Myriad of applications involve the interaction of humans with machines, such as reception agents, home assistants, chatbots or autonomous vehicles' agents. Humans can control the virtual agents by the mean of various modalities including sound, vision, and touch. In this paper, we discuss about designing engaging virtual agents with expressive gestures and prosody. We also propose an architecture that generates upper facial movements based on two modalities: speech and text. This paper is part of a work that aims to review the mechanisms that govern multimodal interaction, such as the agent's expressiveness and the adaptation of its behavior, to help remove technological barriers and develop a conversational agent capable of adapting naturally and coherently to its interlocutor.