Polished Recognition is a <b>voice keyboard</b> powered by your own AI. Switch to it, speak, and polished text lands in whatever app you are typing in — transcribed by the speech provider <b>you</b> choose, refined by the language model <b>you</b> choose, with a prompt <b>you</b> write.<br /><br />Switch to the Polished keyboard and recording starts instantly. Pause and resume mid-sentence, switch to your typing keyboard and back without losing dictation, then tap send: audio goes to your STT provider, the LLM polishes the text — fixing filler words and punctuation, restructuring, or translating to another language — and flawless text is typed into the field. <b>Raw mode</b> skips the LLM entirely and inserts the plain transcription; <b>Polish only</b> refines without translating.<br /><br /><b>Bring your own provider.</b> 18 presets (OpenAI, GROQ, OpenRouter, Google AI, DeepSeek, xAI, Mistral, local Ollama/LM Studio, …) or any custom OpenAI-compatible endpoint. GROQ's free tier covers a complete STT-plus-polish pipeline at zero cost. The system prompt is fully editable, with variables for source and target language; searchable model pickers with per-provider caching handle hundreds of models.<br /><br /><b>Gboard will never use this.</b> Gboard's microphone is hardcoded to Google's voice typing and cannot be redirected to any third-party voice input. Use a keyboard that delegates its mic: the AOSP keyboard (LineageOS and derivatives), HeliBoard (F-Droid only), Fossify Keyboard, or OpenBoard.<br /><br /><b>Privacy-first:</b> No analytics, no tracking, no proprietary SDKs, no account, no central server. All data flows directly between your device and your chosen endpoints — or nowhere at all, if you run Ollama locally.<br /><br />Permissions:<br />• RECORD_AUDIO — required to capture microphone input<br />• INTERNET — required to send audio to your configured STT/LLM endpoints