Fujitsu and NVIDIA Advance Secure AI Agents With Future Path to Robotics

• Fujitsu launched a secure multi-agent AI system as a first step toward integrating digital workflows with real-world robotics.
• The system combines NVIDIA microservices with Takane-based agents, showing early results in automating procurement tasks.

Fujitsu has released Kozuchi Physical AI 1.0, a new technology developed with NVIDIA to integrate physical and agentic AI. It combines NVIDIA’s NIM microservices with Fujitsu’s proprietary AI systems, including the Takane large language model. The platform is designed for secure enterprise automation and is part of Fujitsu’s plan to extend AI from digital workflows to real-world robotics.

The system does two main things. First, it gives businesses a way to visually build automated workflows by linking together multiple AI agents, each focused on a specific task. These workflows are created using components that work with NVIDIA’s NIM microservices, combined with Fujitsu’s own AI models. Fujitsu’s Composite AI handles the setup, automatically choosing the right models for each step. A secure gateway manages how the AI agents exchange data, allowing them to operate across different systems while keeping confidential business information protected. The second part of the system includes specialized agents for procurement tasks such as interpreting documents, analyzing regulations, and performing compliance checks. Fujitsu says these functions are designed to support secure enterprise automation today and will serve as the foundation for future systems where AI agents interact with the real world through robotics.

In testing within Fujitsu’s purchasing department, the system reduced order confirmation workload by approximately 50%. Once compliance was verified, requests for quotations were securely transmitted to external suppliers. With NIM integration, inference speeds are expected to improve by 50%, enabling faster processing of hundreds of internal compliance checks each day. Fujitsu plans to further develop the system into an adaptive AI foundation and extend it into physical environments using real-world robotics.

🌀 Tom’s Take:

Fujitsu’s agent framework shows how secure, specialized AI can move from the back office to the physical world.

Source: Fujitsu