[DigitalToday reporter Dae-geon Seok] Dinotisia said on Tuesday it released its AI agent knowledge platform, "AKB (Agent Knowledge Base)", as open source on GitHub. AKB is a platform that integrates and manages documents, files, databases and members' work records scattered across a company into a knowledge base that AI agents can use. It supports collaboration by enabling multiple departments and AI agents to share the same work context.

Use of generative AI has recently expanded beyond helping individuals write and search documents to real business processes such as development, sales, human resources and marketing. But internal corporate knowledge is scattered across document folders, collaboration tools, presentation materials and databases, making it difficult for AI to continuously grasp work context. As a result, employees face inefficiencies such as re-sharing completed materials, repeatedly confirming the same content with those in charge, or converting documents separately into formats AI can read.

To address these issues, AKB integrates different forms of information, including documents, files and database tables, into a single knowledge base. It applies an ontology-based structure that defines semantic relationships between documents and data, allowing AI agents to use not only individual materials but also the relationships among them. It is designed to go beyond a RAG (retrieval-augmented generation) knowledge repository for search and to accumulate and organize conversations, work logs, rationale for decisions and outputs generated by AI agents during work processes.

Technically, AKB supports MCP (Model Context Protocol)-based integration and standard markdown document management. It can manage various forms of content together, including SQL databases and object storage, and organically connects relationships among work knowledge through graph-based association definitions. Combined with search based on Dinotisia's vector database, "Seahorse", it can explore work context and related information that are difficult to find through keyword search.

In the AI agent field, interest is growing in long-term memory and knowledge repositories for agents such as LLM Wiki and GBrain. AKB extends this trend to corporate environments and is designed on the premise of permission management by organization, department and role. The company stressed that it reflected department-, role- and project-based permission management and controls on access boundaries between users as key design elements, considering both knowledge sharing and corporate security requirements.

Dinotisia plans to provide AKB free of charge to users for non-commercial purposes and to continuously upgrade its features based on feedback from developers and practitioners.

Chief Executive Moo-kyung Jung (정무경) said, "Corporate AI competitiveness is gradually shifting away from which model has been adopted to how well AI can use the data an organization holds." He added, "By releasing AKB as open source, Dinotisia will contribute so that more companies can convert their knowledge into assets for AI use and grow together with AI agents."