Work · local LLM · In use by teammates
AI SQL Assistant
A Chrome side panel that catches the SQL mistakes an analytics editor punishes, and writes queries with a local model.
TypeScriptChrome ExtensionLocal LLM
context
Built at work for analytics users
The problem
The analytics editor rejected common SQL patterns with unclear errors, and nobody remembered the table names.
What I built
- Eleven lint rules for the patterns that editor and its warehouse reject.
- A searchable, pinnable schema browser for every table and column.
- A chat assistant that writes queries grounded in a rules knowledge base and only ever produces SELECT statements.
- Local by default: data goes only to the analytics API the page already uses and the local model.
How it works
Every query is checked against the rules as you type.
Impact
- Fewer failed runs and less time hunting for table names.
- Keeps data on the machine by default.
What I learned
LLM-written SQL needs hard guardrails. A short list of enforced rules mattered more than a bigger model.