A Multi-agent system (MAS) is an open distributed system of agents. The agents in a MAS may form certain specific groups, where agents in each group jointly engage in solving a specific problem. Since problem solving in a specific domain requires agents to have the BDI set that is specific to that application domain, and several groups of agents with each group inheriting one set of BDI, may exist in a large system, the MAS is not only heterogeneous but also has a heavy overhead on the system execution. This complexity must be resolved at the architecture level so that in an implementation the complexity does not arise. In this paper we introduce a formal multi-layered component-based architecture towards developing a dependable MAS. Agent communication and configurations are also discussed. To illustrate our approach, we discuss a solution to a travel planning example using the techniques presented in this paper.