Openclatura: Agentically Constructed, Verifiable Rules for Molecular Nomenclature

Agentic systems for molecular science increasingly rely on domain tools to ground language-model reasoning, but constructing reliable chemical tools from large bodies of expert rules remains challenging. We introduce Openclatura, an open-source and deterministic system for generating IUPAC names from molecular structures. Rather than learning the mapping from examples, Openclatura uses a rule set constructed through an agentic workflow in which candidate nomenclature rules are tested using executable chemical feedback: generated names are parsed back to molecular structures with OPSIN and checked for round-trip consistency. This produces a verifiable symbolic tool whose decisions can be traced to molecular-graph elements and nomenclature rules. Across QM9, PubChem, and ZINC22, Openclatura produces structurally valid names for $100\%$, $99.0 \pm 0.0\%$, and $96.5 \pm 0.1\%$ of molecules, respectively, outperforming the neural STOUT baseline while requiring only a single CPU core. We further expose the structured traces through Robomoleculographer, which converts molecular graphs and naming decisions into natural-language descriptions. Openclatura illustrates how agentic rule construction coupled to hard domain verification can yield transparent and reproducible components for molecular-science workflows.