[Feature Request] Text to SQL with AI/LLM based chat

[pietz]

Describe the solution you'd like

I think it would be great if Dataflare had an AI/chat based interface that could write SQL queries for me like TablePlus does. I would provide my API key, and it could view the schema and generate SQL based on a description.

I could write things like "list all users that signed up in the past 2 months" and it would create the query based on the schema. Of course this gets more useful as queries get more complex.

[wyhaya]

Hi @pietz , thank you for your feedback. It's already on our roadmap.

[shajin-sha]

Huge +1 on this!

@wyhaya , do you mind if I take a crack at this feature? I’ve been a big fan of DataFlare for a while and the UX is honestly the standout for me. I've even got my whole team using it now.

I know you guys might be closed source, but if you are open to outside contributions, I’d love to help out. Keep up the great work!

(Btw, I’m working on designlumo.com if you ever want to check it out.)

[wyhaya]

It's not possible at the moment, but we'll soon begin preparing to develop AI-related features, starting with an AI Chat—do you have any expectations or suggestions for that?

[shajin-sha]

Yah one suggestion will be : an inline chat rather than an chat side by side interface

One reference will be cursor's inline editor

Also make sure it will have all the required contexts like table, columns, index etc.

[pietz]

It's not possible at the moment, but we'll soon begin preparing to develop AI-related features, starting with an AI Chat—do you have any expectations or suggestions for that?

I think the easiest approach would be to have an "AI Optimize" button somewhere in the SQL query window. It could take the current content of the input + a static view of the schema and replace the input with what it generated. This could be used to fix broken sql or even turn natural text into sql.

The more sophisticated way would be to have a separate chat panel and the context to the agent is provided using tools. The llm could have access to a get_schema() tool or write_sql(query) tool, while normal response outputs continue to appear in the chat screen.

As for providers, I'd focus on the openai sdk with the additional option of modifying the base URL on top of the API key. Most providers have an openai compatible API, so this single setup can get you a long way.