Intelligent agents can be classified as internal and external based on user authority. Because of the non-interoperability of information permissions, external agents have to send information requests to users to provide personalized services. In this process, an interaction scenario containing a user and multiple agents is naturally formed. The interaction between one user and multiple agents cannot be understood simply as an overlay of single agents. Agents’ relationships and interacting patterns affect users’ perceptions and behaviors. Therefore, we designed three types of information conveying styles—direct, representative, and triadic. This study explored how agents’ information conveying style affects users’ social perceptions of agents and their information disclosure tendency in a mock-up smart home scenario (N = 36). The results indicated that the representative conveying style was superior to the direct and triadic styles in fostering positive social perceptions, especially perceived warmth. In contrast, the triadic condition elicited the lowest ratings for competence and the highest levels of discomfort. When it comes to disclosure behaviors, the direct condition reported the highest levels of intimate disclosure from participants. These findings have implications for the design of multi-agent-mediated interactions.