It is for both and yes what you are talking about is possible.
Thanks for the speedy reply! That sounds brilliant. Are there any specific, self-hosted LLMs you'd recommend for such a use case?
I like Mixtral 8b7x something with 32k context limit.