News Summary:

The latest NVIDIA Agent Toolkit software, featuring updated NVIDIA NemoClaw blueprints, Nemotron models, OpenShell secure runtime, and CUDA-X libraries, provides a robust open-source framework for enterprise development.
Leading firms, including Cadence, Dassault Systèmes, Siemens, and Synopsys, are pioneering the use of NVIDIA NemoClaw to create autonomous AI engineers, functioning as digital coworkers that significantly expedite simulation and verification workflows, reducing weeks of work to mere hours.
The NVIDIA Nemotron 3 Ultra, a compact and rapid open model designed for enduring agents, offers fivefold increases in inference speed and cuts costs by up to 30% for intricate agent-dependent tasks.
CrowdStrike and Palantir are revolutionizing domains such as cybersecurity and operational strategy through long-term AI agents fueled by Nemotron models, thereby enhancing data analysis efficiency and streamlining operations within complex settings.
A partnership between NVIDIA and Microsoft aims to deliver an integrated Windows environment for personal agents, introducing new security frameworks alongside NVIDIA OpenShell, while Canonical and Red Hat are working to incorporate OpenShell as a safe, open-source runtime with comprehensive policy and privacy controls across various platforms.
NVIDIA CUDA-X libraries, including cuDF, cuOpt, AI-Q, NeMo, PhysicsNeMo, and CUDA-Q, are now available as specialized skills for AI agents, broadening their functional capabilities.

NVIDIA GTC Taipei—NVIDIA unveiled groundbreaking software, open-source models, and crucial alliances with premier software platform providers aimed at developing autonomous AI agents across multiple industries, including engineering, healthcare, software development, and business operations.

Noteworthy software enterprises are leveraging NVIDIA Agent Toolkit software to fabricate secure, persistent AI agents that collaborate as digital coworkers.

These autonomous agents begin with a foundational model before employing a software interface termed a harness, which enables the model to perform orchestration, maintain context, retain memory, utilize tools, and ensure security.

The NVIDIA Agent Toolkit empowers enterprises to construct agents poised to function alongside human employees on a grand scale.

NVIDIA Nemotron open models and NVIDIA NemoClaw blueprints interface with popular harnesses; the NVIDIA OpenShell secure runtime establishes policy and privacy controls; and agents can now access NVIDIA CUDA-X libraries as functional skills.

“The world’s software pioneers are embedding AI agents within operational systems — demonstrating how AI coworkers can enhance cognitive speed and execute complex undertakings to address larger challenges,” asserted Jensen Huang, founder and CEO of NVIDIA.

“NVIDIA NemoClaw furnishes software developers within enterprises with the foundational elements to craft more secure, persistent AI coworkers, thereby augmenting human intelligence and transforming the nature of work.”

Design and Simulation Leaders Develop Autonomous AI Engineers with NVIDIA NemoClaw

Within the realms of semiconductor and industrial engineering, simulation and verification represent among the most laborious tasks, demanding teams to navigate intricate, repetitive workflows spanning days or weeks before designs can advance.

Cadence, Dassault Systèmes, Siemens, Synopsys, Flexcompute, Luminary, Neural Concept, nTop, P-1 AI, PhysicsX, and Synera are trailblazers in deploying autonomous AI engineers alongside human personnel, utilizing NVIDIA NemoClaw.

By transferring these time-intensive tasks to perpetually active AI engineers, organizations can condense what’s traditionally a lengthy cycle into mere hours, redirecting human expertise to areas demanding nuanced judgment.

Cadence employs NVIDIA OpenShell to safeguard its ChipStack AI Super Agent, an entirely autonomous AI engineer responsible for chip design and verification tasks.

Notably, NVIDIA was the inaugural customer to harness ChipStack for autonomous design verification.

Dassault Systèmes is utilizing NVIDIA NemoClaw and OpenShell to enhance the 3DEXPERIENCE platform, enabling long-term, autonomous agents across design, simulation, and manufacturing processes.

Siemens integrates NVIDIA NemoClaw and OpenShell within the Fuse EDA AI Agent, a bespoke autonomous agent designed to strategize and orchestrate multifaceted workflows in the domains of semiconductor, 3D integrated circuit, and printed circuit board design.

Synopsys is collaborating with NVIDIA to fabricate consistently operational autonomous AI engineers aimed at the field of chip design, with an emphasis on attaining comprehensive workflow autonomy.

Moreover, Foxconn has embarked on piloting NVIDIA NemoClaw to energize its Nurabot and CoDoctor platforms, employing specialized teams of AI agents for clinical reasoning, documentation, and care coordination.

Foxconn is also integrating NVIDIA FOX and NemoClaw blueprints to construct MoMClaw, a factory operations agent designed to amalgamate sensor and machine data with AI agents to provide instantaneous insights and actionable strategies, equipped with OpenShell’s privacy protections and safety protocols.

NVIDIA Introduces New Nemotron Open Models for Persistent Agents

To enhance the intelligence of autonomous entities, NVIDIA introduced new open models and datasets created in collaboration with the NVIDIA Nemotron Coalition.

NVIDIA Nemotron 3 Ultra emerges as a 550-billion-parameter mixture-of-experts model that bestows cutting-edge intelligence upon long-running agents, applicable across coding, research, and enterprise operations.

With potential for fivefold acceleration in inference speed and up to a 30% reduction in costs relative to comparable open frontier models, Ultra empowers agents to accomplish tasks with remarkable efficiency and affordability.

Nemotron 3 Ultra is post-trained for several leading agent platforms and harnesses, including Hermes Agent, LangChain Deep Agents, OpenClaw, OpenHands, and OpenCode.

Additional new Nemotron models, addressing safety and speech recognition, further augment the model family’s capabilities for developing specialized, efficient agents tailored for enterprise tasks.

The deployment of NVIDIA Nemotron models is fostering a new generation of long-running AI agents across enterprise ecosystems from organizations such as CrowdStrike and Palantir.

These agents facilitate expedited data analysis, task coordination, and operational efficiency in both cybersecurity and enterprise settings.

CrowdStrike leverages NVIDIA Nemotron models for specialized agents dedicated to the ongoing detection, prioritization, and remediation of vulnerabilities and policy irregularities, significantly abbreviating response times against threats while alleviating workload on security teams.

Palantir integrates NVIDIA Nemotron models into its AI FDE (Forward Deployed Engineer) platform to autonomously perform complex operations while enabling continuous learning from agency interactions to build specialist, air-gapped enterprise frameworks.

Key Software Platforms to Integrate Secure Agent Runtime via NVIDIA OpenShell

Autonomous agents capable of code writing, sub-agent generation, and context retention across different sessions may access local files, familiarize themselves with new tools, and execute intricate workflows with growing autonomy.

As agents advance in capability, the necessity for stringent guardrails becomes paramount. A vital layer comprises a runtime featuring adjustable privacy and security parameters, ensuring that autonomous agents can be deployed healthily and securely at scale.

NVIDIA and Microsoft are collaborating on developing novel Windows security primitives in tandem with the NVIDIA OpenShell runtime to guarantee safe operation of agents while maintaining complete user oversight.

These new Windows primitives provide functionalities surrounding identity, containment, policy implementation, and holistic security, facilitating the native operation and construction of agents.

NVIDIA OpenShell augments these primitives by offering comprehensive policy features and will intelligently direct queries to local models while respecting user privacy policies, including anonymizing personal data in requests dispatched to cloud services.

Canonical is set to incorporate OpenShell alongside Ubuntu utilizing supported snaps and OCI-compliant containers to facilitate the operation of autonomous agents on global enterprise servers.

Red Hat is integrating OpenShell into its comprehensive Red Hat AI platform to ensure systematic control and policy enforcement at the infrastructure level.

Furthermore, the company is contributing significantly to the OpenShell upstream open-source initiative to set standardized protocols for agent management across enterprise platforms.

Today’s announcements build on recent integrations by SAP, which is embedding OpenShell into the Joule Studio runtime, a component of the SAP Business AI Platform for enterprise AI agents, and ServiceNow, which has secured Project Arc, its enterprise autonomous desktop agent, with OpenShell to enhance policy-driven governance for enterprise safety.

OpenShell operates effectively across on-premises, hybrid, and enterprise cloud environments, as well as on local devices, including NVIDIA RTX Spark, NVIDIA DGX Spark and GB10 systems from various vendors, alongside NVIDIA DGX Station for Windows and NVIDIA DGX Station GB300 systems from NVIDIA affiliates.

NVIDIA CUDA-X Libraries Accessible as Skills for Autonomous Agents

NVIDIA CUDA-X libraries are now available to AI agents through domain-specific skills, empowering AI agents to effectively deploy specialized competencies to confront substantial challenges across science, industry, and enterprise. Notable examples include:

NVIDIA cuDF accelerates the data processing and analytics capabilities over extensive structured datasets, allowing agents rapid reasoning over enterprise data.
NVIDIA cuOpt aids agents in addressing intricate routing, scheduling, resource allocation, supply chain, and decision optimization challenges in real-time.
NVIDIA AI-Q equips agents with intelligent routing, persistent context, and built-in assessment functionalities for enterprise research and knowledge workflows.
NVIDIA NeMo enhances agent optimization, evaluation, and oversight through prompts, skills, model routing, and customization for specialized fields and regions.
NVIDIA PhysicsNeMo empowers agents to construct and validate high-fidelity AI physics models essential for complex scientific and engineering simulations.
NVIDIA CUDA-Q enables AI agents to streamline installation processes, generate and validate quantum scripts, simulate quantum systems, and orchestrate quantum applications.

NVIDIA has also launched a significant suite of open-source physical AI libraries, skills, models, and frameworks, facilitating AI agents and developers to swiftly accelerate the development of robotics, autonomous vehicles, and industrial systems.

Availability

NVIDIA NemoClaw is available now. NVIDIA OpenShell is presently accessible in early preview.

NVIDIA Nemotron 3 Ultra is slated for release on June 4 via Hugging Face, ModelScope, OpenRouter, and build.nvidia.com as NVIDIA NIM microservices, alongside a vast network of NVIDIA Cloud Partners, inference platforms, and cloud service providers.