Konnect-linK Co., Ltd. Fully Launches LLM/AI Agent Construction Solution for Closed and On-Premise Environments

Konnect-linK Co., Ltd. (Headquarters: Chiyoda-ku, Tokyo; Representative Director: Kento Komoda; hereinafter "the Company") has officially launched its LLM (Large Language Model)/AI agent construction solution for closed and on-premise environments, as the importance of AI governance and addressing security risks increases with the expansion of generative AI utilization in companies.

This solution systematizes and provides the knowledge and know-how regarding LLM/AI agent construction in closed and on-premise environments that the Company has been developing through internal research and development for over a year and providing to clients, including prime market listed companies. It enables the construction of an environment where generative AI can be utilized without sending confidential data outside the internal network, providing end-to-end support from AI strategy formulation to infrastructure design and construction, business implementation, and operational improvement. This will allow companies in industries such as finance, manufacturing, and healthcare, which require high information security standards, to confidently promote AI utilization.

1. Background: Manifestation of Information Leakage Risks from Generative AI

While generative AI significantly contributes to improving corporate productivity, information leakage risks associated with its use have recently become a serious management issue.

(1) "Shadow AI" Risk Caused by Lack of AI Governance

In many companies, cloud-based AI tools are being used based on individual judgment by employees and contractors, among others, without clear rules or governance structures for the business use of generative AI.

According to IBM's "Cost of a Data Breach Report 2025," data breaches caused by "shadow AI" used without corporate approval were confirmed in 20% of all surveyed companies *1.

(2) Structural Risks Associated with Cloud-based AI Usage

Cloud-based generative AI services carry various security risks (e.g., use of confidential information as training data/information leakage risks) that can arise when user-inputted data is transmitted to external servers.

In fact, a survey by Assured Inc. targeting 300 information system department personnel from major companies with 1,000 or more employees found that over half (58.5%) of companies using SaaS had experienced security incidents such as information leakage caused by AI or AI-powered SaaS *2.

Furthermore, a survey by Gartner Japan predicts that half of all corporate security incidents will be caused by AI-powered applications after 2026 *3.

(3) Privilege Management Risks with the Advancement of AI Agent Utilization

While many companies recognize the usefulness of LLMs, the introduction and consideration of AI agents are also progressing in recent years as a means to achieve more advanced business efficiency and automation. Compared to traditional LLM usage that responds to explicit user instructions, AI agents are expected to autonomously or semi-autonomously process tasks while connecting to multiple tools, business systems, and internal data. While this improves convenience, it also entails a structural risk of being granted broader privileges.

As a result, if misconfigurations, inappropriate privilege designs, or unintended instructions from external data occur, the impact on confidential information, internal data, and core systems could be greater than with standalone LLM usage. Therefore, when utilizing AI agents, it is necessary to address stronger AI governance, including strict privilege management, auditing, and approval processes, even more so than with LLMs.

(4) Acceleration of On-Premise Adoption in Various Industries Due to Stricter Regulations

Against the backdrop of these risks, the transition to on-premise LLMs, which allow AI to be used while keeping data within the company's own environment, is rapidly progressing, especially in industries handling highly sensitive information such as finance, manufacturing, and healthcare.

In response to these market changes, the Company will officially launch its LLM construction solution for closed and on-premise environments.

2. Solution Overview: On-Premise LLM/AI Agent Construction Solution

Our on-premise LLM/AI agent construction solution provides end-to-end support for all phases necessary for generative AI utilization in a company's closed environment, from AI strategy formulation to infrastructure construction, business implementation, and operation/improvement.

*This service has been under development, including research, for over a year and is already being deployed to prime market listed companies, including financial institutions.

〇Main Service Lines

●AI Strategy & Concept Formulation: Business challenge organization, AI utilization roadmap formulation, ROI calculation, support for AI strategy formulation for management.

●AI Governance Design: AI usage policy formulation, shadow AI countermeasures, risk assessment framework construction, guardrail design (output verification, prompt injection countermeasures, etc.), privilege management and control design for AI agent utilization.

●On-Premise LLM Infrastructure Construction: GPU server design and construction in closed/on-premise environments, LLM model selection and introduction, model optimization through quantization, RAG (Retrieval Augmented Generation) environment construction, API design and business system integration.

●Business Implementation (Social Implementation): AI integration into business processes, user flow design, business stabilization support, implementation of next-generation AI integration utilizing MCP (Model Context Protocol), A2A (Agent-to-Agent Protocol), etc.

●Operation & Improvement: Continuous monitoring of model accuracy, standardization of evaluation design (benchmarks, verification procedures), fine-tuning support, operational cost optimization.

●Security & Safety Verification: Vulnerability assessment, data access control design, prompt injection resilience testing, compliance suitability verification, access control and external linkage risk verification for AI agent utilization.

Example of implemented models (a partial list focusing on public models)
・Qwen3-Next-80B-A3B-Thinking
・gpt-oss-120b
・Llama-3.1-Nemotron-Ultra-253B-v1

*The above is an example, and the optimal model will be selected based on customer requirements, usage environment, license conditions, etc.

3. Our Strengths

(1) End-to-End Support from Foundational Technology to AI Strategy Formulation/Business Review/System Development

The Company specializes in strategy formulation and business development, and for LLM implementation in closed and on-premise environments, we can provide end-to-end support from foundational technology development to AI strategy formulation, business review, and system development. We are equipped to allow customers to select the necessary areas according to their challenges and objectives, not just limited to technology introduction, but also considering integration into business operations and continuous improvement.

(2) Track Record in AI Infrastructure Construction, Including Closed and On-Premise Environments

The Company has traditionally handled AI infrastructure construction in closed environments isolated from the internet and on-premise environments for clients, including prime market listed companies. We have accumulated know-how through implementation実績 for major companies with strict security requirements, and this is integrated into this solution.

4. Anticipated Target Companies/Organizations for Introduction

This solution primarily targets companies and organizations with the following challenges and requirements:

・Companies that have advanced generative AI introduction and want to fully establish security and AI governance systems.

・Companies that want to develop and enhance internal rules and control systems with the expansion of AI utilization.

・Organizations such as finance, manufacturing, healthcare, and government agencies that require high information security standards.

・Companies that require AI utilization in closed/on-premise environments due to handling confidential and personal information.

・Companies that are considering AI agent introduction and want to establish a secure implementation infrastructure, including privilege management and internal controls.

・Companies that want to move beyond PoC and trial introductions to build an operational system for company-wide deployment.

・Companies that want to promote controlled generative AI utilization, including countermeasures against shadow AI.

5. Representative's Comment

6. Company Overview

・Konnect-linK Co., Ltd.

　Location: 5F Sotokanda S Building, 5-2-1 Sotokanda, Chiyoda-ku, Tokyo

　Representative: Kento Komoda

　Established: May 2022

　Business Activities: New business development, AI/IT support, system development, management support

　URL: https://konnect-link.co.jp/

　Inquiries: [email protected]

Sources, etc.

*1 Cost of a Data Breach Report 2025 (IBM)