This is basically what you’re asking for. The tradeoff is that we add a dependency to Google, and requires storing translation models offline on disk and memory which can take a good amount of space. This is also what Amethyst does on Android. I’m happy to explore this translation approach as an option for Damus.