The growing adoption of wearable fitness devices and health applications has led to an exponential increase in fitness recommendations. However, privacy concerns remain significant barriers to user trust and regulatory compliance. Federated Learning (FL) offers a privacy-preserving paradigm by training models across decentralized devices without exposing raw data. However, FL introduces new challenges, including data heterogeneity, computational overhead, and the need for explainable AI (XAI). This work presents XFL, an integrated, explainable FL approach for personalized fitness recommendation systems. Our approach integrates FL with XAI techniques, SHAP, and LIME, to enhance transparency and interpretability while preserving privacy. By leveraging global and client-specific explanations, our framework empowers users to understand the rationale behind personalized recommendations, fostering trust and usability. Experimental results demonstrate that XFL performs better than centralized models while maintaining strong privacy guarantees. Furthermore, we evaluated the computational impact of integrating XAI in FL environments, providing insights into the efficiency of different explainability techniques. Our findings contribute to developing user-centric, privacy-aware, and interpretable AI-driven fitness solutions.