For many years I was feeling my skills and my time were wasted on one-off queries.
Requests were coming from everywhere, slack, teams, tickets.
At some point I decided that it would be better to build the tool
that will save me time. So after studying the last months about AI in
general and text to sql concepts more specifically I decided to build Savvina -
my daughter's name.
Privacy was my first concern building that so I worked towards that with multiple
security guardrails. Self-hosted was another security aspect and as one can see
at https://www.youtube.com/watch?v=zK-n0cM88vI it takes less than 15 minutes to
run your first question.
Community edition can be found at https://github.com/savvina-ai/savvina
and supports PostgreSQL and MySQL.
The biggest thing I learned from this journey is that semantic layer
sometimes matters more than the LLM you choose. Or that parse-time query validation is more reliable than confidence scores.
If you're tired of being the SQL middleman — or just want fast insights from your data try it on real schemas and tell me where it breaks.
File issues, start discussions.