To model complex systems, sofiware agents need to combine cognitive abilities to reason about complex situations, and reactive abilities to meet hard deadlines. We propose an operational hybrid agent model which mixes well known paradigms (objects, actors, production rules and ATN) and real-time peflonnances. This agent model combines cognitive and reactive modules. The cognitive module uses a rule-based system and is provided with a synchronization mechanism to avoid the possible inconsistencies of the asynchronous execution of several rule bases. The reactive asynchronous perception and communication modules allow the agent to dynamically adapt its behavior to changes in the environment. To manage the interactions between these fundamentally diflerent modules (reactive and cognitive), we use an ATiV-based module.