The system we present is an intelligent database assistant based on a knowledge level view of databases: When a user wants to retrieve data from a database, he first conceptualizes the question (usually expressed by means of other concepts common to persons doing the same job, or "trade concepts"), then translates this "mental" concept into a (Set) query. In our L/sub ex/ system, concepts are implemented in a hybrid language closed to mental concept formulation. Each well-formed concept in L/sub ex/ corresponds a safe extended relational calculus of tuples formula, and, therefore, a SQL query. In our system, a relational database is first reified (a "concept vision" of the database is built in L/sub ex/). Domain knowledge and user knowledge are expressed by a set of L/sub ex/ concepts. A query is viewed as computing the interpretation of its corresponding concept, which is done by (automatically) rewriting it into a SQL query.>