I also tried their local AI for tagging and grouping media. It worked to some extent with photos, though it's very basic experience.
I hope that at a later stage users will be able to extend the local models. No idea if it's on their roadmap. Fingers crossed.