Whisper Terminal · v0.1.0 · preview

Dictate anywhere. Open a terminal with a key.

A light app for Windows and Linux that lives in the tray. Press a key, speak, and the text is pasted where your cursor is. Press another and a terminal drops over whatever you are doing, with your usual programs in it.

Windows 10 22H2 or 11 (64-bit) and Linux (x86_64). The Windows build is a first preview that has not yet been run on a real Windows PC. Linux needs a recent distribution: glibc 2.43 (Ubuntu 26.04, Fedora 44). macOS is not available yet: it is planned, with no date.

What it does

Dictate anywhere

Ctrl+Space: press, speak, press. The text is pasted into the window you were in; if it cannot paste, it stays on the clipboard and says so. Other keys paste into the terminal, with or without Enter at the end.

Terminals that belong to no project

One key drops a window of tabs and splits over anything. On Linux, Claude Code, vim, less, top and bash run in it; on Windows it opens Git Bash by default (see below). Hiding it keeps the terminals alive.

Speech is transcribed on your machine

With whisper.cpp, on the graphics card (Vulkan) when there is a useful one and on the CPU when there is not. On Linux it prefers a discrete card and uses an integrated one only if it turns out faster than the CPU; on Windows it is built to use the Vulkan card your driver offers, and Settings → Speech is meant to say in words which device ran it and, if it fell back to the CPU, why. The speech engine starts with the first dictation and closes by itself two minutes after the last.

It does not lose what you said

Every dictation is in the history before the first audio sample is recorded, and the audio is kept if something fails, so you can try again.

Light at idle

On Linux, freshly started, in the tray with no window open, it takes about 21 MB and almost no CPU (measured on the development machine). With windows open it takes more. It has not been measured on Windows yet.

With agents, if you want them

It opens Claude Code profiles in its tabs (Claude Code is not included: you need your own install and account). On Linux it also includes an MCP server so an agent can see and drive the desktop; that part has not been tried on a real desktop yet, and by default it asks for your yes before every keyboard or mouse action and before commands that are not on an allowed list: it is a consent prompt, not isolation. The Windows build cannot drive the desktop yet.

The shortcuts

Four keys, the same idea on both systems. All of them can be changed.

What it does Windows Linux
Dictate: press, speak, press again Ctrl+Space Ctrl+Space
Open the terminals Win+Enter Super+Return
Dictate into the terminal Ctrl+Shift+Space Ctrl+Super+Space
Dictate into the terminal and press Enter Ctrl+Shift+Enter Ctrl+Super+Return

On Windows, Win+Space and Ctrl+Win+Space belong to Windows' own input switcher and cannot be registered, which is why the terminal keys use Ctrl+Shift. If another program already uses a combination (an input method often takes Ctrl+Space), Settings → Shortcuts says Windows refused it: change the key in %APPDATA%\whisper-terminal\settings.toml and press "Rebind shortcuts".

On Linux, GNOME registers them itself; on other desktops some cannot, and ~/.local/bin/whisper-terminal --doctor says which and how to assign them by hand.

Before you download

It is a preview (0.1.0)

This is the first version we publish. It can break and it will change. It has no automatic updates yet: to update, quit the app and install the new one over the old one (Linux: extract the new tarball over it; Windows: run the new installer, or unpack the new zip in another folder).

WindowsThe Windows build has never run on a Windows PC

It was built on a Linux machine. What could be checked there was checked: it builds and links for Windows, and its tests were compiled for Windows, but nobody has run it on a Windows PC yet, the installer included. The shortcuts, the tray icon, pasting into other programs and the graphics card are what only a real PC shows. You would be the first test: if something fails, the last part of the guide below says what to send us.

It is not signed

There is no code-signing certificate. Compare the SHA-256 below with your file's before you run it. On Windows your browser may ask you to confirm the download, and SmartScreen shows "Windows protected your PC" for any new program from a small publisher: choose "More info", then "Run anyway". Windows Defender may take a few seconds to look at the first start.

WindowsGit Bash is the default shell

The terminals open Git Bash, so install Git for Windows first (gitforwindows.org). Without it they fall back to PowerShell, then Command Prompt (cmd); both are profiles too. WSL is not supported.

WindowsYour graphics card, through Vulkan

Dictation runs on the graphics card through Vulkan when there is a useful one (NVIDIA, AMD and Intel drivers bring the Vulkan library: keep yours current) and on the CPU when there is not. Settings → Speech says in words which device runs it and, if it fell back to the CPU, why and what to do; every dictation also writes a log line with the device and the time. That path has not been run on a real PC yet: whether your driver accepts it is the first thing a test shows.

WindowsWhat the Windows build does not have yet

Driving the desktop (the agent that clicks and types for you), the connector hub (Telegram, Slack, the web page and the phone), the WhisperKey import, the sounds and the DIY OS parts. Their windows are not in the menu. Windows refuses a paste into a program that runs as administrator: the text stays on the clipboard and the orb says "copied".

WindowsWhat it writes on Windows

Your settings are in %APPDATA%\whisper-terminal; the history, recordings, downloaded models and logs are in %LOCALAPPDATA%\whisper-terminal. The installer adds a Start-menu entry and an entry in Settings → Apps. "Start at login" (off until you turn it on in Settings → System, or tick "Start Whisper Terminal when I sign in to Windows" on the installer's last page) writes one value under HKCU\Software\Microsoft\Windows\CurrentVersion\Run. Nothing needs administrator rights, and nothing is sent anywhere except the speech model download.

LinuxLinux: a recent distribution

x86_64. It was built on Ubuntu 26.04 and needs glibc 2.43 or newer: on an older distribution (Ubuntu 24.04 or Debian 12, for example) it does not start. Check yours with ldd --version. GNOME is the reference desktop; on others some keys cannot register themselves: ~/.local/bin/whisper-terminal --doctor says what is missing and how to assign them by hand.

LinuxRemote access is off

Remote access (chat, phone) ships switched off and turns on only when you ask for it: by pairing a device or a chat at the machine, or with ~/.local/bin/whisper-terminal --hub on. The Windows build does not have it.

LinuxIt writes outside its folder

If you turn on autostart, it creates an entry in ~/.config/autostart. On GNOME 47 or older it registers keyboard shortcuts of its own (whisper-terminal-*). And at start it links four skills into ~/.claude/skills (for Claude Code). The guide below says how to remove all of it.

Closed source, and no usage licence yet

It is proprietary software. The LICENSE file in the package says: "No licence to use, copy, modify or distribute it is granted except by a separate written agreement with the copyright holder". It also says this is a provisional value while the licence is decided. If you need that agreement in writing, write to support@vibedout.app. Third-party notices (THIRD_PARTY_NOTICES.md) are partial: the full list for a release is not generated yet.

The speech model

On Linux the package includes the base.en model (141 MB), which only understands English; the Windows package carries no model. On the first dictation the app starts downloading, from Hugging Face, the one it recommends for your machine (for example small.en-q5_1, 181 MB; on Windows with the curl that comes with Windows); that first attempt does not transcribe, so try again when it finishes. To dictate in another language, pick a model without ".en" in Settings → Speech.

Download

Windows

v0.1.0 · preview

Windows 10 22H2 or Windows 11 · 64-bit (x86_64) · a microphone · for the graphics card, a current NVIDIA, AMD or Intel driver (it brings the Vulkan library) · no administrator rights, no Visual C++ Redistributable

First preview: it has not yet been run on a real Windows PC. You would be the first test.

Download the installer 14 MB

Per user: it installs in your own folder, with a Start-menu entry and an uninstaller, and asks for no administrator rights. It is not signed: Windows will show SmartScreen (see "Before you download").

How to install it on Windows

File WhisperTerminal-0.1.0-windows-x86_64-setup.exe

SHA-256 b890c69fcf991409ee90ec908596cc56bb6985548dd8cfd0b8b9e3fe8439fae4

Portable zip

Download the zip 31 MB

Unpack it in a folder you own (not Program Files) and run whisper-terminal.exe. Nothing is installed; "Start at login" stays off until you turn it on.

File WhisperTerminal-0.1.0-windows-x86_64.zip

SHA-256 c86b90eeb6fbe8f4e4398a454ee570f92a299ad7df8c26c12840a40f71817eb3

Test kit, only for people helping to test

The program's own tests as .exe files and a run-tests.cmd. It is not the app and does not install it. Not recommended unless you were asked for it.

Download the test kit 7.9 MB

File WhisperTerminal-0.1.0-windows-x86_64-tests.zip

SHA-256 d11a46e99ff2397df455f3f6f0e438d2fc6e852b14e372caf682046116c98472

Linux

v0.1.0

x86_64 · glibc 2.43 or newer (Ubuntu 26.04, Fedora 44) · GNOME as the reference desktop · a Vulkan driver (Mesa or NVIDIA)

Download 171 MB

The guide unpacks it in ~/.local/opt; the programs, the libraries and a speech model are inside. The program installs into your user folder; only the libraries of step 2 ask for sudo.

How to install it on Linux

File whisper-terminal-linux-x86_64.tar.gz

SHA-256 3fb6e1dfbf3686ec4e5fe30c4ac187475e1d5b417085bd32fce20450647c78f8

macOS

Not yet

There is no macOS version yet. It is planned, with no date. There is no button because there is nothing to download.

Install

Pick your system. Both install in your user folder, with no administrator rights; only the Linux libraries of step 2 ask for sudo.

Windows · Linux

Windows

The installer is per user: no administrator rights, nothing outside your own folders. The portable zip further down is the same program without an installer.

A first preview that has never run on a real PC: these steps say what it is built to do. Tell us where it differs.

1. Check the download. In PowerShell, in the folder where you saved it (usually Downloads: cd $HOME\Downloads gets you there):

$expected = "b890c69fcf991409ee90ec908596cc56bb6985548dd8cfd0b8b9e3fe8439fae4"
(Get-FileHash .\WhisperTerminal-0.1.0-windows-x86_64-setup.exe -Algorithm SHA256).Hash -eq $expected

It must print True. If it prints False, do not run it: download it again.

2. Optional but recommended: install Git for Windows so the terminals open Git Bash. Without it they open PowerShell.

3. Run the installer.

Double-click it. Windows will probably say "Windows protected your PC" (SmartScreen) because the program is not signed yet: choose "More info", then "Run anyway". The installer asks for a folder (on a first install it suggests %LOCALAPPDATA%\Programs\WhisperTerminal) and, on its last page, offers to start Whisper Terminal now (ticked) and to start it when you sign in to Windows (not ticked).

4. Start it and see what it does.

From the last page of the installer, or Whisper Terminal in the Start menu. A small icon appears in the notification area (click ^ if Windows hides it) and the terminals open once: that is what starting it does (when it starts at login it starts quietly, with only the icon). Right-click the icon for the menu (Start dictation, Terminals, History, Settings, Quit). Then press Ctrl+Space in any program, speak, and press it again: the first dictation starts downloading a speech model (a progress line shows in Settings), so dictate again when it finishes.

The portable zip instead of the installer

Check it like step 1, with the zip's name and its SHA-256 from the card above:

$expected = "c86b90eeb6fbe8f4e4398a454ee570f92a299ad7df8c26c12840a40f71817eb3"
(Get-FileHash .\WhisperTerminal-0.1.0-windows-x86_64.zip -Algorithm SHA256).Hash -eq $expected

Then unpack it in a folder you own (right-click, "Extract All") and double-click whisper-terminal.exe inside the folder, or in PowerShell:

Expand-Archive .\WhisperTerminal-0.1.0-windows-x86_64.zip -DestinationPath $HOME
Start-Process "$HOME\WhisperTerminal-0.1.0-windows-x86_64\whisper-terminal.exe"

SmartScreen may show up here too, at the first start. To have it start with your session, turn on "Start at login" in Settings → System.

Remove it

If you used the installer: Settings → Apps → Installed apps (on Windows 10: Apps & features) → Whisper Terminal → Uninstall (or run uninstall.exe in the install folder). It quits the program, deletes its files, the Start-menu entry and the start-at-login value, and asks whether to also delete your settings, history and downloaded models (the default answer is No).

If you used the zip: right-click the tray icon and choose Quit; if you turned on "Start at login" (Settings → System), turn it off first; then delete the folder.

Your settings (%APPDATA%\whisper-terminal) and your history, recordings, models and logs (%LOCALAPPDATA%\whisper-terminal) stay until you delete them. To delete them, with the app quit:

Remove-Item -Recurse "$env:APPDATA\whisper-terminal", "$env:LOCALAPPDATA\whisper-terminal"

If something does not work

Save what this PC looks like to the program, in the folder that has whisper-terminal.exe (the install folder, %LOCALAPPDATA%\Programs\WhisperTerminal unless you chose another, or the unpacked zip's folder; --doctor does not start the app):

.\whisper-terminal.exe --doctor | Out-File doctor.txt

Send doctor.txt and the log (%LOCALAPPDATA%\whisper-terminal\whisper-terminal.log, and wt-stt.log beside it, from the speech engine) to support@vibedout.app, with what you pressed and what you expected. If a shortcut says Windows refused it, tell us which key. whisper-terminal.exe --status and whisper-terminal.exe --stt probe help too.

Linux

Seven steps, and only the second asks for sudo. Everything else stays in your user folder.

1. Check your glibc: the first line must end in 2.43 or more.

ldd --version | head -1

If it is lower, the program does not start (it says version GLIBC_2.43 not found) and there is no point in going on.

2. Install the libraries it uses. On Debian, Ubuntu and derivatives:

sudo apt install libxkbcommon-x11-0 libvulkan1 mesa-vulkan-drivers alsa-utils

On Fedora:

sudo dnf install libxkbcommon-x11 vulkan-loader mesa-vulkan-drivers alsa-utils

On GNOME the tray icon needs the AppIndicator extension (Ubuntu ships it on; on Fedora it is a separate install). Without it the app works, without an icon.

3. Verify, unpack and link the command, in the folder where you saved the file (usually ~/Downloads):

echo "3fb6e1dfbf3686ec4e5fe30c4ac187475e1d5b417085bd32fce20450647c78f8  whisper-terminal-linux-x86_64.tar.gz" | sha256sum -c - \
  && mkdir -p ~/.local/opt ~/.local/bin \
  && tar -xzf whisper-terminal-linux-x86_64.tar.gz -C ~/.local/opt \
  && ln -sf ~/.local/opt/whisper-terminal/bin/whisper-terminal ~/.local/bin/whisper-terminal

The first command must say "OK". If it says "FAILED", do not unpack anything: download again. If you already had it installed, quit it first with ~/.local/bin/whisper-terminal --quit. If you move the folder later, repeat the last line of step 3 (and step 7), then step 6.

4. A launcher in the application list (in bash or zsh):

mkdir -p ~/.local/share/applications
cat > ~/.local/share/applications/app.vibedout.WhisperTerminal.desktop <<EOF
[Desktop Entry]
Type=Application
Version=1.0
Name=Whisper Terminal
Comment=Dictate anywhere and open overlay terminals
Exec="$HOME/.local/bin/whisper-terminal"
Icon=utilities-terminal
Terminal=false
Categories=Utility;
Keywords=dictation;voice;speech;whisper;terminal;overlay;
StartupWMClass=app.vibedout.WhisperTerminal
StartupNotify=false
X-GNOME-UsesNotifications=true
EOF

This one does show in the list. The package does not carry one, which is why it is written here.

5. See what works on your machine:

~/.local/bin/whisper-terminal --doctor

--doctor does not start the app: it marks what is missing with FAIL (keys, tray, model) and says how to fix it. On a fresh install it is normal to see FAIL on "autostart" (you have not turned it on), "dictation model" (the recommended model downloads at the first dictation), "claude" (if you do not have Claude Code) and "skills" (they are linked at the first start). The default keys are in the shortcuts table above.

6. Start it now, in the tray with no window, and make it start with your session:

setsid -f ~/.local/bin/whisper-terminal --background
~/.local/bin/whisper-terminal --autostart on

The second command creates an autostart entry that is not shown in the application list (on purpose: it is a different file from the launcher of step 4). ~/.local/bin/whisper-terminal --autostart off removes it.

7. Optional: use the model that comes in the package, with nothing to download (English only):

mkdir -p ~/.local/share/whisper-terminal/models
ln -s ~/.local/opt/whisper-terminal/share/whisper-terminal/models/ggml-base.en.bin ~/.local/share/whisper-terminal/models/ggml-base.en.bin

Then pick it in Settings → Speech. If you had already downloaded that model, the command leaves it as it is.

Remove it

~/.local/bin/whisper-terminal --quit
~/.local/bin/whisper-terminal --autostart off
~/.local/bin/whisper-terminal --shortcuts remove
~/.local/bin/whisper-terminal --skills unlink
[ -L ~/.local/share/whisper-terminal/models/ggml-base.en.bin ] && rm ~/.local/share/whisper-terminal/models/ggml-base.en.bin
rm -f ~/.local/bin/whisper-terminal ~/.local/share/applications/app.vibedout.WhisperTerminal.desktop
rm -rf ~/.local/opt/whisper-terminal

This stops it, removes its GNOME shortcuts and the skill links, and deletes the program. Your settings (~/.config/whisper-terminal), your history and downloaded models (~/.local/share/whisper-terminal) and its log (~/.local/state/whisper-terminal) stay until you delete them.

Coming from WhisperKey on Linux? ~/.local/bin/whisper-terminal --migrate plan shows what it would import from your settings and history, changing nothing of WhisperKey's. The Windows build does not import yet.

WhisperKey is still available for Windows and Linux. WhisperKey · See all the tools