Climate change, driven by greenhouse gas emissions, and the widespread issue of hidden malnutrition represent two compelling and interconnected global challenges. Both are exacerbated by the increasing consumption of ultra-processed foods, which are high in energy density but low in nutritional value. For these reasons, a fundamental shift in eating habits is essential. However, changing something so deeply tied to our identity and daily routines is difficult without the right information and motivation. This work presents an approach to encouraging healthier and more environmentally sustainable eating habits through an LLM-based conversational agent called E-Mealio. Its objective is to help users better understand the impact of food on both climate change and human health, while suggesting new recipes to gradually build a more sustainable and healthy lifestyle. The agent is backed by a comprehensive dataset of recipes and ingredients, each enriched with sustainability indexes computed from reliable sources, as well as detailed nutritional information. This ensures data reliability and minimizes the risk of generating misleading content that could confuse users. Additionally, the conversation flow is structured through well-defined steps that trigger specific functionalities, reinforcing the agent’s role as a specialized tool within its domain. The experiments show that users who used the system reported positive satisfaction with both the user experience and the quality of the recommended recipes. Furthermore, the results indicate that the choice of the underlying LLM, provided it is sufficiently capable, does not significantly affect user perception, reinforcing the idea of a model-agnostic architecture.