In our paper, we describe a new architecture of cooperative systems to im- prove information retrieval on the WWW. Two basic ideas emerge from our model : 1 rst, the use of metadata embedded within documents of the WEB in order to be explicit about the representation and the description of them; 2 second, a self-con guring framework based on generalized and specia- lized search tools in order to localize the knowledges and the competencies on the WEB, and to make them participate in the overall auto-organizing information process.