NVIDIA launches Agent Toolkit with NemoClaw to build secure autonomous AI agents at scale

NVIDIA Introduces Comprehensive Framework for Enterprise AI Agents

NVIDIA announced a major expansion of its Agent Toolkit at GTC Taipei during COMPUTEX, delivering open-source foundations for building secure autonomous AI agents across industries. The NVIDIA Agent Toolkit includes NemoClaw blueprints, Nemotron models, OpenShell secure runtime, and CUDA-X libraries with agent skills, providing developers with comprehensive building blocks to create long-running digital coworkers 4

4

. According to NVIDIA CEO Jensen Huang, "AI agents are revolutionizing software development, and that shift is now coming to physical AI, extending into the systems that will transform transportation, manufacturing, healthcare and robotics" 3

3

.

The release addresses critical challenges in deploying AI agents at scale, particularly around orchestration, security, and workflow automation. While large language models have proven capable as coding assistants, they struggle with complex business and operational tasks that require persistent memory, tool integration, and multi-agent collaboration 5

5

. NVIDIA's framework provides the infrastructure layer needed to transform frontier models into fully functional autonomous systems.

Enterprise Software Leaders Deploy Autonomous AI Engineers

Enterprise software leaders including Cadence, Dassault Systèmes, Siemens, and Synopsys are among the first to build autonomous AI engineers using NVIDIA NemoClaw, compressing weeks of engineering work into hours 4

4

. These AI agents automate end-to-end industrial engineering workflows surrounding simulations, including computer-aided design, meshing, simulation setup, debugging, and post-processing 1

1

.

Cadence is building an autonomous register-transfer level engineer with NemoClaw that orchestrates Cadence Design Systems ChipStack for design and verification, cutting time for RTL verification from weeks to hours 1

1

. Siemens is integrating NVIDIA NemoClaw and OpenShell into Fuse EDA AI Agent, a purpose-built autonomous agent that plans and orchestrates multi-tool workflows across semiconductor, 3D integrated circuit, and printed circuit board system design 4

4

. Synopsys is collaborating with NVIDIA to apply agents to end-to-end engineering workflows, with Ansys Icepak being demoed to mesh, simulate, and optimize GPU electronics cooling designs 1

1

.

NemoClaw Framework and OpenShell Security Runtime

NemoClaw serves as an open blueprint for building specialized, long-running agents with a secure runtime and frontier models 1

1

. The framework includes a choice of harness that can be integrated with various orchestration frameworks enterprises use to deploy and coordinate agents, such as OpenClaw and Hermes, as well as a model router and NVIDIA NeMo libraries for customization 1

1

.

The OpenShell runtime provides policy-based security and privacy governance on local or cloud hardware, addressing the massive security risks that emerge when organizations give autonomous agents freedom to access sensitive files and make code changes 5

5

. Developed in collaboration with Microsoft, Canonical, and Red Hat, OpenShell integrates with native Windows security primitives to ensure AI agents remain under full user control 4

4

. The runtime can intelligently mask sensitive data before sending queries to cloud-based models and ensure the most sensitive workloads are routed to local hardware only 5

5

.

Physical AI Reaches Edge Devices With Jetson Platform

NVIDIA announced JetPack 7.2 and NemoClaw support on Jetson, bringing agentic AI from servers and workstations into the physical world across robotics, inspection, and industrial automation 2

2

. JetPack 7.2 brings agentic AI skills, Yocto project support, NVIDIA CUDA 13 on Jetson Orin, substantial performance gain on Jetson AGX Orin 32GB module delivering 241 TOPS of AI compute (up 20% above original spec), and Multi-Instance GPU support on Jetson Thor 2

2

.

The release includes a new layer of agent skills that automate developer tasks like Linux customization, memory optimization, and model benchmarking, reducing tasks that previously took weeks to just days 2

2

. Solomon uses NVIDIA NemoClaw to coordinate AI agents on a humanoid robot, integrating reasoning, perception, sensor fusion, locomotion, and manipulation into a single workflow 2

2

. Advantech is deploying an agentic factory brain within its manufacturing facilities using NVIDIA NemoClaw, Nemotron 3, and Jetson Thor to enable AI-native operations 2

2

.

Open-Source Agent Tools for Physical AI Development

NVIDIA released a major collection of open-source agent tools and skills spanning NVIDIA Omniverse, Cosmos, Alpamayo, and Metropolis for robotics, autonomous vehicles, vision AI, and digital twins 3

3

. These physical AI skills turn complex training, evaluation, and deployment workflows into repeatable, optimized, and agent-executable instructions 3

3

.

Industry leaders including Agile Robots, Delta Electronics, Foxconn, Pegatron, PTC, and TSMC are using NVIDIA physical AI tools to accelerate development 3

3

. The skills cover the entire physical AI development pipeline, from generating perception and mobility training data to simulation, automating navigation training, and tuning Jetson-based edge systems for deployment 3

3

.

CUDA-X Libraries Become Agent Skills

NVIDIA introduced CUDA-X libraries as reusable agent skills, giving AI agents access to specialized capabilities without extensive training 5

5

. The plug-and-play skills include cuDF for processing massive structured datasets, cuOpt for solving complex routing and supply chain problems, AI-Q for intelligent routing with persistent context, NeMo for accelerating agent optimization and governance, PhysicsNeMo for scientific and engineering simulations, and CUDA-Q for quantum computing applications 5

5

.

Nemotron Models Power Long-Running Agent Intelligence

NVIDIA unveiled Nemotron 3 Ultra, a 550 billion-parameter mixture-of-experts model built specifically for long-running autonomous agents 4

4

. The model delivers frontier-level reasoning across coding and research workflows, with up to five times faster inference speeds and 30% lower running costs than comparable frontier models 5

5

. CrowdStrike and Palantir are transforming cybersecurity and operational decision-making with long-running AI agents powered by Nemotron open models 4

4

.

Startups are also extending the reach of agentic AI using NemoClaw. Flexcompute is applying OpenShell to its Tidy3D and PhotonForge agents for multiphysics co-packaged optics design, with autonomous workflows exploring thousands of design variants overnight 1

1

. PhysicsX is partnering with Microsoft Surface team to build an electronics thermal simulation agent that compresses weeks of manual CAE workflows into automated, AI-driven design cycles 1

1

. The agent automates the full thermal simulation lifecycle for consumer devices like Microsoft Surface laptops, from mesh sensitivity analysis through physics AI model training to continuous accuracy monitoring 1

1

.