Routing Mode for Cortex Analyst

Routing Mode is a query strategy that prioritizes semantic SQL and falls back to standard SQL only when needed. It acts as a simpler version of SQL, with guardrails coming from your semantic views. Routing mode uses your semantic views to drive higher accuracy and consistency. As a result, metrics, joins, and filters follow governed definitions from the semantic view.

You can use Routing Mode with no change to your workflow, except higher text-to-sql quality.

Note

Routing Mode does not change permissions. Semantic views are Snowflake objects with standard privileges; access is enforced the same way as tables or views.

Benefits of Routing Mode

Routing Mode offers the following benefits:

  • Consistent metrics: Queries use definitions from semantic views, not SQL.

  • Safer defaults: Dimensions, metrics, and joins come from governed metadata.

  • LLM-friendly: Shorter SQL is easier for an LLM to produce correctly.

Routing Mode could be beneficial in the following situations:

  • You have one or more semantic views that define core business entities and metrics.

  • You want consistent answers for common questions, with flexibility for edge cases.

For example, consider the following scenarios and how Routing Mode handles them:

  • Ask for a governed metric by a business dimension

    • User intent: “Average order value by customer segment.”

    • Routing behavior: Tries semantic SQL first, so joins and metric calculations come from the view.

      SELECT *
FROM SEMANTIC_VIEW(
  tpch_analysis
  DIMENSIONS customer.customer_market_segment
  METRICS orders.order_average_value
)
ORDER BY customer_market_segment;
      Copy

    • Benefit: No manual joins or metric formulas. Results align with your BI definitions.

  • Multiple governed metrics with one dimension

    • User intent: “Show total revenue and order count by year.”

      SELECT *
FROM SEMANTIC_VIEW(
  tpch_analysis
  DIMENSIONS orders.order_year
  METRICS orders.total_revenue, orders.order_count
)
ORDER BY order_year;
      Copy

    • Benefit: Both metrics use the same definitions and filters as in the semantic view.

  • Fallback for uncovered asks

    • User intent: “Show a raw column or transformation not modeled in the view.”

    • Routing behavior: If the semantic view cannot satisfy the request, Cortex Analyst automatically routes to standard SQL on base tables.

    • Benefit: Flexibility without blocking the user.

How it works

The following procedure outlines the steps that Cortex Analyst takes when using Routing Mode.

  1. Cortex Analyst uses Routing mode in the playground, API, and all product surfaces.

  2. Cortex Analyst tries to produce semantic SQL.

    SELECT … FROM SEMANTIC_VIEW(...).
    Copy

  3. If the produced SQL fails validation, Cortex Analyst routes to standard SQL on physical tables.

Note

Routing Mode only results in semantic SQL for about 10% of queries.

Considerations

  • Routing Mode doesn’t hide missing definitions. If the semantic view cannot satisfy a question, Cortex Analyst falls back to Standard SQL. You should expand the semantic view to reduce fallbacks over time.

  • You should create a semantic view using Snowsight or CREATE SEMANTIC VIEW, then centralize core metrics and joins before using Routing Mode.

Language: English