I have seen many applications augmenting LLM/GPT responses with unstructured text data using frameworks like LangChain/Vector databases. The idea is basically querying similar document chunks and plug them into the prompts.

But for structured data stored in SQL-like databases, there doesn't seem to be a widly adopted way.

I have seen exapmles from the LangChain library using GPT to generate a SQL query and execute it using Python, but this would create security issues if the LLM generates queries that tamper with the database.

Thanks for any inspirations or input.