Each table in a semantic model has a storage mode that determines how data is loaded and queried. Semantic Modeler lets you view and change the storage mode for any table directly from the table properties.
Available Modes
Mode
How It Works
Import
Data is loaded into memory during processing. Queries are fast because they run against the in-memory copy. This is the most common mode.
No data is stored in the model. Each query is sent to the underlying data source in real time. Useful when data must always be current or the dataset is too large to import.
Reads data directly from Delta tables in OneLake (Microsoft Fabric). Combines the performance of Import with the freshness of DirectQuery by accessing Parquet files without a full data copy.
Changing the Storage Mode
Select a table in the Explorer or Diagram View.
Open the Properties panel.
Change the Mode property to the desired storage mode.
Things to Keep in Mind
Changing a table from DirectQuery to Import is straightforward. Going from Import to DirectQuery has restrictions because downstream tables and relationships may depend on the imported data being available in memory.
Direct Lake mode is only available for models deployed to Microsoft Fabric. It is not compatible with SSAS or Azure Analysis Services. See Direct Lake overview for details.
Changing a table to Direct Lake mode will upgrade the model's compatibility level to 1604 or higher. This is expected and required for Fabric. See Develop Direct Lake semantic models for compatibility level requirements.
In composite models, you can mix Import and DirectQuery tables.