Best practices for semantic views¶

Best practices for semantic views are organized into two guides, plus a set of accuracy-tuning topics for Cortex Agents. Use this page to find the right one.

Modeling¶

Design semantic views that produce accurate, trustworthy answers in Cortex Agents: scoping and organization, high-quality descriptions, relationships, metrics and filters, and how to increase accuracy and iterate. See Best practices for modeling semantic views.

Developing and deploying¶

Manage semantic views as part of a data engineering pipeline or data product: ownership and data access (RBAC, masking, and row access policies), creation and update options, automated (CI/CD) deployment, dbt, and BI integrations. See Best practices for developing and deploying semantic views.

Increasing accuracy¶

Detailed guidance for the features that most improve Cortex Agents answer quality:

• Add verified (“gold”) queries — see the verified query repository.
• Optimize a model from its verified queries — see Optimize an existing semantic view or model with verified queries.
• Steer SQL generation and question handling — see custom instructions.
• Improve literal matching with Cortex Search — see the literal search guide.