Speak instead of typing.

Tap the bubble, talk, and clean text appears where you were typing, in any app. It uses the speech and AI models you choose, including your own.

Free and open source under the GPLv3. About 4 MB. Android 9 or later.

Why Sasayaki

It is just the app. You pick the servers, and nothing else gets your voice.

Works in every app: The bubble shows up when your keyboard opens, in chats, mail, notes and browsers. You can also start it from your keyboard's mic key.
Cleans up what you say: Filler words go, and “8, no wait, 9” becomes “9”. Each app can get its own style, casual for chats and full sentences for mail.
Use any OpenAI-compatible server: Run whisper.cpp, faster-whisper or llama.cpp at home, or use a hosted service. Cleanup is optional.
Private: No cloud backup, and audio only goes to the servers you set up. It won't record in password or banking fields.
Gets names right: Add names and jargon once, and both the speech and the cleanup model spell them your way.
Small, with no bloat: About 4 MB. No ads, no trackers, no account and nothing running except the bubble.

Install with Obtainium for automatic updates, or download the APK. You also need a speech-to-text server. The setup guide lists some options.