With the rise of Generative Artificial Intelligence (GenAI) technologies, such as ChatGPT, humans are getting easier to interact with AI through natural languages, which reduces time and effort in obtaining information via Large Language Models (LLMs). Human-centered AI design and implementation are crucial for positive value co-creation between humans and AI. This research aims to engage humans with GenAI technologies to achieve knowledge sharing and co-creation. We tackled two significant challenges in human-GenAI collaboration for knowledge management: the absence of human-like, natural language-based knowledge management services to lower knowledge exchange costs and the lack of autonomous mechanisms to connect knowledge seekers and givers proactively. Leveraging GenAI’s natural language capabilities, a knowledge assistant will act as an intelligent agent, facilitating knowledge activities through interactions with its human master. A knowledge assistant can help explicate its master’s tacit knowledge and manage the explicit knowledge objects. The directory connects to other knowledge holders, which sustains its transactive memory of knowing what and who. A multi-agent system will also manage transactive memory by integrating individual knowledge maps into a collective system, fostering collaboration among knowledge holders and their knowledge assistants. We adopted an elaborated Action Design Research (eADR) approach to design, implement, and evaluate the proposed framework and prototyping service system and obtained five principles to meet the needs of human-AI collaborative knowledge management. This study contributes to human-centered AI service system innovation by addressing these challenges by developing ethical, cooperative, and efficient human-agent collaborative knowledge service systems.