Guidance on how to implement a text-2-sql style agent

[stokedout]

I would like to implement a text-2-sql style agent using the teams-ai library.

I want to allow users to ask questions to specific data sources, in this case a database. The difference being that the user input needs converting to SQL before seemingly calling an action that gets the data from the data source. I can see the samples talk about data sources. If I use the built in DataSource class then it will add the output to the prompt. Therefore I'd need to:

1. Ask a question
2. Data source takes question temp.input then somehow calls another prompt to get the SQL?
3. Trigger an action to get the data and render it

The above doesn't seem like a simple way forward. I'm thinking this can be achieved using a mix of LLM calls/prompts and function calls but how that maps to code in the library I'm unsure. Can you help?

I'm a bit lost with the library in general:

• Understanding augmentation types is confusing and not very well defined
• Understanding what the default prompt is meant to do is confusing
• Understanding what happens if I add multiple prompts in separate folders is not clear how it will be picked up by the planner?

[heyitsaamir]

We're clearing discussions and starting fresh. This discussion has not been acted upon. Please reopen the discussion, or start a new one if this question is still important. Thank you for your understanding.