Large language models (LLMs) have achieved multimodal perception across vision, audio, and text. Olfaction remains the only major human sensory channel without a corresponding digital input modality for LLMs. This paper proposes the Molecular Olfaction Architecture (MOA), a conceptual framework in which a molecular detection layer identifies volatile organic compounds (VOCs) and passes them as structured input to an LLM, which performs semantic reasoning to produce a natural language description of the detected scent. An informal proof-of-concept demonstrates the viability of the reasoning layer. Limitations and directions for future empirical validation are discussed.