Changelog

What's new, what's improved, what's fixed.

Minor July 21, 2026
v1.1.0
Smarter dictation, faster control

Added

  • Smart local formatting with Qwen3 1.7B and llama.cpp for punctuation, paragraphs, lists, and filler removal
  • Content-safety validation for formatted text: if the local model drops or changes too much, Ditto keeps the original transcription
  • Hybrid Ctrl + Windows shortcut: hold for push-to-talk, double-press for hands-free, and press again to finish
  • Silero VAD voice detection with frequency-aware bars and waves that react only when speech is detected
  • Status sounds for recording start, hands-free lock, stop, cancel, and completed transcription
  • In-app download and removal of the optional Qwen formatter model
  • Local transcription history in Settings with up to 100 recent entries, one-click copy, individual delete, clear all, and an on/off setting

Improved

  • Redesigned floating pill with a slimmer idle state, instant shortcut tooltip, click-to-record, dedicated cancel button, and compact error feedback
  • Whisper model management now supports deleting user-downloaded models while protecting the last available transcription model
  • Model storage now reports both Whisper and formatter downloads and opens the shared Ditto data folder
  • Existing users on the original default shortcut are migrated to hybrid mode while custom shortcut choices are preserved
  • Voice visualisation now uses local VAD, RMS intensity, and frequency bands without trimming or changing the audio sent to Whisper

Fixed

  • Stop and cancel requests are preserved while microphone access is still starting, so quick shortcut or pill actions no longer get lost
  • Whisper silence markers such as [BLANK_AUDIO] and [AUDIO_EN_BLANCO] are no longer copied, pasted, formatted, or counted in usage stats
  • Microphone and transcription failures now return the pill to idle automatically instead of leaving the recording flow stuck
Major May 10, 2026
v1.0.0
First public release

Added

  • Local voice-to-text transcription with whisper.cpp, no audio leaves the machine
  • Floating pill with click-through, drag, and always-on-top
  • Tray app lifecycle: close hides the pill, quit only from the tray menu
  • Global shortcut (Ctrl+Shift+Space by default, fully configurable)
  • Auto-paste to the active window with optional clipboard restoration
  • Settings window with seven panels: General, Shortcut, Audio, Appearance, Models, Advanced, About
  • Five pill themes (Matte, Onyx, Aurora, Sunset, Mint) with voice-reactive bars on the colored ones
  • Eight UI languages with system auto-detect (EN, ES, FR, DE, IT, PT, JA, ZH)
  • On-demand model download with progress and cancel: Tiny, Base, Small, Medium, Large-v3
  • GPU acceleration via CUDA for NVIDIA cards (CPU fallback works without one)
  • Welcome window on first launch to pick and download a Whisper model
  • Microphone selector, input gain, and Chromium noise suppression toggle
  • Pill position grid (corners and center top/bottom), per-display preference, remembers last drag position
  • Auto-recovery: if the active model is deleted while the app is running, the welcome reopens to redownload