Vanna

Open-source framework for text-to-SQL

MIT-licensed Python framework that trains a RAG model on your schema, docs, and past queries, then answers plain-language questions with SQL, results, and charts. Works with most databases and local LLMs via Ollama.

Visit Vanna → Data & Analytics

What it does

Vanna is an open-source text-to-SQL framework: you train a retrieval layer on your schema DDL, documentation, and known-good queries, and it generates SQL for plain-language questions, executes it against your database, and can chart the results with Plotly.

Privacy is structural — database contents are not sent to the LLM by default; the retrieval layer sees schemas, docs, and queries. It works with hosted or fully local models via Ollama, and connects to most major databases.

MIT-licensed and community-maintained, it earned Google Cloud's BigQuery-ready designation after technical validation. It is a building block for internal data tools rather than a polished end-user app.

AI stack

RAG over your schema and query history, feeding any LLM — hosted or local via Ollama.

From the AXIS Launch team

The train-on-your-own-queries loop is the honest answer to why generic text-to-SQL fails on real schemas, and keeping row data away from the LLM by default is the right architecture. Question for the maintainers: what accuracy do teams see on legacy schemas with poor naming, before and after training?

Similar apps