Agents have emerged as a major breakthrough in human-computer interaction and have been widely adopted across multiple sectors. However, despite their success, Arabic-language agents remain relatively scarce due to the inherent linguistic complexity of Arabic. To address this challenge, we propose an innovative approach that combines Large Language Models (LLMs) with a Retrieval-Augmented Generation (RAG) knowledge graph. Our pipeline begins with the annotation of a portion of the dataset using LLMs, followed by AraBERT for efficiently extracting key medical entities such as symptoms, diseases, and treatments. These extracted entities serve as the foundation for constructing a structured RAG-based knowledge graph, which acts as a robust repository of medical knowledge. This RAG-enhanced graph provides rich contextual information that significantly improves the quality of responses generated by the language model, ensuring greater precision and relevance to users’ specific needs. The final stage of our pipeline involves leveraging the LLM in conjunction with the RAG graph to generate accurate, context-aware, and reliable medical responses. This hybrid approach represents a major advancement in the development of Arabic medical conversational agents.