This article presents a multi-agent dialogue system. We show how a collection of relatively simple agents is able to treat complex dialogue phenomena and deal successfully with different deployment configurations. We analyze our system regarding robustness and scalability. We show that it degrades gracefully under different failures and that the architecture allows one to easily add new modalities was well as porting the system to different applications and platforms. General Terms Algorithms, Design, Reliability. Keywords dialogue, natural language processing, agent-oriented software engineering, emergent behavior, scalability, robustness, performance, case studies.