ChatDiet is an LLM-powered framework for personalized nutrition-oriented food recommendation chatbots. It integrates personal and population models with an orchestrator to retrieve and process relevant information, enabling dynamic and explainable food recommendations. The personal model uses causal discovery and inference to assess personalized nutritional effects, while the population model provides generalized food nutritional information. The orchestrator combines these models' outputs with the LLM to generate tailored recommendations. A case study demonstrated ChatDiet's effectiveness, achieving a 92% success rate in food recommendations. The framework emphasizes explainability, personalization, and interactivity, with the LLM providing detailed explanations and adapting to user preferences. ChatDiet's architecture includes an orchestrator, personal model, population model, and LLM. The personal model uses causal discovery to determine the impact of nutrition on health outcomes, while the population model provides general food knowledge. The framework's evaluation showed high accuracy in personalized food recommendations and effective explanations. However, challenges such as hallucination and limited data scope remain, requiring further research to enhance accuracy and adaptability. ChatDiet represents a significant advancement in leveraging technology for personalized dietary guidance.ChatDiet is an LLM-powered framework for personalized nutrition-oriented food recommendation chatbots. It integrates personal and population models with an orchestrator to retrieve and process relevant information, enabling dynamic and explainable food recommendations. The personal model uses causal discovery and inference to assess personalized nutritional effects, while the population model provides generalized food nutritional information. The orchestrator combines these models' outputs with the LLM to generate tailored recommendations. A case study demonstrated ChatDiet's effectiveness, achieving a 92% success rate in food recommendations. The framework emphasizes explainability, personalization, and interactivity, with the LLM providing detailed explanations and adapting to user preferences. ChatDiet's architecture includes an orchestrator, personal model, population model, and LLM. The personal model uses causal discovery to determine the impact of nutrition on health outcomes, while the population model provides general food knowledge. The framework's evaluation showed high accuracy in personalized food recommendations and effective explanations. However, challenges such as hallucination and limited data scope remain, requiring further research to enhance accuracy and adaptability. ChatDiet represents a significant advancement in leveraging technology for personalized dietary guidance.