Contemporary conditions of project activity management are characterized by a high degree of dynamism, uncertainty and competition for limited resources. This requires adaptive and intelligent approaches to decision support, especially in the context of project portfolio management. This article proposes an agent-based decision support model based on a multi-agent system (MAC), in which each object of the project environment (portfolio, project, task, resource) is represented as a software agent with autonomous behavior logic. The developed architecture and operational algorithms of the system ensure decentralized decision-making, coordination of agents’ actions and adaptation to changing conditions of project implementation. Special attention is paid to the formalization of agent types, scenarios of their interaction, events that initiate decision-making and appropriate algorithms for resource redistribution and task rescheduling. Model verification was conducted through comparison with traditional centralized approaches. The results confirm the effectiveness of the agent-based approach to increase adaptability, consistency and strategic validity of decisions in a multi-project environment.