Changelog

Notable changes to Kass for users. Each release gets a section here, newest first. The website shows this file at kass.mrgnhnt.com/changelog.

0.7.1 — September 30, 2026

New

Try new features early with beta updates. Turn on Beta updates in Settings › General to get new features before they’re public. Turn it off to go back to public releases; you stay on your beta until the next public one is newer.
Fix what Kass just typed by voice (beta). Start a dictation with “fix that” to change your last one: “fix that, Morgan not Megan”, “fix that, delete actually”, or spell a name out. No selecting needed. Works where Kass can edit the field directly (not yet in Safari pages or Firefox).

Improved

Dictionary names can leave similar names alone (beta). With Beta updates on, you can stop a name from respelling similar ones. A Meghan entry used to also turn Megan and Meagan into Meghan. Turn off Also fix words that sound like it on an entry to fix only its exact spelling.
Say a phrase again without doubling it (beta). With Beta updates on, put the cursor mid-sentence and dictate the new words, running on into the words already after the cursor: Let's meet | at noon tomorrow. plus “on Friday at noon tomorrow” now gives Let's meet on Friday at noon tomorrow. Kass drops the words you said again, even with a small spelling difference, and keeps the field’s own. Live text may show them for a moment before it settles.

Changed

Herga is now Kass. Your captures, dictionary, writing styles and settings come along on their own. Because macOS treats a renamed app as a new one, Kass asks again for Microphone, Accessibility and Input Monitoring the first time it opens. The website moved to kass.mrgnhnt.com.

Fixed

⌘H keeps Kass out of the way. Dictating after hiding Kass with ⌘H used to bring its window back along with the pill. Now only the pill shows, and the window returns when you click Kass in the Dock or ⌘Tab to it.
The global keys work right after reinstalling Kass. They used to wait until you brought Kass’s window to the front. Now Kass notices on its own when macOS confirms the permissions you’d already given it.

0.6.2 — September 30, 2026

Fixed

Writing styles learn when you leave off the final period. Fixing a word in a capture used to count as keeping its period, so a style kept adding periods you’d removed. Now only the punctuation you actually change counts, and Herga recounts your past corrections the first time it opens.
Dictionary words are spelled right from the start. A word in your dictionary used to be fixed only where Whisper already heard it right, so a misheard one stayed wrong until you’d corrected it. Now Herga also fixes near misses like Kubernetis or cuber netes, and leaves everyday words you actually said alone.
The global keys work right after Herga opens at login. They used to wait until you’d opened Herga’s window once.
Add a word to the dictionary while you correct it. Select a word in a capture, even mid-correction, and just type its correct spelling. You no longer have to save the correction first.

0.6.1 — September 30, 2026

Improved

Updates download in the background. When a newer version is out, Herga downloads it while you work. Click Restart in the sidebar (or Restart to update in Settings › General) to switch to it, or it installs the next time you quit. Your permissions carry over. Versions before this one need a one-time manual update from the download page.

0.6.0 — September 30, 2026

The first release of Herga as a dictation app. This fork of jamiepine/voicebox 0.5.0 drops text-to-speech and focuses entirely on turning your speech into ready-to-send text on Apple Silicon Macs.

Voicebox is now Herga, from jerga, Spanish for slang. Your captures, styles, dictionary and settings carry over the first time Herga opens. macOS treats it as a new app, so it asks for Microphone, Accessibility and Input Monitoring once more. The old Voicebox.app can go in the Trash; the install script does that for you.

New

Guided setup. First-run onboarding downloads the models in the background while it walks you through permissions and your hotkey, then has you say your name, read a messy line, reply to a text in your own words to teach your writing style, and rewrite a paragraph by voice. If macOS needs Herga to quit for a permission, setup reopens on the same step.
Native, streaming dictation. The microphone is captured in native code and streamed while you speak, and cleanup runs a sentence at a time, so there’s less to wait for when you let go.
Text lands where you were typing. Text is inserted through Accessibility, typed keystrokes or the clipboard, whichever works for the app. When it has to use the clipboard, it puts yours back.
Writing styles per app. Every app gets a style, and each style learns on its own. Switch style by saying so at the start of a dictation (“use formal mode”). The pill shows the style’s name and plays a sound when it changes.
Teach by replying. Teach a style how you write by replying to a few short conversations.
Dictionary. Add terms and spoken replacements for everywhere, a style or specific apps. Select a word in a capture to add it.
Command Mode. Select text in any app, hold right ⌘ + right ⇧ and say how to rewrite it. It comes with the Polish and Prompt Engineer transforms, and you can save your own.
Spoken commands. Say line breaks, lists, quotes, brackets, braces, slashes, pipes and carets. Spell things out letter by letter (“capital C…”). Say “paste from clipboard” to insert your clipboard.
Self-corrections. “Tuesday, no actually Wednesday” becomes “Wednesday”. Repeats, stutters and restarts are removed.
Correction learning. Fix a capture and the fix is used right away. Older fixes are folded into rules, which are kept only if they do at least as well on your past corrections.
Captures redesign. Captures are grouped by app and show the audio next to what Whisper heard and the cleaned-up text. A Check badge flags cleanups worth a second look.
Insights. See words dictated, speaking pace, time saved and your most-used apps.
History retention. Choose how long to keep captures. Nothing is deleted until you confirm.
Escape to cancel a dictation, even while it’s still finishing.
Sound cues for start, stop and errors, with a volume setting.
Launch at login, on by default, with the window hidden.
Command palette. Press ⌘ K to jump to any setting or action.
Update notices. Herga tells you when a newer release is out.

Improved

After an update, Herga replaces a local server still running from the old version instead of reusing it, so the new version’s backend is always the one running.
Model weights stay loaded between dictations, and the first dictation after launch is as fast as the rest.
The pill shows on the display you’re working on, shows recording as soon as you press the keys, and never takes keyboard focus.
Audio without a voice in it is never transcribed, and Whisper’s repeated-phrase loops are removed.
Cleanup is rejected and your own words are used instead when it answers your question instead of writing it down, copies an example, or adds words you didn’t say.
The microphone is released after every dictation.
The DMG opens to a drag-to-Applications install window.

Removed

Text-to-speech, voice profiles and cloning, stories, effects, the MCP server, and Herga Cloud.
Support for Windows, Linux and Intel Macs. Herga now runs only on Apple Silicon.