Metal-organic frameworks (MOFs) provide a highly modular platform for adsorptive gas separation, yet their vast reticular design space complicates inverse design under constraints of chemical validity, separation performance, and structural diversity. We present the LEMO Agent, a large-language-model agent framework for closed-loop inverse design of gas-separation MOFs in MOFid space. LEMO Agent integrates language-based candidate generation, MOFid standardization, explicit validity checks, Transformer-based property prediction, structured design memory, and multi-island exploration. Through iterative generate-validate-evaluate-remember cycles, the agent leverages feedback from both successful and failed candidates to guide chemically constrained searches across linker, metal, and topology choices. We evaluate LEMO Agent on CH$_4$/N$_2$ and CO$_2$/N$_2$ separation tasks. Compared to representative generative, optimization, and agentic baselines, LEMO Agent enriches high-performing candidates, enhances predicted separation performance, and maintains broad chemical and topological diversity. Selected candidates are further reconstructed, evaluated by GCMC simulations, and filtered through an experimental down-selection workflow based on chemical feasibility and ligand purchasability, leading to initial wet-lab synthesis and SEM characterization. These results demonstrate that large language model agents can serve as interpretable and scalable design engines for accelerating MOF discovery beyond conventional fixed-library screening.
Blogger's Review: The introduction of LEMO Agent signifies a major advancement in the MOF design realm. By merging language models with structured search, it efficiently explores complex material spaces. This innovative approach not only enhances separation performance but also showcases the potential of language models in chemical design, hinting at new directions for intelligent material design in the future.