Fujitsu announced the development of Fujitsu Kozuchi Physical AI 1.0, a new technology designed to seamlessly integrate physical and agentic AI. The technology, which marks the first achievement of Fujitsu’s collaboration with NVIDIA, announced on October 3, 2025, integrates NVIDIA’s software stack with Fujitsu’s proprietary technologies.
As core functions offered through the new technology, Fujitsu has developed a multi-AI agent framework that enables secure automation of highly confidential business workflows and a set of specialized AI agents based on its large language model (LLM) Takane, to support the automation of procurement operations in purchasing departments.
Moving forward, Fujitsu will continue its collaboration with NVIDIA to evolve the newly developed technology for sovereign domains. By the end of its fiscal year 2025, Fujitsu plans to transform the technology into an agentic AI foundation where AI autonomously learns and evolves within customer environments. The technology will then be expanded into the physical AI domain, enabling AI agents to directly interact with the real world through physical robots. Fujitsu envisions a society where AI agents and robots seamlessly collaborate to perform complex tasks based on a deep understanding of real-world operations.
Fujitsu remains committed to advancing research and development to address diverse customer needs and solve challenges in specialized business areas, thereby unlocking new possibilities for enterprise utilization of agentic and physical AI.
Addressing the limitations of agentic AI
Agentic AI has seen remarkable advancements, yet its adoption by enterprises remains limited, particularly in complex, inter-departmental/inter-company workflows. Achieving advanced automation requires specialized AI agents tailored to specific industries and enterprises, capable of securely processing confidential information and ensuring workflow maintainability. Similar challenges exist in the realm of physical AI, where AI agents interact with the real world via robots, necessitating secure information sharing and processing in physical environments combined with a robust understanding of on-site operations.
Fujitsu Kozuchi Physical AI 1.0 was developed to address these challenges. It integrates its technologies with NVIDIA NIM microservices (NIM), which enhance maintainability by providing version control and update functionalities. Fujitsu is now offering trial environments for the two core functions within the new technology.
- Multi-AI agent framework for secure automation of highly confidential business workflows
- Multi-AI agent framework provides a visual UI for building business workflows
- Leveraging Fujitsu Composite AI, the framework automatically combines NIM-compatible core technologies from the Fujitsu Kozuchi AI platform with specialized Takane agents
- Enables the rapid construction of highly maintainable business workflows using multiple AI agents, and ensuring secure workflows through Fujitsu secure inter-agent gateway
- Specialized AI agents to support automation and efficiency in enterprise procurement operations
- Three new specialized Fujitsu Kozuchi AI agents based on Fujitsu’s Takane LLM enable the automation of procurement workflows in enterprise purchasing departments.
- Multiple agents support procurement regulation analysis, and compliance checking:
Document comprehension agent: accurately interprets complex documents and converts them to structured data
Procurement regulation analysis agent: analyzes regulations to generate compliance check prompts
Compliance checking agent: automated compliance verification using structured document data and prompts, with verified requests for quotations securely sent to external suppliers via Fujitsu secure inter-agent gateway
A proof-of-concept within Fujitsu’s purchasing department demonstrated these agents reduced order confirmation workload by approximately 50%; NIM compatibility is also expected to boost inference speed by 50%, significantly accelerating hundreds of daily internal regulation compliance checks.
Figure: Procurement workflow construction by multi-AI agent framework
If you have an interesting Article / Report/case study to share, please get in touch with us at editors@roymediative.com roy@roymediative.com, 9811346846/9625243429.
