Agent-based models have proved their utility to study complex systems. Nowadays, they are used to tackles numerous issues: disease spreading, urban growth, car traffic, climate change adaptation, etc. A recent tendency is the development of more and more realist models. However, if models have made a lot of progresses concerning the integration of data, the agents used to represent the different actors influencing the dynamic of the system (inhabitants, decision makers...) are often simplistic (reactive agents). Yet, for some socio-ecological models, being able to integrate these cognitive agents, i.e. agents able to make complex reasoning such as planning to achieve their goals, is mandatory to improve the realism of models and test new scenarios. A classic paradigm to formalise the internal architecture of such agents is the BDI (Belief-Desire-Intention) paradigm. This paradigm allows to design expressive and realistic agents, yet it is rarely used in simulations. One explanation is that most agent architectures based on the BDI paradigm are complex to understand and to use by non-expert in Artificial Intelligence. Moreover, they are often very time-consuming in terms of computation and thus not adapted to simulations with thousands of agents. In this work, we propose a new BDI architecture integrated into the GAMA platform. GAMA is an open-source modeling and simulation platform for building spatially explicit large-scale agent-based simulations. It provides modelers with a complete modeling language (GAML) that makes it usable even with low level programming skills. Our architecture allows to directly and simply define the agent’s beliefs, desires and plans through the GAML language. A case-study concerning the impact of a technological hazard on drivers in the city of Rouen (France) was carried out. This case study showed that our plug-in allows to simply define rich cognitive agents and to simultaneously simulate several thousands of agents.