Deep dive into Direct Lake on OneLake and creating Direct Lake semantic models in Power BI Desktop

In March 2025, we announced the public preview of creating Direct Lake semantic models in Power BI Desktop. Microsoft Fabric’s OneLake data is visualized in Power BI without duplicating data using the new Direct Lake storage mode. Power BI semantic models with Direct Lake tables can give you the latest data from the OneLake to visualize insights quickly and dynamically in Power BI reports and provide the right context for success with Copilot. Now, it’s even easier than ever, as you can create and edit Direct Lake semantic models in Power BI Desktop. And, for the first time with Direct Lake models, you can add tables from multiple data sources, giving you the full flexibility to use the OneLake data in Power BI.

Let’s get started by demonstrating how to create these Direct Lake semantic models in just a few clicks.

First, if you haven’t done so already, in the Preview features of Options select Create semantic models in Direct Lake storage mode from one or more Fabric artifacts.

Then, in a new Power BI Desktop instance, select a Lakehouse or Warehouse from the OneLake catalog, then Connect.

Give the semantic model a name, pick a workspace, and select the tables you want to include then OK.

The semantic model is created in the service, accessing the data from the OneLake storage and now you are live editing the semantic model in Desktop, easy as that!

To bring in other tables from another Lakehouse or Warehouse, return to the OneLake catalog.

Like when you created the semantic model, just pick a Lakehouse or Warehouse and click Connect.

This time you are already in a semantic model, so the name and workspace options are removed, just pick the tables and click OK.

And it’s added to the semantic model!

From here you can continue data modeling: add relationships, measures, calculation groups, hierarchies, and more. DAX query view is available to view data in the tables and to try out calculations. TMDL (Tabular Model Definition Language) view is also available to make changes using code.

The tables are all stored in the same OneLake, so regular relationships can be created between the tables, similar to the import experience.

To create a report, go to File > Blank report and then live connect to the semantic model you are also editing. To find it, go to OneLake catalog > Power BI semantic models and it should be at the top of that list as you were just editing it, then click Connect.

This gives you two instances of Power BI Desktop, one live editing the model and the other editing the report with a live connection. If you have multiple monitors, or a single large monitor, you can now edit them side by side.

You can save the report PBIX as you would any live connected report and publish when ready.

For the model being live edited in Power BI Desktop, there is no local PBIX file created as the semantic model is already in the workspace. You can choose to export to Power BI Project to have a local copy of the metadata.

When created in Desktop, these Direct Lake tables are the new flavor of Direct Lake, called Direct Lake on OneLake. The existing Direct Lake, now called Direct Lake on SQL, behaves just like Direct Lake on OneLake when accessing data from the OneLake delta tables. The difference is what they can do in addition to Direct Lake mode.

Direct Lake on OneLake never uses DirectQuery to access data. Direct Lake on SQL also can talk to the SQL endpoint using DirectQuery. Views are accessed using DirectQuery mode, not Direct Lake, unless they materialized as delta tables.
Direct Lake on OneLake is multi-source. You can use multiple Lakehouse or Warehouse tables in the same semantic model. Direct Lake on SQL is single source.
Direct Lake on OneLake permission is only dependent on each source itself. Currently, this is either a Lakehouse or Warehouse. Use the ‘ReadAll’ permission to enable access to the delta tables. Shortcut tables can only be accessed with OneLake security early access. More information about shortcut tables is described below. Direct Lake on SQL permission is dependent on the SQL analytics endpoint of the source. Use the ‘ReadData’ permission to access delta tables through the SQL endpoint.
Direct Lake on OneLake semantic models are created and edited in Power BI Desktop. Support for creation and full editing in web is planned. Limited web modeling is supported at the start of the public preview. Direct Lake on SQL semantic models are created from the web in the Lakehouse or Warehouse by clicking New semantic model and can be edited in either web or Power BI Desktop.

Direct Lake on SQL does and will continue to have the fallback to DirectQuery option to be able to utilize the SQL endpoint. There is no more fallback to DirectQuery available when you create a Direct Lake on OneLake semantic model in Power BI Desktop. This Direct Lake storage mode only connects to the OneLake tables and is not downstream or impacted by the SQL analytics endpoint. The Direct Lake behavior option, found in Model view > Data pane > Model explorer > Semantic model node properties pane, will be greyed out to indicate this new Direct Lake on OneLake storage mode.

Direct Lake on OneLake doesn’t show or allow the use of views in the semantic model, unless they materialized as delta tables. Direct Lake on SQL shows and uses views in DirectQuery mode.

During the initial public preview, Direct Lake on OneLake doesn’t support the use of shortcut tables in the semantic model, or using any table in a Lakehouse opted into the public preview of ‘Manage OneLake data access (preview)’. Accessing any table, including shortcut tables, is supported if you sign up for early access of the upcoming OneLake security.

Migrating an existing Direct Lake on SQL semantic model to Direct Lake on OneLake is possible in Power BI Desktop now TMDL view is available in live edit.

Create a test Direct Lake on OneLake semantic model using the same data source in Power BI Desktop. Remember views should be materialized and for shortcut tables, they are not yet supported unless in early access of OneLake security. Go to the OneLake catalog > Pick the Lakehouse or Warehouse > Connect.
Navigate to TDML view, and from the Data pane > Model explorer, drag the Semantic model node to the script window to script the entire model.
Scroll to the bottom to find the expression, copy the code starting with ‘let’.
Open a new instance of Power BI Desktop by going to File > Blank report. Then, live edit your existing Direct Lake on SQL semantic model. Go to OneLake catalog > Power BI semantic models. Pick the model, then on the drop-down on Connect choose Edit.
Navigate to TDML view, and from the Data pane > Model explorer, drag the Semantic model node to the script window to script the entire model.
To give yourself a way to undo the migration, you have two options. The first options are you can create two TMDL scripts of the semantic model so you can apply the un-altered one to return to Direct Lake on SQL. The second option is you can navigate to drop-down below the name to click Version history and create a version to return to.
To continue, scroll down to the bottom of the script to find the expression in the Direct Lake on SQL model.
Paste in the one you copied from the test Direct Lake on OneLake model. Don’t hit apply just yet!
If this is a Lakehouse without schemas or folders, there is one additional step. If you are using a Warehouse or Lakehouse with schemas, you do not need to do this step. Click Replace in the ribbon and look for ‘schemaName: dbo’, changing ‘dbo’ with what your schema happens to be. Keep the Replace box empty to remove all these references.
Now click apply.
If you have calculated tables or calculation groups, you may need to go to Model view and click refresh. You can test out the semantic model by going to DAX query view and running any query. Quick queries are available in the right-click menu of any table, column, or measure in the Data pane to generate a DAX query for you.

To recap, you can create semantic models using Direct Lake on OneLake storage mode in Power BI Desktop from one or more Fabric artifacts. At this time, only Lakehouses and Warehouses are available, but other artifacts will be added during the public preview.

To create the semantic model with Direct Lake tables, follow these steps.

Open Power BI Desktop and turn on the public preview for Create semantic models in Direct Lake storage mode from one or more Fabric artifacts. It is recommended to also turn on Live edit of Power BI semantic models in Direct Lake mode too, if it is not already turned on to edit the model you create later in Power BI Desktop.
Go to OneLake catalog in the ribbon.
Pick a Lakehouse or Warehouse with the tables you want to add and click Connect.
Give your semantic model a name and pick the tables you want to use then click OK.

Now the semantic model in Direct Lake mode is created in the service and you are live editing the model in Power BI Desktop.

To add tables from other OneLake Fabric artifacts, such as Lakehouses or Warehouses, follow these steps.

Go to OneLake catalog again in the ribbon.
Pick another Lakehouse or Warehouse with the tables you want to add and click Connect.
Pick the tables you want to use then click OK.

That’s it! Now you can continue to build your semantic model or add more tables from other Lakehouses or Warehouses.

In addition, you can use the Power BI Project by going to File > Export > Power BI Project.

To create a report from this newly created semantic model there are many paths, but here is how you can do it in Power BI Desktop to get you started.

In Power BI Desktop go to File then select Blank report. This will open a second instance of Power BI Desktop on your machine. If you have multiple monitors, you can then build your semantic model on one screen and build your report on the other screen.
Go to OneLake catalog in the ribbon.
Pick the semantic model you just created and click Connect. And now you can create fully featured Power BI reports just like you can with any published semantic model in Power BI Desktop.
When you are ready you can click the Publish button in the Home ribbon to publish it.

As with any published Power BI semantic model, you can create reports, explorations, DAX queries, and paginated reports in the service, as well as connect to the model via Excel.

Microsoft Fabric Copilot in Power BI may be available to help you create reports in Power BI Desktop or the web.

For more information and any limitations about Direct Lake on OneLake during public preview see the documentation at aka.ms/DirectLake and these resources may also be helpful.

Try it out today and let us know your feedback by commenting below!

Blog consacré à Microsoft Power BI