A semantic layer is a shared, versioned description of selected data: catalogs, schemas, tables, columns, business context, and generated metadata or embeddings. Semantic layers are top-level resources, so several projects can use the same one.
A project needs a built semantic-layer version before anyone can use chat, workflows or metrics in it.
Create a semantic layer
- Open Semantic Layers and choose Create semantic layer.
- Enter a durable name. You can’t change it after creation.
- Select the active cluster when a cluster choice is shown.
- Browse the catalog and schema. Choose at least one table and the columns AgentFlow should understand.
- Add table and business context, including domain language, measures, known relationships and important rules.
- Optionally configure an available enricher, such as OpenMetadata.
- Create the layer. AgentFlow starts its first build automatically.
- Wait until the version is Built. If a build fails, inspect and resolve it before continuing.
Versions
Each build creates a new version. Projects adopt a version deliberately, by pinning it, so a rebuild never silently changes a project’s data context. See Project settings.
Use external Cube instead
A project can query through the built-in semantic layer, or through an external Cube source when one is configured. See Connect data sources and systems. Last modified on October 1, 2026
