We propose a deductive database model containing two kinds of rules: update rules of the form L0←L1, where L0 and L1 are literals, and query rules of the form of normal logic program rules. A basic feature of our approach is that new knowledge inputs are always assimilated. Moreover, updates are always deterministic and they preserve database consistency. We consider that update rules have higher priority than query rules, i.e., update rules may generate exceptions to query-driven derivations. We introduce a semantics framework for database update and query answering, based on the well-founded semantics. We also suggest an alternative approach based on extended logic programs and we show that our database model can be defined in terms of non-monotonic formalisms.