Most automated reasoning tasks with prac tical applications can be automatically reformulated into a constraint solving task. A constraint programming platform can thus act as a unique, underlying engine to be reused for mu ltiple automated reasoning tasks in intelligent agents and systems. We identify six key requirements for such platform: expressive task modeling language, rapid solving method custom ization and combination, adaptive solving method, user-friendly solution explanation, efficient execution, and seamless integration within larger systems and practical applications. We then propose a novel, model-driven, component and rule-based architecture for such a platform that better satisfies as a whole this set of requirements than those of currently available platforms.