Stress in the workplace is a significant challenge people face in their everyday lives that affects their well-being, and productivity. Large Language Models (LLMs) may function as complementary tools in a wider, work-related, stress management approach, helping occupational mental health professionals handle employees’ stressors. Our work explores the use of LLMs in a stress management recommendation system that accepts smartwatch-derived stress indicators and user feedback from scientifically verified questionnaires to provide results. The system employs Retrieval-Augmented Generation (RAG) to access a curated library of stress management literature, delivering evidence-based recommendations. Our experiments highlight how the input form affects producing grounded results and compare the produced recommendations using different library compositions. While the presented approach may only be used under supervision by mental health experts, it may also help in promptly identifying certain cases that require immediate attention while providing straightforward guidance for milder cases.