Hey Angelina! I'm actually developing a text2sql app (no RAG in it yet) and after comparing both, seemed to me like DataLine had better quality results for most queries, even with additional context in Vanna.

I do think RAG is helpful, in some cases necessary to add a 'knowledge base' for bigger DBs, but one can get pretty far with just Text2SQL and a proper state machine loop! The issue with most Text2SQL applications is that they just do a plain LLM query with all the metadata and expect things to work well. In reality an approach closer to 'agents' is much better as it can error correct and use tools and such.

DataLine is also open source and privacy first! Check it out, would love to hear what you think: https://dataline.app

Did a basic comparison with Vanna here: https://ramiawar.medium.com/vanna-ai-vs-dataline-4829b1d2fad5