This paper presents an implemented architecture for easy learning, reorganizing and navigation into a Boolean knowledge base. As the base grows with new definitions and constraints, it is normalized by the closure of a completion operator. This normalization allows arbitrary formats for Boolean expressions. It ensures basic reasoning abilities and spontaneously organizes intermingled taxonomies of concepts. User interaction is done with three operators~: activation, inhibition and forced contradiction. Complex constraints, such as instantiations of first-order expressions, may be programmatically added. A complete illustrative scenario is developed around a knowledge base on elementary school geometry and the approach is situated in the AI field.