Building a mental-health companion that never phones home — thoughts?

Hey PH, I'm launching NilaMind on July 16, and wanted to start a conversation before then.

The core bet behind it: a mental-health companion app should never need your data to leave your phone. No account, no backend, no analytics - the AI model, the crisis-safety checks, even voice transcription all run on-device.

I'm curious what this community thinks: does "fully on-device" actually matter to you when it comes to AI apps that touch something personal, or is convenience/cloud-power usually going to win out? And if you've used other on-device AI apps, what made them feel trustworthy (or not)?

Happy to answer anything about the architecture too - happy to go deep on the crisis-detection design, the model choice, or why I skipped a backend entirely.