In the fast-paced context of Generative Artificial Intelligence (GAI) applied to intelligent interfaces, the implementation of Large Language Models (LLM) have emerged among the most intriguing applications due to their development flexibility in many different domains. However, this approach brings with it peculiarities and challenges. The legislative topic, which is one of the most profitable applications for Artificial Intelligence (AI), immediately showed the advantages and the constraints of these technologies, especially in those cases where the conversational systems are very complex and advanced. The Crea3 project, building on the foundations of its previous version, has the primary objective of investigating potential challenges of LLM-based conversational agents in the specific field of conflict resolution exploiting legal knowledge bases from different European countries. This study presents the progress of this initiative following a line of continuity that joins the technical development of the system and the evaluation of its functioning in an uncontrolled environment, highlighting notable positive aspects but also strong weaknesses thus offering possibilities for future developments and improvements.