Skip to main content
Version: 0.0.41

Data Sources

The Data Sources tab is where you connect structured data sources to a Custom Model. The connected data sources form the foundation of the model's semantic layer — they determine what data the model can query and reason about.

Prerequisite: Semantic Extraction Must Be Complete​

Before connecting a data source to a Custom Model, the data source must have completed semantic extraction. Semantic extraction reads the schema and generates the descriptions, hierarchies, and synonyms the AI uses to understand the data.

The status of semantic extraction is visible in the data source editor (Operations tab). Only data sources with a successfully completed semantic layer are ready to be connected to a Custom Model.

important

Connecting a data source that has not yet been semantically extracted will result in an empty or incomplete semantic model. Always run semantic extraction at the data source level first.

The recommended preparation sequence for each data source before connecting it to a Custom Model:

  1. Create the data source and configure its connection
  2. Set schema, table, and column filters in the data source Filters tab
  3. Run Semantic Extraction from the data source Operations tab
  4. Review and enrich descriptions in the data source Descriptions tab
  5. Only then connect it to a Custom Model

Connecting Data Sources​

In the Data Sources tab, use the selector to add one or more data sources to this Custom Model. Only data sources accessible to your role are shown.

After selecting and saving, the connected data sources become the input for the model's semantic layer.

Using a Timeline Data Source​

Best Practice: Connect a Timeline Data Source

For any Custom Model intended for time-based analysis, connect a Timeline data source in addition to your primary data sources. A Timeline data source provides a structured time dimension that enables accurate date-based grouping, filtering, and trend analysis.

Without a Timeline, the AI must rely on date columns in the transactional data itself, which may produce inconsistent or inaccurate temporal queries — especially across fiscal years, custom periods, or sparse date ranges.

The Timeline data source type is specifically designed for this purpose. See Timeline for configuration details.

Connecting Multiple Data Sources​

A Custom Model can connect multiple data sources. This is useful when the analytical domain spans several systems (e.g., sales data from a database + uploaded CSV files for targets).

When multiple data sources are connected, the semantic layer is built across all of them. The AI can reference tables from any connected source in a single query.

Recommendations:

  • Connect only data sources that are relevant to the model's intended use case. Unrelated data sources reduce query precision.
  • Ensure all connected data sources have completed semantic extraction before building the semantic model.
  • After adding or removing data sources, run Update Semantic Model from the Operations tab.

After Connecting Data Sources​

Once data sources are connected:

  1. Go to the Filters tab to define which part of each data source is exposed to this model's users.
  2. Go to the Operations tab to build the semantic model.