The paper explores the application of recommender systems within a Large Language Model (LLM)-based Agent platform, emphasizing the unique characteristics of these agents—robust interactivity, intelligence, and proactiveness. It introduces a novel recommendation paradigm called Rec4Agentverse, which consists of two key components: Agent Items and Agent Recommender. Rec4Agentverse is developed in three stages to enhance interaction and information exchange among users, Agent Recommender, and Agent Items. - **Stage 1: User-Agent Interaction** involves the interaction between users and Agent Items, where Agent Items can engage in multi-round dialogues with users to exchange information and gather preferences. - **Stage 2: Agent-Recommender Collaboration** enables information exchange between Agent Items and Agent Recommender, allowing Agent Items to share user preferences with the Agent Recommender and receive new instructions. - **Stage 3: Agents Collaboration** supports collaboration among multiple Agent Items, facilitating the exchange of information and enhancing personalized information services. The paper also discusses potential application scenarios, such as travel, fashion, and sports agents, and highlights research topics like evaluation, preference modeling, efficient inference, knowledge update, and environmental friendliness. A preliminary case study demonstrates the feasibility of the Rec4Agentverse paradigm, showcasing its potential for practical implementation. The authors conclude by emphasizing the need for further exploration and quantitative research to fully realize the potential of Rec4Agentverse.The paper explores the application of recommender systems within a Large Language Model (LLM)-based Agent platform, emphasizing the unique characteristics of these agents—robust interactivity, intelligence, and proactiveness. It introduces a novel recommendation paradigm called Rec4Agentverse, which consists of two key components: Agent Items and Agent Recommender. Rec4Agentverse is developed in three stages to enhance interaction and information exchange among users, Agent Recommender, and Agent Items. - **Stage 1: User-Agent Interaction** involves the interaction between users and Agent Items, where Agent Items can engage in multi-round dialogues with users to exchange information and gather preferences. - **Stage 2: Agent-Recommender Collaboration** enables information exchange between Agent Items and Agent Recommender, allowing Agent Items to share user preferences with the Agent Recommender and receive new instructions. - **Stage 3: Agents Collaboration** supports collaboration among multiple Agent Items, facilitating the exchange of information and enhancing personalized information services. The paper also discusses potential application scenarios, such as travel, fashion, and sports agents, and highlights research topics like evaluation, preference modeling, efficient inference, knowledge update, and environmental friendliness. A preliminary case study demonstrates the feasibility of the Rec4Agentverse paradigm, showcasing its potential for practical implementation. The authors conclude by emphasizing the need for further exploration and quantitative research to fully realize the potential of Rec4Agentverse.