Live-streaming commerce refers to a form of e-commerce that uses live streaming as a channel to achieve marketing objectives, a product of the bidirectional integration of live streaming and commerce in the digital era. However, utilizing large language models for sales in live-streaming commerce faces the challenge of low interaction quality. In this paper, we introduce ChatLsc, a Vtuber designed for live-streaming commerce. Drawing inspiration from leadership behaviors in human societies, we assign different identities and capabilities to agents and encourage their collaboration. In the preparation phase, leaders plan and prepare for the upcoming broadcast based on initial information received from humans. During the live streaming phase, ChatLsc perceives audience comments in real-time through the collaboration of observers, creators, broadcasters, and leaders, combining these with past live streaming activities to generate visual and auditory feedback for the audience. We demonstrate how to use LLMs to interact with audiences in real-time live-streaming commerce, revealing the potential of LLMs for new commercial applications.