1.0.2

Faster custom vocabulary

• The model behind custom vocabulary is now about a third smaller (123 MB instead of 190 MB) and runs about twice as fast. It downloads automatically after you update; vocabulary keeps working with the previous model until it finishes.

Fixes

• Muffle: switching away from Muffle shortly after it failed to start no longer leaves other apps' audio routed through its filter.

1.0.1

Superwhisper Linux 1.0.1

New

• ARM (aarch64) builds, for Omarchy and other Linux desktops on ARM. Offline needs an ARMv8.2 CPU with dot product (Apple M1 and later under Asahi Linux, Snapdragon X, Raspberry Pi 5); Cloud works on any.
• Search and play back dictations in History.
• Failed dictations are kept in History with their audio, so you can retry them.
• No more two-minute limit on a dictation.
• A notice when your free cloud words are running low and when they run out, with the option to switch to offline models.

Improved

• The first offline dictation after installing is much faster on a GPU. Graphics drivers now prepare while the models load, instead of during your first dictation.
• If a graphics driver gives wrong results, Superwhisper notices and uses the CPU instead. superwhisper doctor reports it.
• Offline is faster on processors with many cores, especially on ARM.

Fixed

• GNOME 51 support.
• Uninstalling now undoes the accessibility settings that vocabulary learning turned on.
• Shortcuts no longer stay paused after the service restarts or the settings window is closed unexpectedly.

Existing preferences, vocabulary, history, and downloaded models are preserved.

1.0.1-beta.2

Superwhisper Linux 1.0.1 beta 2

New

• ARM (aarch64) builds, for Omarchy and other Linux desktops on ARM. Offline needs an ARMv8.2 CPU with dot product (Apple M1 and later under Asahi Linux, Snapdragon X, Raspberry Pi 5); Cloud works on any.
• The first offline dictation after installing is much faster on a GPU. Graphics drivers now prepare while the models load, instead of during your first dictation.
• If a graphics driver gives wrong results, Superwhisper notices and uses the CPU instead, so you get correct text rather than garbage. superwhisper doctor reports it.

Improved

• Offline is faster on processors with many cores, especially on ARM.

Existing preferences, vocabulary, history, and downloaded models are preserved.

1.0.1-beta.1

Superwhisper Linux 1.0.1 beta 1

New

• Search and play back dictations in History.
• Failed dictations are kept in History with their audio, so you can retry them.
• No more two-minute limit on a dictation.
• A notice when your free cloud words are running low and when they run out, with the option to switch to offline models.

Fixed

• GNOME 51 support.
• Uninstalling now undoes the accessibility settings that vocabulary learning turned on.
• Shortcuts no longer stay paused after the service restarts or the settings window is closed unexpectedly.

Existing preferences, vocabulary, history, and downloaded models are preserved.

1.0.0

Superwhisper Linux 1.0.0

The first stable release, the same as 0.1.0 beta 16. Beta installs follow stable updates from here.

Existing preferences, vocabulary, history, and downloaded models are preserved.

0.1.0-beta.16

Superwhisper Linux 0.1.0 beta 16

New

• KDE Plasma and GNOME. Superwhisper now has its own settings window on desktops without Omarchy's panel. On KDE Plasma it can insert text directly through fcitx5 instead of pasting, and on GNOME the recording pill and tray menu are drawn through a Shell extension.
• Superwhisper can take over Omarchy's Dictate spot in the bar, and Omarchy's dictation key when one of your Superwhisper shortcuts uses it.
• The panel has an on/off switch in place of Quit.
• Superwhisper says when the free cloud allowance is used up.

Fixed

• Dictations paste where you are when you stop dictating, even if you switched windows while speaking.
• Recording overlays no longer fade out on Hyprland.
• Muffle no longer pauses other apps when Superwhisper starts or stops.
• Push-to-talk release, vocabulary learning, the recording glow and the Settings start-up are more reliable.
• History counts every saved dictation, not just the 50 it lists.
• Fixes from testing on CachyOS and KDE Plasma.

Existing preferences, vocabulary, history, and downloaded models are preserved.

0.1.0-beta.15

Superwhisper Linux 0.1.0 beta 15

Fixed

• Dictating with the Right Alt hold no longer opens the app's menu. Slack, Discord, VS Code, Obsidian, Firefox and other apps that open their menu bar when Alt is tapped on its own focused their File menu when you released Right Alt, and the dictation went into the menu instead of the message box. When Left or Right Alt on its own is your dictation shortcut, Superwhisper now keeps that key press from reaching the app. Alt still works in combinations such as Alt+F, and AltGr keyboard layouts are unaffected.

Existing preferences, vocabulary, history, and downloaded models are preserved.

0.1.0-beta.14

Superwhisper Linux 0.1.0 beta 14

New

• Dictation goes straight into the text field on Omarchy and Hyprland. Superwhisper now hands the text to the focused field through fcitx5, Omarchy's input method, instead of pasting it. Your clipboard is never touched, the keyboard layout and the app's paste shortcut no longer matter, and the text arrives about four times faster.
  • Line breaks work in regular apps without sending messages in chat apps. Multi-line dictations into a terminal are still pasted, so the lines don't run.
  • When there's no text field focused, or another window or menu has taken focus, Superwhisper pastes as before.
  • The installer restarts fcitx5 once to load Superwhisper's addon. Your input method settings stay as they are.
  • superwhisper doctor shows whether it's active. To go back to pasting everything: superwhisper general set linux.textCommit false.

On niri and other desktops, dictations are pasted as before.

Existing preferences, vocabulary, history, and downloaded models are preserved.

0.1.0-beta.13

Superwhisper Linux 0.1.0 beta 13

Fixed

• Dictation pastes into terminals outside Omarchy. In foot, Alacritty, kitty, Ghostty or WezTerm with their default settings, dictations pasted nothing. Terminals now paste with Ctrl+Shift+V, which works in all of them and on Omarchy.
• No more false "Copied to clipboard" in Discord, Slack, Notion and other web or Electron editors. Superwhisper checks that a paste landed, and in these editors it couldn't see the text, so it showed the notice even when the paste worked.
• A paste could pick up the previous clipboard. Superwhisper now waits until the clipboard holds the new dictation before pasting.
• superwhisper commands piped into head or similar no longer crash.

Improved

• When a paste does fail, the error report (if error logging is on) says more about where it happened: the kind of app, the toolkit and how the text was checked. It never includes the text, window titles or app names outside a fixed list of well-known apps.

Existing preferences, vocabulary, history, and downloaded models are preserved.