Skip to content

Home

Multi AI Agent Framework
Fujitsu Kozuchi evolves with the Multi AI Agent Framework. It transforms from a platform where humans use AI into a platform where agents autonomously leverage AI.

Challenges in AI Agents

  • For companies without AI expertise, even when trying to pilot AI agents, they lack a place to run agents and don't know how to build them.
  • Applying AI agents to various business domains requires AI experts to solve challenges in development, operation, and improvement phases.

With Multi AI Agent Framework

  • Provides a platform where you can easily try cutting-edge AI agents from our research labs
  • Provides an enterprise multi-AI agent development environment that even users with limited AI knowledge can use

Values of Multi AI Agent Framework

  1. Easy access to cutting-edge agents from research labs
    • Advanced technologies from research labs are integrated as tools and agents for easy trial
    • Technologies previously provided through Fujitsu Kozuchi are also available as agents/tools
  2. Create original multi-AI agents by combining advanced technologies
    • Provides a low-code service that enables users with limited AI knowledge to build agent workflows
    • All provided technologies support A2A or MCP protocols and can be freely combined

Technology Overview

Target Industry/Users

Companies seeking to automate and streamline various operations such as customer support, software development, and manufacturing/logistics field work through AI agent adoption, but lacking expertise in AI and agents

Technical Challenges

  • For companies without AI expertise, even when trying to pilot AI agents, they lack a place to run agents and don't know how to build them.

  • Applying AI agents to various business domains requires challenges to be solved in the development, operation, and improvement phases, which typically require AI experts.

    Phase Challenges
    Development Establishing simple Multi-AI Agent System development and evaluation methods
    Collaboration with other Multi-AI Agent Systems
    Ensuring flexibility for dynamic tasks
    Operation Collaboration with other Multi-AI Agent Systems
    Ensuring flexibility for dynamic tasks
    Improvement Autonomous behavior improvement of Multi-AI Agent Systems

  • Therefore, the number of AI experts becomes a bottleneck for business domain application of AI agents and scaling takes time, requiring a platform that enables development of Multi-AI Agent Systems without dependence on AI experts for broad business domain application.

Solution

  • Provides a platform from Fujitsu Kozuchi where you can easily try cutting-edge agents from our research labs
    • Advanced technologies from research labs are integrated as tools and agents for easy trial
    • Technologies previously provided through Fujitsu Kozuchi are also available as agents/tools
  • Create original multi-AI agents by combining advanced technologies
    • Provides a low-code service that enables users with limited AI knowledge to build agent workflows
    • All provided technologies support A2A or MCP protocols and can be freely combined

Fujitsu Kozuchi Multi-Agent Framework

Fujitsu's Technical Advantages

  • Incorporates proprietary technologies for collaboration and quality assurance into the platform (planned)
    • Agentic Memory: Incorporates memory that mimics human-like memory, enabling agents to act based on past experiences
    • Monitoring & Routing: Autonomously improves agent workflows by repeating self-evaluation and improvement (including Human-in-the-Loop)
    • Collaboration: Coordinates multiple AI agents to maximize benefits while communicating toward goal achievement
    • Security: Monitors inter-agent communication to prevent confidential information leaks and hallucinations

Values Provided by Multi AI Agent Framework (Details)

Even users with limited AI knowledge can use this technology to specialize AI agents for each customer and improve workflow quality securely.

  • Domain Specialization: Agentic Memory that mimics human-like memory enables learning from people, robots, and other agents, executing workflows while growing in specific domains
  • Quality Improvement: By appropriately monitoring complex interactions throughout the workflow, supports workflow execution while considering diverse constraints

Multi AI Agent Framework Overview

Case Studies/Use Cases (Platform-loaded Agents)

Meeting Agent

The meeting agent joins Teams meetings and autonomously understands and solves issues based on participants' statements. Using conversation data from the meeting as input, it mimics the human discussion process of repeatedly converging and diverging, proposing tasks in two directions aligned with the discussion flow. Specifically, it ①extracts key statements, ②supplements the discussion context using inferred statement intentions, and ③instructs the agent to generate tasks that lead the discussion in two directions. The meeting agent automatically posts the obtained suggestions and graphs to Teams chat, supporting participants' decision-making and discussions. In one internal trial example, when asked "Why did the gross profit margin in August change significantly compared to July?", the meeting agent visualized a graph showing the proportion of each product in total product sales for each month, demonstrating that product sales proportions differed between July and August with bias toward certain products, providing a quantitatively grounded answer to the question.

Field Work AI Agent

The Field Work AI Agent supports worker operations by autonomously proposing improvement plans and creating work reports by analyzing camera videos installed at manufacturing or logistics fields with spatial recognition and referencing document information such as work instructions and rules. Various abilities required for field work support, such as the ability to understand spatial relationships between workers and objects, field-specific object recognition, and recognition of individual worker tasks, can be realized by adding them to the AI agent.

Auto Data Science Agent

Auto Data Science Agent aims to realize fully automated data science through natural language via a collection of AI agents that support the Model Context Protocol (MCP). Specifically, when users provide table data they want to analyze along with data analysis requirements in natural language, the agent automatically combines various AI agents for data science according to the requirements and derives appropriate answers. This enables users to easily perform advanced data analysis without detailed knowledge of data science or AI.

  • Demo Video