NVIDIA announced the Factory Operations Blueprint (FOX), a reference design for building autonomous factory manager agents that continuously monitor real-time factory data and coordinate specialized AI agents and machines to address operational issues at scale during GTC Taipei at COMPUTEX on May 31 (31/05).

The blueprint is designed to help manufacturers connect machine signals, quality systems, work instructions, and operational alerts into a unified decision-making layer. FOX provides a framework for building factory manager agents that deliver centralized visibility across factory operations.

FOX Blueprint Connects Factory Systems and AI Agents

FOX enables developers to build secure factory manager agents that orchestrate specialized industrial AI agents responsible for quality control, material transport, and worker safety.

The blueprint is built using NVIDIA NemoClaw, AI-Q Blueprint, and NVIDIA Nemotron open models. It provides a customizable foundation for connecting factory systems, automating model development, and operating intelligent workflows at scale.

FOX is optimized to run on NVIDIA DGX Station, which is powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip.

According to NVIDIA, the system delivers 20 petaflops of FP4 performance and 748GB of coherent memory. It can run AI models with up to one trillion parameters locally.

The superchip combines the NVIDIA Blackwell Ultra GPU and NVIDIA Grace CPU through the NVIDIA NVLink-C2C interconnect to support high-speed communication between NemoClaw and AI models.

Blueprint Supports Training Automation and Intelligent Operations

FOX integrates with industrial data sources, machines, applications, and robot fleets. It can also connect with specialized agents developed by software providers through standard application programming interfaces and agent skills.

Using NVIDIA TAO skills, factory manager agents can automate the model training lifecycle. This includes identifying accuracy gaps, generating or sourcing training data, fine-tuning models, and redeploying them into production.

The blueprint also supports visual inspection, process compliance, and material transport workflows using NVIDIA open models and blueprints, including the NVIDIA Metropolis Blueprint for video search and summarization.

Real-time factory information can be visualized through operational digital twins built with NVIDIA Omniverse libraries.

Taiwan Manufacturers Begin Deploying FOX-Based Factory Manager Agents

Taiwanese manufacturers Foxconn, Advantech, Pegatron, and Wistron are the first companies to deploy autonomous factory manager agents using the FOX blueprint and NemoClaw.

Foxconn is using the blueprint to develop MoMClaw, a manufacturing operations multi-agent system. The platform connects sensors, machine signals, and digital systems with hundreds of specialized agents through a single agentic layer.

The company said MoMClaw provides plant managers and operators with real-time answers and action plans through a natural language interface supported by NVIDIA OpenShell privacy controls and safety guardrails.

Foxconn projects an 80% improvement in root-cause analysis time, a 15% increase in labor productivity, and a 10% reduction in machine failure rates.

Pegatron is building a factory manager agent that coordinates specialized agents for material transport, AI inspection, standard operating procedure guidance, and machine-to-machine coordination.

The company estimates the system can reduce asset redundancy costs by 15% through more efficient robot utilization and reduced reliance on standby equipment.

Advantech introduced AI Factory Brain, a multi-agent system led by a factory manager agent built with FOX and NemoClaw.

The company has deployed the system in its own factories to autonomously manage energy across HVAC and lighting agents and projects a 10% reduction in energy consumption.

Wistron is using the FOX blueprint together with NVIDIA Cosmos, NVIDIA Nemotron open models, and the NVIDIA Metropolis VSS blueprint to develop surface-mount technology agents that analyze production-line operations and support real-time root-cause analysis and quality control.

Technology Partners Develop Specialized Manufacturing AI Agents

Several companies are developing specialized agents using NVIDIA AI technologies and the NVIDIA VSS blueprint.

DeepHow is using the Metropolis VSS Blueprint and Cosmos 3 to develop a standard operating procedure agent for Foxconn that supports the assembly of Bianca boards for NVIDIA GB300 servers.

Running on NVIDIA RTX PRO Servers, the agent is designed to understand complex assembly motions and help improve first-pass yield by 3%, reducing rework and production waste.

Spingence is using NVIDIA Defect Image Generation skills, the NVIDIA Cosmos open vision language model, and the NVIDIA TAO Toolkit to develop a factory manager agent for Cooler Master.

According to NVIDIA, the system achieved 99.6% defect recall, reduced defect escapes by 78%, and increased inspection capacity by three times.

Overview: AI is using NVIDIA agent skills for defect image generation and NVIDIA Cosmos to support Amphenol through its Advanced GenAI Toolkit.

The toolkit generates synthetic defect data and deploys visual inspection AI models 12 times faster, reducing time to first inference to less than 30 minutes across more than 300 products.

Roboflow is using NVIDIA Cosmos to develop a model-building agent for Corning Fiber Optics that generates synthetic defect images when training data is limited.

The company reported near-perfect detection rates and demonstrated the potential to reduce daily manual image review.

NVIDIA Expands Access to Video Analytics AI Tools

NVIDIA also announced that Metropolis VSS Blueprint 3 is now generally available.

The latest version includes skills that allow external agents, including Claude Code, Codex, Hermes, and NemoClaw, to access VSS components and rapidly build and operate video analytics AI agents.

NVIDIA is also accepting registrations from those interested in being notified when the Factory Operations Blueprint becomes available.

PHOTO: NVIDIA

This article was created with AI assistance.

We make every effort to ensure the accuracy of our content, some information may be incorrect or outdated. Please let us know of any corrections at [email protected].