Skip to main content

dbt models & semantic layer

When a repository syncs, Zingle AI builds a rich catalog from your dbt project. This catalog is what makes the agents context-aware - they don't guess at your schema, they read it.

Models & columns

For every dbt model, Zingle AI stores:

  • Its name, path, materialization, and description.
  • Its columns, with dual provenance - a column can be discovered in the dbt manifest (declared in YAML), in the warehouse (present in the physical table), or both. This lets the UI show you where a column is defined versus where it actually exists.
  • Its lineage - upstream and downstream dependencies, rendered as an interactive graph in the Data Explorer.

Semantic layer

If your project uses dbt's semantic layer, Zingle AI syncs the full definition:

ObjectDescription
Semantic modelA semantic wrapper over a dbt model.
EntityJoin keys / grains (primary, foreign, unique).
DimensionAttributes you slice by (categorical, time).
MeasureAggregatable quantities.
MetricNamed, reusable calculations built from measures.

Zingle's agents use the semantic layer to reason in business terms (metrics and dimensions) rather than requiring only raw table structures.

How agents use the catalog

  • The Build agent looks up existing models to avoid duplication and to write changes consistent with your project's conventions.
  • Refactor and Optimize analyze the whole model graph to find merge candidates, duplicates, layering violations, and optimization opportunities.

Explore all of this visually in the Data Explorer.