AI Dictation on Linux

What's actually available for Linux today

Most AI dictation apps do not run on Linux at all. We checked the platform pages of four of the best-known ones on 31 August 2026, directly on their own sites:

• Wispr Flow — wisprflow.ai states it is "Available on Mac, Windows, iPhone, and Android." No Linux.
• superwhisper — superwhisper.com's own header reads "AI Voice to Text for macOS, Windows & iOS," and the product section repeats "Available on Mac, Windows and iOS." No Linux.
• Willow Voice — willowvoice.com says "Available on Mac, Windows, and iPhone." No Linux.
• Aqua Voice — aquavoice.com lists its platforms as macOS, Windows and iOS. No Linux.

We also tried to check CleverType. As of the same date, both clevertype.ai and clevertype.com redirect to a GoDaddy "domain for sale" page — there is no product site left to check, on Linux or anywhere else.

This isn't a knock on those four products — they were simply never built for Linux, and their own sites say so plainly. If you sit down at a Linux desktop, none of them is an option.

The Linux-native options that do exist

Two open-source projects are worth naming, because they are real, maintained, and run entirely on the machine in front of you:

• nerd-dictation is a command-line speech-to-text tool built specifically for Linux, using the offline VOSK engine. You start and stop listening yourself — there is no push-to-talk out of the box, you script that part — and it types into whichever window has focus using xdotool, ydotool, dotool or wtype, depending on your display server. It is a single, hackable Python script under the GPL-3.0 license.
• Speech Note (the dsnote project) is a Linux desktop app, also packaged for Sailfish OS, for note-taking, reading and translation, built around fully offline models — Vosk, whisper.cpp, Faster Whisper, Coqui STT and others for recognition, plus its own offline text-to-speech and machine translation. It installs from Flathub, the Arch AUR, or openSUSE Packman.

Both are genuinely offline and genuinely free. Neither does what Qenzo does, though: they turn speech into text, not into a finished message. There is no tone, no chosen recipient, no "write this the way a lawyer writes," no translating what you said into another language while keeping your meaning intact. That layer needs a model doing more than recognition — a different kind of project from either of these.

Qenzo on Linux

Qenzo's Linux client is a desktop app, not a keyboard: hold a key, speak, release, and the finished text is typed into whatever window had focus when you started — a terminal, a browser field, a chat window, a code editor, anything with a text cursor. It is built from the same codebase as the macOS and Windows clients.

Installing it

The install command is one line:

curl -fsSL https://qenzo.ai/install.sh | sh

It does not ask for a password. Everything it puts on the machine goes into your home directory: the app itself at ~/.local/share/Qenzo/Qenzo.AppImage, a menu entry at ~/.local/share/applications/qenzo.desktop, and an icon under ~/.local/share/icons. The download is around 130 MB. Running the same command again updates the app in place. Removing it is one rm -rf command against those same paths — nothing is left in /usr, /etc, or anywhere outside your own account.

If you would rather not pipe a script into a shell, the same AppImage is downloadable directly, and a .deb is offered for Debian- and Ubuntu-based systems — Mint, Pop!_OS, elementary and the rest. AppImages are built to run on most current x86_64 desktop distributions without installation, Fedora, openSUSE and Arch included, though Qenzo does not publish a tested-distribution list beyond that.

X11 and Wayland

On X11, Qenzo works natively. On Wayland, an ordinary application is not allowed to type into an arbitrary window by itself, so Qenzo pastes the finished text through the ydotool helper instead. ydotool works by talking to the kernel's uinput device to simulate keystrokes; that means keeping its ydotoold service running with access to /dev/uinput, which ydotool's own documentation says usually requires root permissions (some distributions offer a udev-based workaround instead), set up once. Qenzo's site names ydotool as the requirement without detailing that setup.

What needs a connection

The Linux client is a dictation client, not the full keyboard the phone apps are — there is no on-device typing layer to fall back to, because on the desktop there is no typing layer at all; its only job is turning speech into finished text. That job needs a connection every time: the audio goes to Qenzo's servers, is understood, cleaned up, translated or re-toned if you asked for that, and the text comes back. Nothing about it runs on the machine. If the network is down, dictation does not work.

The recording itself is never written to disk on either end — it lives in memory long enough to be turned into text, then it is discarded. Private mode skips the history row entirely. The privacy policy at qenzo.ai/privacy covers who processes what in more detail.

Languages and price

The Linux client speaks the same 42 languages as the phone apps, mixed mid-sentence if that is how you talk, with the same seven tones and eight kinds of recipient to write for. Pricing is the same everywhere: the first 20 actions need no account at all, linking an email adds 50 more, and Qenzo Pro is $9.99 a month or $79.99 a year through the app stores, or $79 a year bought directly on qenzo.ai. Most countries pay a lower regional price than the US number, down to $1.99 a month. There is no free trial; nothing is charged until you subscribe.

Questions Linux users actually ask

Does it need root?

No. The installer never asks for a password; uninstalling is deleting the same three folders it created.

What exactly does it install?

One AppImage under ~/.local/share/Qenzo, one .desktop menu entry, and one icon. Nothing runs as a system service, and nothing runs unless you launch it yourself.

Does it phone home?

For the AI part, yes, necessarily: every dictation is sent to Qenzo's servers, because that is where recognition and writing happen, with no offline fallback. The audio itself is never written to disk and is discarded once turned into text; private mode skips saving the resulting text too. The privacy policy has the detail on who processes what.

Does it work in a terminal, in Emacs, or in a browser text field?

It should, by construction, rather than by app-specific testing: Qenzo types the finished text into whichever window had focus, using the same OS-level mechanism (X11 input, or ydotool on Wayland) regardless of what that window is. A terminal, Emacs, and a browser's text box are all just windows to that mechanism.

Which distributions are supported?

No tested list is published. AppImage is built to run across most current x86_64 desktop distributions without a package manager; the .deb targets Debian- and Ubuntu-based systems specifically.

Are nerd-dictation or Speech Note good alternatives?

For offline, free, open-source speech-to-text with no server involved, yes — both are real, maintained projects. For speech turned into a finished message with a chosen tone, translated into another language, or written the way a given profession writes, neither does that; they transcribe what you said, they do not rewrite it.

Qenzo against the alternatives, on Linux

Axis	Qenzo	Wispr Flow / superwhisper / Willow Voice / Aqua Voice	CleverType	nerd-dictation / Speech Note
Linux desktop app	Yes — install script, AppImage, .deb	No — checked 31 Aug 2026, none lists Linux on its own site	Unverifiable — clevertype.ai and .com are parked "for sale" domains as of 31 Aug 2026	Yes — Linux (and, for Speech Note, Sailfish OS) is what they are built for
What it does	Speech to finished, translated, re-toned text	AI dictation, on the Mac/Windows/phone platforms they support	Unknown	Speech to raw text, plus offline translation in Speech Note
Where speech is processed	Server-side; needs a connection every time	Not checked for this page	Unknown	Fully offline, on the machine
Cost on Linux	Free tier, then $9.99/mo or $79/yr on qenzo.ai	N/A — no Linux version to price	Unknown	Free and open source

The honest summary: if the goal is dictation on a Linux desktop that stays entirely offline and a raw transcript is enough, nerd-dictation or Speech Note are real, free options worth trying first. If the goal is speaking messily and getting back a finished, correctly-toned message in the language you chose — the way Qenzo already works on Android, macOS and Windows — none of the four commercial competitors checked here offer that on Linux at all, and Qenzo does.