Calliop v0.1.4
Calliop v0.1.4
Feature release — live editing during dictation, update notifications, richer dictation controls, and model management in Settings.
Changes
Live dictation
- Live editing pipeline — punctuation on pause, list formatting, and pipelined LLM cleanup while you dictate
- Improved list formatting in transcript processing
- Stronger LLM cleanup validation so prompt fixes apply without rebuilding the sidecar
Dictation controls
- Clearer toggle vs push-to-talk behavior with busy hints during transcription and injection
- Cancel during transcription by releasing the shortcut (no text injected)
- Updated user guide with a shortcut action table (EN)
Updates
- Update notifications and background download with in-app toasts
- Streamlined update check logic
Settings & models
- Delete and reinstall models from Settings (Whisper and LLM)
- Better handling and messaging for corrupted model files (EN/FR)
- New model management actions and status feedback in the Models panel
Dictionary & injection
- Dictionary learning ignores appended tokens and case-only changes to reduce pollution
- More reliable paste injection error handling on Windows
Hotkeys & LLM tooling
- Hotkey capture includes modifier keys for more accurate shortcut mapping
test-llmsupports raw input and model selection for easier debugging
Installation (Windows)
- Download
Calliop_*_x64-setup.exe(NSIS) or.msi. - Install WebView2 if needed.
- Accept the SmartScreen warning (unsigned binary in v1).
- On first launch, the Whisper model downloads (~466 MB).
See docs/installation.md.
macOS / Linux (experimental)
DMG, AppImage, and deb artifacts are provided unsigned and untested for v1. Use at your own risk.
Known limitations
- Windows binaries are unsigned (SmartScreen)
- ML models download separately on first use
- Automatic updates are enabled by default; disable in Settings → Advanced if needed
- No official macOS/Linux support in v1
License
AGPL-3.0