The growing global burden of chronic conditions like diabetes and osteoarthritis necessitates innovative, scalable rehabilitation strategies. This paper proposes a conversational AI system designed to deliver dynamic, personalized physical rehabilitation plans. The system leverages large language models (LLMs) and physiotherapist supervision to adapt recommendations based on user feedback and physiological data. A key novelty of this work is its avatar-based AI agent, designed with the potential for integration into extended reality (XR) environments, thus aligning with emerging AR glass technologies. We present an early-stage prototype supported by a modular architecture and propose a pathway toward implementation and evaluation to enhance patient adherence, safety, and engagement in long-term rehabilitation.