This work describes an agent interaction model (ACLT, Agent Communicating through Logic Theories) rooted in the concept of logic theory. ACLT agents and their behaviour are conceived as inferential as well as procedural activities within a multiple theory space. The communication unit (CU) abstraction is exploited, subsuming traditional communication models (both shared memory and message passing) based on explicit and extensional knowledge, while allowing agents to exploit partial/incomplete knowledge through deduction. Agent synchronization is reconducted to the concept of theory evolution, by allowing agents to wait for theory modification until facts can be deduced from a CU. Agent cooperation/competition is re-interpreted in terms of knowledge generation/consumption. A coherent notion of logic consequence in a time-dependent environment is proposed. As a result, the traditional dichotomy between reactive and symbolic systems is here exploited as a feature rather than a problem, leading to an integration of behavioural and planning-based approaches.