Speech → draft → refined text
Turn a rough voice message into something clean, concise, and ready to send.
The app captures a spoken draft with the browser speech API, then asks WebLLM to rewrite it in your chosen tone. Use the lite model for quick cleanup or the deep model for stronger edits.
Fast passLightweight rewrite for quick notes, reminders, and short replies.
Deep passHeavier model for emails, summaries, and more nuanced restructuring.
Capture
Best in Chromium browsers with microphone permission and WebGPU enabled.
Mic idleModel not loadedChecking WebGPU
No model loading yet.
Refined output
Prompt preview
Notes
- Speech recognition quality depends on the browser engine and language pack.
- The refinement step runs fully in-browser through WebLLM once the model is downloaded.
- Deep models can take several gigabytes of GPU memory; keep the lite option as the default path.