AI engineers targeted at simulation and verification workflows. Cadence is utilizing OpenShell to secure its ChipStack AI Super Agent, with NVIDIA acting as the inaugural customer for its autonomous verification of chip designs.
In another partnership, Microsoft is collaborating with NVIDIA to create a native Windows experience for personal agents using OpenShell and new security features. Canonical and Red Hat are also integrating OpenShell into their enterprise platforms. Meanwhile, CrowdStrike and Palantir are deploying Nemotron models for cybersecurity and decision-making operations.
NVIDIA has released additional open-source physical AI libraries, skills, models, and frameworks for applications in robotics, autonomous vehicles, and industrial digital twins. These tools aim to help developers accelerate the development of physical AI systems.
GTC Taipei runs from June 1 to June 4, featuring sessions focused on AI factories, scaling infrastructure, and agentic AI applications.
