SmartQL turns natural language into SQL, database first. Ask for the users who spent over a thousand and it writes the query against your actual schema.

A YAML semantic layer describes what the tables and columns mean, so the model works from a description of your data rather than guessing from column names. Every generated query is parsed and validated with sqlglot against a real AST before it runs, which catches invented columns and malformed SQL rather than trusting the model.

Model-agnostic through LiteLLM, so it runs against whichever provider you prefer. Available as a Python library, a CLI shell, and an optional REST API. Installable from PyPI.