Skip to main content

Wren SQL

Legible uses Wren SQL, a SQL dialect based on ANSI SQL and optimized for the semantic engine. When you ask questions or write queries, Wren SQL is used internally by the AI to generate results.

How It Works​

  1. You ask a question in natural language
  2. The AI generates a Wren SQL query against your data models
  3. The semantic engine translates it to native SQL for your data source (PostgreSQL, BigQuery, etc.)
  4. Results are returned to the UI

Key Differences from Standard SQL​

Wren SQL operates on your data models rather than raw database tables:

  • Table names refer to model names you defined, not database table names
  • Column names include calculated fields and relationships
  • Joins are inferred from defined relationships

Viewing Full SQL​

When you receive a query result, you can inspect the SQL in two ways:

Wren SQL View​

The default view shows the Wren SQL query — the semantic query against your models. This is useful for understanding the AI's intent.

Original SQL View​

Click "Show original SQL" to see the native SQL that was actually executed against your database. This is the translated query in your data source's dialect (e.g., PostgreSQL syntax, BigQuery syntax).

tip

Use the original SQL view when you want to:

  • Debug query performance
  • Copy the query to run directly in your database client
  • Verify the exact SQL being executed

Writing Custom SQL​

When creating Question-SQL pairs or adjusting AI-generated SQL, write in Wren SQL format:

-- Reference model names, not table names
SELECT customer_name, total_orders
FROM customers
WHERE region = 'US'
ORDER BY total_orders DESC
LIMIT 10

The semantic engine handles translating this to the appropriate native SQL for your connected data source.