Gabble changelog

0.1.7 beta · 8 October 2026

A livelier welcome: the duck quacks, rings of sound leave it and “It types.” types itself
With Reduce Motion turned on in macOS the welcome screen stays still

The pill shows over full screen apps again. After Gabble had been running for a while it could go missing there, even though dictation still worked
Starting a new dictation right after the last one no longer leaves the pill hidden for a moment

The disk image opens as a proper window: drag the duck onto Applications
Other languages keep their words. Gabble removed “er” and “um” as hesitations everywhere, which ate real words in German, Danish, Portuguese and more. It now only does that in English
Numbers said twice stay: “555 555 1234” no longer loses a 555
Pasting works on the Dvorak keyboard layout
Clipboard managers no longer fill up with your dictations
Lighter when idle, and a long history opens faster
Switching speech models while one is still loading no longer leaves the wrong one in use

The menu bar icon is the duck now. Its bill opens while Gabble is listening
Redesigned menu bar window: one clear dictate button, mode and model in a single panel, and the last dictation with the app it went to

The duck is livelier: it rocks and stretches as you talk, its tuft sways and it blinks
The pill no longer shows the mode name while you dictate

Send after dictating. Tick Send next to an app in the per-app rules and Gabble presses Return when you let go of the key, so the message goes without touching the keyboard. On for the Claude desktop app, off everywhere else
The pill has a duck in it now. Its bill follows your voice and the sound ripples away from it
The pill sits at the top of the screen, under the notch, instead of at the bottom

Automatic updates. Gabble checks this site once a day and installs new versions for you; turn that off, or check by hand, in Settings → Updates
New app icon: a duck, mid-quack

If you installed 0.1.0, download this version once from the home page. From here on Gabble updates itself.

First public build.

Hold-to-talk dictation into any app, with hands-free tap mode and Esc to cancel
On-device speech recognition: NVIDIA Parakeet v2 and v3, Whisper Tiny, Base, Small, Distil Large v3 and Large v3 Turbo
Modes: Clean, Verbatim, Polished, Email, Message, Notes, Developer and custom modes, with per-app rules
Optional AI cleanup through Ollama, Anthropic or any OpenAI-compatible API
Command mode: rewrite selected text by voice
Custom dictionary, voice snippets and spoken formatting commands
Searchable history with Markdown and JSON export, and usage stats
File transcription for audio and video
Menu bar controls, floating waveform pill, launch at login