Allowing for private data/information to become less secure.

[DJJones66]

Your 3rd motivation raised valid concerns with regard to private data, I feel you should continue this mindset through out the project then and use a local LLM and VectorDB this way you have zero risk of your data being out in the wild.

[mkwatson]

I agree having a local LLM could be a useful option. As it exists now, even using an external LLM and making a network request, none of your private data should be made public because the network isn’t training on it, only using it for inference.