Hey Jérôme! Nice to hear from you again!
I created GrampsChat to be able to use (and easily switch) between language models. You can use ones that cost money, and ones that are free. Unfortunately, the models like pixtral-12b-2409 are not supported by the open source stack. But of course, as you have shown, you can easily wrap the request up in Python and make an API. Luckily, there are some free versions of such models. However, the hardware requirements probably exclude most of the users of Gramps.
In any event, if a user was willing to pay Mistral, you could probably create a nice Gramplet for the Media view.