Skip to main content
Airform is Oxygen’s modeling engine. It turns raw tables into clean, tested ones with dbt-style SQL models that live in the same workspace as your agents and semantic model, so the tables your metrics build on are defined in code rather than assembled ad hoc. It works in three stages:
  • Compile resolves each model into plain SQL and works out the dependencies between models.
  • Build runs that SQL on the warehouse in dependency order.
  • Test checks the tables it built.
Airform architecture. A project following dbt conventions, with SQL models written in SQL and Jinja using ref() and source(), schema and data tests, and Git review and deployment, is compiled by Airform into plain SQL. Airform builds the models in dependency order and tests every built table, and tracks model- and column-level lineage from the compiled graph. A highlighted box notes that every table is built from SQL you can read before it runs. The output is clean, tested tables in a warehouse database from config.yml, such as Snowflake, BigQuery, Postgres, Redshift, MySQL, ClickHouse, DuckDB, or MotherDuck.

Airform compiles a dbt-style project into plain SQL, builds the models in dependency order on your warehouse, and tests the tables the semantic model relies on.

​
Why it matters

  • Transforms are code. Models are reviewed and deployed through the same Git flow as the rest of the workspace.
  • Tested foundations. Every model can carry tests, so the semantic model compiles over tables that are checked, not assumed.
  • Readable and traceable. Every model compiles to plain SQL you can read before it runs, with model- and column-level lineage across the project.
  • No separate tool. Airform follows dbt conventions, so existing projects carry over with one added mapping file, and without a separate dbt installation or Python environment.

​
Where it fits

Airform builds models on the warehouses already connected to the workspace, including Snowflake, BigQuery, Postgres, Redshift, and DuckDB, and its output is what the semantic model defines measures over.

Modeling reference

Project layout, configuration, operations, and supported databases

The semantic model

Build measures over the tables Airform lands