A new open-source project called Rocky wants to fix what's broken about managing SQL pipelines. Built in Rust by data engineer Jacob Sensiba, Rocky sits on top of existing data warehouses like Databricks and Snowflake. It handles the messy parts: dependency management, schema drift detection, tracking where your data actually comes from.
The dbt comparison is unavoidable, but Rocky takes a different approach to correctness. Instead of Jinja templates and manifest files, it uses a type-safe compiler that catches errors before any data moves. A misspelled column name or type mismatch fails at compile time, not three hours into a production run. Column-level lineage traces a single field back through its transformations without reading every model. And branches let you test changes in isolation, see the downstream impact, then promote or drop them.
The AI angle is practical. Rocky includes a natural language interface that generates SQL models, but with a twist: it compiles and validates the output in a loop, retrying if the first attempt fails. The system flags contract violations and schema drift as diagnostic codes, not runtime surprises. Hacker News commenters have already asked for ClickHouse support and model versioning, signs that data engineers are paying attention. The project is Apache 2.0 licensed and includes integrations for Dagster and VS Code.