mloda.ai
A plugin-based, declarative data-access layer where an AI agent or user describes the data it needs and mloda resolves how to fetch and transform it, tracing every result back to its source.
mloda separates what data is needed from how to get it, through three plugin types: FeatureGroup plugins define data transformations, ComputeFramework plugins provide an execution backend such as pandas or Spark, and Extender plugins hook in logging, validation or monitoring. Three roles use these plugins: providers who define them, users who request data, and stewards who govern execution.
Built-in lineage traces every result back to its source, and the same plugins are meant to work anywhere from a notebook to production. An LLM tool function lets an agent request data by describing what it needs rather than writing retrieval code itself. mloda positions itself against feature stores, semantic layers, DAG frameworks, data catalogs and ORMs, each of which it says solves only part of the same problem. It installs with pip and is licensed Apache-2.0.