What are ephemeral models used for in dbt?

Ephemeral models in dbt are designed to perform transient transformations, which means they are temporary constructs that exist only within the scope of a single dbt run. When a model is defined as ephemeral, dbt compiles it into a Common Table Expression (CTE) that can be referenced by downstream models. This approach allows for efficient transformations that do not need to be stored in the database as permanent tables. The primary benefit of ephemeral models is that they enable complex data transformations to be broken down into manageable pieces without creating unnecessary physical tables in the database, thus optimizing storage and performance during the transformation process.

The other options pertain to different functionalities that dbt models can serve. Permanent database tables are typically created by tables or views in dbt, which store transformed data for consistent access. Aggregating large data sets is a common use case for more traditional model types but does not reflect the transient nature of ephemeral models. Similarly, managing historical data compliance is not the purpose of ephemeral models; rather, it is often associated with snapshot models or other mechanisms designed for archiving and tracking historical changes.