Text-to-SQL tools often struggle with enterprise security requirements, but Ashish Sinha is addressing this head-on with his library, schemagate. After running the tool against a real Oracle database, Sinha implemented critical features to satisfy developer demands for better access control and transparency.
Community-Driven Security Features
The update comes directly from user feedback on DEV.to. Two commenters specifically requested column-level restrictions and an audit record of selected tables. Sinha responded by adding the restrict_column() function and updating the Selection.to_ method to include these capabilities, moving beyond simple table filtering.
Real-World Oracle Testing
Testing against a live Oracle database provided the necessary stress test for these new features. While the source text is heavily compressed, the core narrative highlights the transition from a basic table-picking library to one that respects granular data permissions. This shift is essential for teams integrating LLMs with sensitive enterprise data.
Key Takeaways
- Schemagate now supports column-level restrictions via restrict_column().
- Audit logging for table selection has been added to the Selection.to_ output.
- The updates were driven by direct community feedback on DEV.to.
- The library was tested against a real Oracle database, validating its enterprise readiness.
The Bottom Line
Security isn't an afterthought in enterprise AI. Sinhaβs quick implementation of column-level controls shows that the dev community is demanding more than just syntactic correctness from text-to-SQL tools.