Conversational agents are becoming increasingly intelligent due to artificial intelligence (AI) advances, improving human-machine interaction. As conversational agents are already being used in various real-world applications, research shows the enormous potential of exploring advanced areas such as companionship or learning support. Particularly, those intelligent and future-oriented application areas rely on using AI-based technologies. However, implementations are often prototypical without comprehensive technological information. Moreover, the possibilities for practical implementation of conversational agents are growing, making the choice of architecture and tools for implementation complex. We address this problem by developing a taxonomy that helps characterize AI-based tools and services for intelligent interactions. We then derive archetypes for AI-based tools using cluster analysis and give meta-architectures that exemplify how to integrate AI-based services into typical technical conversational agent architectures. With our research, we aim to expand the knowledge base for developing conversational agents and support developers in their implementation.