Fujitsu developed the Kozuchi Physical AI 1.0 framework to integrate physical robots and AI agents with each other. Nvidia is behind the scenes.
Fujitsu launches Kozuchi Physical AI 1.0, a technology that seamlessly integrates physical and agentic AI to automate confidential business processes. The development is a first result of the collaboration with Nvidia. At the heart is a multi-AI agent framework
This makes it possible to automate confidential workflows via a visual interface. The system combines technologies from the Fujitsu Kozuchi AI platform with specialized AI agents based on the Takane Large Language Model. These agents can communicate securely with each other via a proprietary gateway. In this way, the framework supports the automation of complex workflows, even when sensitive data is processed.
Fujitsu also offers test environments for core functions of the platform. Thanks to integration with Nvidia NIM microservices, users can count on version management and automatic updates. After its own internal proof-of-concept within the sales department, Fujitsu found that the technology halved the workload for order confirmations, which should also be possible for compliance checks, among other things, the company claims.
Physical AI
In addition to applications within digital workflows, Fujitsu is also working on an expansion towards physical AI. In doing so, AI agents will be able to interact with the real world via physical robots. In collaboration with Nvidia, Fujitsu wants to further develop this technology for use in environments with high security requirements. The goal is to allow agentic AI to evolve into a self-learning system within customer environments.
According to Fujitsu, this is necessary to address the current limitations of the technology, such as automating complex processes across different departments. By developing specialized agents that can handle data securely and have insight into operational context, the company wants to lower those barriers.