Abstract. Searching objects within a catalog is a problem of increasing importance, as the general public has access to increasing volumes of data. Constraint programming has addressed the case of searching databases of complex products that can be customized and build from components, through constraint-based configurators. We address the issue of searching objects within catalogs of simpler items for which no logic description is available. We propose to embed constraint technology in a search assistant supporting dynamic and concise dialogs based on the exchange of constraint formulations between the client and the server. At each iteration, the server analyses the data available in the catalog, computes abstractions, cluster decompositions and relaxations in order to provide the user with alternative for either explicitly focusing the search (adding constraints to the user’s wish) or enlarging it (relaxing, to some extent, a subset of the user’s constraints). This cooperative s...