The introduction of large language models in recent years transformed the use of artificial intelligence in our daily life, making it available to common users and allowing them to interact with natural language processing machines. However, the common and general knowledge of large language models is not reliable when it comes to the quality of the response and its veracity. The proposed work integrates the use of retrieval-augmented generation with large language models with a multi-agent system framework to enable the creation of trees of knowledge where multiple agents have domain specific knowledge. The proposed agent-based solution was tested in a realistic case study where multiple knowledge domains are distributed amongst six different agents. The case study evaluates the capacity of agents to delegate questions according to its topics and the capacity of correctly replying to questions. The results are promising, demonstrating the ability of large language models to assess and process information that wasn’t used in its training process.