Domestic Private AI Platform "InfiniCloud® AI" Significantly Enhances AI Governance and Data Utilization with Ver2 Release Featuring Agent Functions

InfiniCloud Inc. (Headquarters: Shizuoka City, Shizuoka Prefecture, hereinafter referred to as InfiniCloud) has significantly updated its domestic private AI platform, "InfiniCloud® AI," designed to enable the utilization of corporate confidential information, personal information, and internal documents under user control without inadvertently exposing them to external AI.

The updated Ver2 includes various functional enhancements such as a new UI with improved operability, an "Evidence-Driven Thinking Mode (EDTM)" for evidence-based reasoning, Source of Truth (SoT) management for original information, Role-Based Access Control (RBAC/ACL), audit logs, and skill/workflow automation. It has evolved from an AI for PoC to an "enterprise AI platform for safely utilizing confidential information, personal information, and original internal data in business operations."

InfiniCloud® AI Chat Interface InfiniCloud® AI Management Screen (Intelligence Registration) InfiniCloud® AI Management Screen (User Group Settings) InfiniCloud® AI Management Screen (Permission Settings)

About "InfiniCloud® AI"

"InfiniCloud® AI" is a "domestic private AI platform" that supports environments from private clouds to fully isolated on-premises installations, designed to securely solve issues involving confidential information that is difficult to handle with large global public AIs, as well as company-specific contexts and local business challenges.

InfiniCloud® AI fundamentally operates by handling prompts, conversation logs, registered documents, reference data, and generated results entirely within the server in an environment managed by the user. Therefore, this structure prevents this information from being used for secondary purposes such as training shared models, user profiling, advertising, behavioral analysis, or service improvement. It is an AI that can be managed by the IT department and safely provided to employees.

It is particularly suitable for the adoption by the following organizations and companies:

Organizations with Strict Governance and Security Requirements

Financial institutions, medical institutions, local governments/public organizations, defense-related companies, universities/research institutions, and other organizations where the external transfer of information is restricted by laws, regulations, or infrastructure policies.

Companies Wishing to Protect Their Unique Intellectual Assets

Manufacturing design and development departments, companies handling patented technologies, companies dealing with pre-release management information, and other organizations that want to securely input internal data into AI and "own" it without hesitation.

IT Infrastructure Providers and System Integrators

Businesses that want to embed Private AI services into their own SaaS or data centers and offer them to customers as OEM or packaged solutions.

◼️ Key Features

Ensuring Data Sovereignty Through Complete Isolation (Air-Gap)

Supports operation in a closed network environment not connected to the internet, ensuring that prompts, internal knowledge, conversation logs, etc., are never transmitted to external cloud AIs or servers. *Exceptions apply when the web search engine function is enabled or when offloading to external remote LLMs. These can be configured by the administrator.

Elevation from "Knowledge (RAG)" to "Wisdom (Fine-Tune)"

Beyond simply answering by referencing manuals and regulations ("RAG" - Retrieval-Augmented Generation), it enables the development of "business-specific AI" that incorporates the organization's unique thought patterns and tone by performing fine-tuning (LoRA) using automatically generated Q&A datasets from internal data. *The fine-tuning here is a function used to improve AI responses within a "intelligence domain" that can only be used by users with access to explicitly specified data. It is structured such that it cannot be used for training/improvement of shared models provided by InfiniCloud to third parties, models for other companies, or user profiling.

Strict Permission Management (RBAC/ACL) and Internal Controls Optimized for Enterprise Use

Employs a two-layer structure of Role-Based Access Control (RBAC) to control "which screens can be opened and which operations can be performed," and Access Control Lists (ACL) to control "access to which intelligence domains and to what extent." Furthermore, it includes "step-up authentication" and audit log functions, supporting strict internal controls and governance.

Major Updates in InfiniCloud® AI Ver2

1. Revamped to a New UI for Intuitive Use, with Automatic Chat History Saving

More Intuitive UI Design

Significantly improved the appearance and operability of the login screen, sidebar, input fields, settings screen, and dark mode, creating a refined interface that anyone can use without confusion.

"Server Saving" for Unlost Conversation History

Supports secure automatic saving on the server side, transitioning from temporary browser saving. Past conversations can be accessed anytime, regardless of the computer, smartphone, or browser used. *Saved conversation history is retained in the user-managed environment and the memory used for saving is only for conversations with the user utilizing it, and will not be shared with third parties.

Command Completion with Just "/"

By simply typing "/" in the input field, available business automation menus and descriptions are displayed, allowing for smooth activation of automation programs (workflows) directly from chat.

2. Equipped with an Advanced Thinking Mode Where AI "Searches and Verifies Evidence" Itself

Evidence-Driven Thinking (EDTM Mode)

A highly intelligent thinking mode is now available, where the AI "autonomously gathers relevant evidence and constructs answers while verifying the content" for complex questions and specialized business challenges.

Enhanced Search to Not Miss Internal Documents You Want to Find

Strengthened hybrid search combining "vector search" for semantic meaning and "sparse search" for keywords. The inclusion of "Agentic RAG" (a function that automatically corrects search terms based on conversation context) allows for more accurate retrieval of desired information from vast internal documents.

3. Strict "Source of Truth" (SoT) Function to Protect Data Integrity and Advanced Analysis of Office Documents

Secure File Storage Exclusively for Users

Secure storage areas are added for each user, allowing original documents attached to chats to be securely managed and confirmed on the server side.

"SoT (Source of Truth) Management" to Ensure Data Accuracy

The system strictly manages the accuracy, version, approval status, and reliability of input data. This reduces the risk of AI providing answers based on "outdated manuals or incorrect drafts."

Improved Analysis of Voice and Images, and Reading Accuracy of Office Files

Significantly improved automatic transcription of audio files and character recognition (VLM/OCR) for scanned PDFs and images. Furthermore, it is now possible to read Word, Excel, and PowerPoint files with high accuracy while preserving the structure of diagrams and tables within them.

4. Load Balancing Across Multiple Servers and "Remote Integration" to Supplement Resource Shortages

Automatic Operation That Keeps the System Running

Supports load balancing that automatically distributes processing across multiple AI servers and rolling updates that allow system updates without any downtime.

Hybrid Operation with "Remote LLM Pool"

Even in small-scale environments with limited internal GPU resources, it is possible to combine external secure AI environments to function as a single brain, enabling cooperative operation. This allows for offloading some processing.

Examples of Usable Remote LLMs (/v1/chat/completions compatible)

When customers have a general-purpose LLM foundation within their LAN

"Sakura AI Engine," which explicitly states that input information is not saved in principle and not used for AI training, etc.

Cyberattack Countermeasures and Isolation of Secret Information

Implements defensive functions to block unauthorized access targeting the system's interior (SSRF attacks) during remote integration. Furthermore, critical secret information such as API keys required for connection is strictly isolated only within the core database, ensuring a secure design that prevents leakage to the outside or peripheral processes.

5. Expansion of AI Agent Functions "Skills/Workflows" to Dramatically Improve Efficiency of Administrative Tasks

Workflows: Automated Functions for Files, Folders, and Conversations

Equipped with functions to read specified files, files within folders, or conversations, and save the results processed by AI (skills) to a designated folder, etc.

Input formats supported include text, HTML, Markdown, Word, Excel, PowerPoint, PDF, images (png, jpeg, etc.), audio files (mp3, m4a, etc.), JSON, CSV, etc.

Output formats supported include text, HTML, Markdown, and Word/Excel file formats.

Skills: Functions for Automatic Processing of Read Data by AI

Allows for routine processing of tasks specified by workflows using AI.

Various files can be summarized or converted using AI skills.

Standard Implementation of Specialized Tools for Office Work

Possible to have AI aggregate information from multiple files (PDFs, Word documents, etc.) and output summaries, records, or documents.

Includes pre-built template files to support daily administrative tasks, such as a skill that compares two files and automatically creates a comparison table of new and old regulations, or a skill that creates meeting minutes in Word format from audio files or memos.

Definition of Custom Skills/Workflows Possible

Allows for the definition of unique workflows and skills for adopting companies. Enables the creation of Word reports through comprehensive judgment using AI from confidential company data files.

For example, by operating in an on-premises or private cloud environment, AI skills/workflows can be applied to tasks involving original data that is difficult to input into external AI, such as creating routine reports from daily work reports, aggregating and formatting measurement data (csv/json) from equipment and generating reports, or extracting necessary items from application documents containing personal information and internal documents. Processing targets, reference ranges, and output destinations are limited to those permitted by the administrator and are subject to permission management and audit logs under user control.

6. Enhanced Management Screen and Test Function to Eliminate "Connection Failures" During System Integration

Streamlined Management Operations

Internal members can be easily invited (including bulk invitations) from the screen, and policies for signup eligibility and templates for various notification emails can be conveniently edited from the management screen.

API Connection Test for Developers

Enhanced "OpenAI-compatible API" for integrating internal systems and company SaaS with AI. A button for one-click connectivity testing (checking if the connection is successful) has been added. This allows for immediate resolution of common development issues such as "unknown URL" or "cannot connect for unknown reasons" directly on the API issuance screen.

Inquiries Regarding This Product

"InfiniCloud® AI" is offered through system integrators (SIers) and OEM partners as an integrated "hardware + software" package or as a standalone "software subscription." For detailed capacity planning and PoC (Proof of Concept) implementation, please feel free to contact us via our website.

Company Overview

InfiniCloud Inc.

InfiniCloud flexibly combines "storage," "server," and "network" technologies through software, adhering to "Japanese quality" to provide customers with mission-critical, stable, and cost-effective cloud services.

Location: Gofurakan 5F, 2-1-5 Gofukucho, Aoi-ku, Shizuoka City, Shizuoka Prefecture Representative: Yasushi Taki, Representative Director & CEO Established: November 2001 Business: Cloud infrastructure leasing, hosting, cloud software development, network and server infrastructure construction, maintenance, and management services URL: https://infinicloud.com/