Skip to main content
Version: Next

Descriptions

The Descriptions tab contains the metadata generated for the data sources connected to this Custom Model. Descriptions provide context about tables, columns, and data types, helping the model deliver accurate responses to user queries.

How Descriptions Are Generated​

Descriptions are generated automatically during Create Semantic Model or Update Semantic Model operations. The system analyzes the data source structure and creates descriptive metadata for each element.

Managing Descriptions​

System-generated descriptions are sufficient for most use cases. Manual management is an advanced option.

Adding Descriptions​

You can add additional descriptions for tables or columns that the system did not describe, or where the generated description is insufficient.

Modifying Descriptions​

Existing descriptions can be modified or disabled. Only change a description if the system-generated version is incorrect or misleading for the model's use case.

Declaring Measure Semantics​

For a numeric column, you can declare its role (Measure, Dimension, Identifier, Temporal, Status), default aggregation (Sum, Average, Count, Count Distinct, Min, Max, None), unit, display name, and synonyms, instead of relying on the AI to infer them. This works the same way on a Custom Model's star views as it does on a data source — see Declaring Measure Semantics for the full field reference. Since the Custom Model is the layer you and your users interact with directly, this is the primary place to curate measure semantics for star views built from multiple data sources.

Columns you curate this way show a Curated badge with their aggregation in the collapsed column list, and carry a provenance label (Manual, LLM, or Annotator).

Derived Measures​

For metrics that combine several columns — like gross margin from revenue and cost — add a derived measure below the column list: a name, an aggregated SQL expression over the table's columns, a unit, synonyms, and a description. See Derived Measures for the field reference. Save persists the whole list, so make all your changes before saving.

Best Practices​

  • Trust system-generated descriptions as the default — they are optimized for the model's analytical functions.
  • Only modify descriptions if you have a clear understanding of how the change will affect query accuracy.
  • After making manual changes, test the model's responses to verify that the adjustments improve, not degrade, the output quality.