The technology of embodied conversational agents in virtual worlds provides an attractive way to achieve natural and realistic human-computer interaction if the interaction designs is handled sensitively. In view of this, we aim at building an embodied tour guide that is able to engage conversationally with system users about gallery exhibits, and capable of behaving according to social norms in terms of gesture and facial expression. The research has focused on the attributeds of agent autonomy and believability, not truthfulness. To achieve agent autonomy, we present a three-layered architectural design to ensure appropriate coupling between agent's perception and action, and to hide the internal mechanism from users. We then work towards agent believability: we utilize the notion of schema to support structured and coherent verbal behaviors; careful attention has also been paid to the design of non-verbal agent interactions that help to establish social facts within the virtual world. Our discussion focuses on how reasoning, planning and generation of the verbal and non-verbal behaviors are performed using the schema-based framework. Finally the agent-user interactions are illustrated with a case study.