Writing named SQL queries for a data warehouse

Adding queries

Click Queries on a warehouse row to open the query editor panel. Each query has a Name (e.g. "Customer Orders") and a SQL statement. Click Add Query to add more, then Save Queries.

Placeholder variables

SQL queries can reference the current conversation's customer data using curly-brace placeholders. These are safely substituted (SQL-escaped) before execution:

• {customer_id} — the customer ID
• {email} — the customer's email address
• {name} — the customer's display name
• {session_id} — the current conversation ID

Example:

SELECT * FROM orders WHERE customer_id = '{customer_id}'

Safety constraints

• Only SELECT statements are permitted. Any attempt to save a query containing DML or DDL (INSERT, UPDATE, DELETE, DROP, etc.) is rejected.
• A LIMIT 100 clause is automatically appended if not already present, preventing accidental full-table scans.

How results appear in conversations

On the first message of a conversation, all active warehouse queries are executed in parallel. Each query's results are injected as a separate section:

## Warehouse Data: {warehouse name} — {query name}

Results are cached on the conversation, so subsequent messages reuse the same data. Queries that fail are silently skipped and never block the AI response.