Not sure what you mean by “train.” If you mean “test” then yes, they are in alpha and anyone can submit bugs. As other users have said, we have only their word that they intend to keep the source available. If you mean “track users to train ML models,” then no. The whole point is that the software is private. All of the processing - the gesture typing, the audio processing, the LLM, etc are all performed on-device. And the source is visible for all to see, so it offers similar protections to FOSS software in this regard.

The software doesn’t “phone home” - it can’t even check itself for updates. It just sends you a little message on a predetermined schedule to manually check. (or you can use a repo/software manager)

In theory and in practice, any Open Source project could be purchased by a for profit company who takes down the source code. However, any prior code would remain under the previous open license. Apparently one of the issues with this license is that it contains no durable license for the code itself. You can’t just fork it and make your own version, although you can use any of the code with certain noncommercial and attribution requirements.

I doubt it. All of the AI stuff is local, and I don’t think there’s any network interaction other than checking for updates (you could check that, or block it with something like RethinkDNS or NetGuard), so there wouldn’t be any way for them to get your data. The other app of theirs I use, Grayjay, doesn’t even sync between instances of their app across devices, so I think it’s fair to say they explicitly avoid touching the network whatsoever.