Skip to content

featlas

An implementation of the Feature Engineering Specification, covering forward engineering, reverse engineering and validation of feature definitions. Its first external target is Databricks.

A single Rust core holds all the semantics: parsing, schema and lineage validation, canonical identity hashing, planning and an exact local evaluator. The CLI, a read-only MCP server and the Python bindings all call the same API, so every interface returns the same results.

featlas on GitHub

Check the status table first

The README has a status table that shows what is done, what has been tested live and what is not implemented yet. Read it before you rely on featlas. The Databricks code generation has passed live acceptance on Databricks Free Edition. Azure Databricks has not been tested, and online serving is not implemented.

Parts

Part Role
featlas-core The Rust core, with no vendor SDKs
featlas-adapters Local reference, in-memory and Databricks (Spark SQL, Unity Catalog) targets
fe-spec CLI Command line and MCP server, with output rendered by rs-rich
featlas (Python) Bindings to the core, plus reverse engineering of existing feature tables