That's kinda cute, yeah-- love how they struggle to explain their besoke "AI Tokens" system and incompatability with existing local models. That would have made sense when they started development (and generally for accessability), but I wonder what network calls it makes, and how well it would work if we MITM-proxied those to eg. Mixtral.
It was a mix of a time constraints we gave ourselves to finish and ship the product by mid December, keeping costs lower for us and players, while also maintaining a level of product quality that we were comfortable with. We didn't necessarily do extensive research into the options, but from what we did find we didn't believe we'd be able to have unique voices that we'd be satisfied with across 31 current characters without also having to pay top shelf prices for them or do lots of manual tweaking ourselves.
Given the silly cartoony nature of the game, the current implementation is working fine for now. We are of course always looking at options when they present themselves, and as costs go down with the advancement of tech, it's more possible we'll see our characters speaking as well someday in the near future.
That's our answer from a design and product standpoint at least. Hope it's helpful!
[0]: https://www.playsuckup.com/