Lium acts as an agent harness that allows language models to work with massive, unstructured physical-world data sets from sectors like agriculture and climate science. It enables subject matter experts to query data using natural language, automatically generating repeatable workflows and tools that capture human domain expertise for scaled use.