Changelog — OmniVox

Voice Commands — say "new line", "new paragraph", or "delete last word" while dictating to control formatting with your voice; toggle on/off in Settings
Writing Styles — choose between Formal, Casual, and Very Casual output styles; Formal adds proper capitalization and trailing punctuation, Casual keeps capitalization but skips forced periods, Very Casual lowercases everything for chat-friendly output; configurable globally in Settings and per context mode
Ship Mode — automatically presses Enter after transcription to send messages instantly in chat apps like Slack, Discord, and Teams; toggle in Settings or from the pill overlay
Auto Context Switch Fallback — when switching between apps, OmniVox now automatically returns to your General context mode if the foreground app doesn't have a binding, instead of staying stuck on the previous mode
Ghost Mode — hide the floating pill overlay completely while keeping it interactive; click or right-click the invisible area to reveal it again
Reliable Text Output — replaced SendInput-based keystroke simulation with clipboard paste for the TypeSimulation output mode; eliminates dropped characters and repeated-character glitches on Windows, especially for longer transcriptions; clipboard is saved and restored automatically
Zero-Allocation Audio Capture — audio callback no longer allocates heap memory per frame; mono conversion and resampling happen inline directly into the shared buffer
Stack-Only Denoise — RNNoise processing now uses stack-allocated frame buffers instead of creating intermediate heap vectors, reducing GC pressure during recording
Pill Overlay Polish — toggle buttons use a consistent dark amber accent color for better visibility on light backgrounds; Ghost mode button matches the popup menu background
Fixed transcription corruption caused by style-conditioning sentences leaking into the Whisper initial prompt