Maha Transcribe v3, as a native Android app. Shares the Talk to Type provider core as a submodule.
# Maha Transcribe for Android Maha Transcribe v3, as an app you launch from the home screen instead of a server you start from Termux. Same interface, same design rules, same promise: **the audio is committed to storage before transcription is ever attempted**, so pulling the wifi or killing the process loses nothing. Transcription is AssemblyAI. The text features are Anthropic, Google Gemini and Groq. Keys come from a file picker, never a text box, and never leave the process in anything but their masked form. ## The shared core is not copied The provider layer lives in `markoboskoauroville/DictateKeyboard` at `lib/dictate-core`, and is consumed here as a git submodule at `ttt/`, compiled in place. Two copies with a rule for keeping them in step are still two copies, and a rule that has to be remembered is one that eventually is not. ``` git clone --recurse-submodules https://github.com/markoboskoauroville/MAHA_TRANSCRIBE_ANDROID.git ``` Read `HANDOFF.md` before changing anything. It describes the app that exists.