In order to design a large system for knowledge retrieval by “casual” users on a semantically complex domain, archeology, one must integrate several technological components: data bases, knowledge-based systems, interfaces, and architectures. This paper discusses briefly the main methodological, technical, and conceptual problems encountered in such an attempt, the partial solutions developed, and the architecture proposed for the system, which should be based on agents cooperating in a “weakly” coupled distributed structure.