1. Wispr Flow’s Linux status, as of October 2026
Wispr’s downloads page offers a “Vote to build for Linux” button with the note “Help us prioritize Linux support by casting your vote. We’ll notify you when our native app is ready.” Its help article on terminals is blunter: “There is no native Linux application.” The same downloads page lists a web version with a “Get notified” button rather than a link, so there is no browser route to Flow on Linux either.
Whatever the platform, Wispr’s Data Controls page says transcription always occurs in the cloud, which Wispr describes as the best way to provide accurate, low-latency transcription. So a native Linux app, if one ships, would change where the hotkey lives rather than where your audio is recognized, unless Wispr changes that policy. If you need speech to stay on the machine, the tools below are the route regardless of Wispr’s roadmap.
2. What to compare in a Linux dictation tool
Linux adds two questions that Mac and Windows users rarely face. The first is how text reaches the focused window: on X11 most tools use xdotool, while on Wayland they rely on helpers such as wtype, dotool, or ydotool, and support varies by compositor. The second is the global hotkey: Handy’s documentation says that on Wayland, system-level shortcuts must be configured in the desktop environment, and Speech Note says its global shortcuts need the XDG Desktop Portal GlobalShortcuts interface.
Then come the familiar questions: does recognition run locally or in a cloud service, which package format matches your distribution, and how much assembly the tool expects from you.
- Text insertion: xdotool on X11; wtype, dotool, or ydotool on Wayland.
- Hotkey: an in-app shortcut or a desktop shortcut bound to a command.
- Speech path: local model, cloud API, or both.
- Packaging: deb, rpm, AppImage, Flatpak, or a Python script.
3. Desktop apps: Spokenly, Handy, and OpenWhispr
Spokenly for Linux (version 1.0.3, updated October 2026) is a system-wide dictation app with free local Whisper and Parakeet models that run offline, plus optional cloud engines through your own API keys or a Spokenly Pro subscription. It ships x86_64 deb, rpm, and AppImage packages; the AppImage supports Ubuntu 20.04 and later, while the native packages need glibc 2.39 or newer (Ubuntu 24.04+, Fedora 40+, Debian 13+). Spokenly states X11 and Wayland support and cautions that shortcuts and text insertion vary by compositor; ARM64 builds are not listed.
Handy is the free, MIT-licensed, open-source app that IraVoice is built on. It transcribes locally with Whisper or Parakeet and installs on Debian and Ubuntu with sudo apt install ./Handy_*.deb. Its README is candid about Wayland: support is limited, text input needs wtype or dotool (dotool requires joining the input group), Ubuntu 26.04 needs ydotool because wtype does not work there, and global shortcuts are bound in the desktop environment to Handy’s command-line flags.
OpenWhispr is MIT-licensed and lets you choose local models, including Whisper and NVIDIA Parakeet, or cloud processing with your own keys; its README says it collects no data and has no telemetry. Its GitHub releases include Linux deb, rpm, AppImage, and tar.gz builds (1.10.2, September 2026). Check its current documentation for your compositor before relying on it under Wayland.
4. Fully offline tools: Speech Note and nerd-dictation
Speech Note (dsnote) is an MPL-2.0 Linux app for notes, reading, and translation that processes voice and text entirely offline, with speech engines including Vosk, whisper.cpp with Parakeet, Faster Whisper, and april-asr. Install it from Flathub as net.mkiol.SpeechNote, with optional NVIDIA and AMD add-ons, or from the AUR. Its “Insert into active window” mode, started from a global shortcut or the command line, turns it into a voice typing tool; on Wayland it needs the ydotool daemon running, and its global shortcuts need the GlobalShortcuts portal, which it says only the latest KDE Plasma and GNOME support.
nerd-dictation is a GPL-3.0, single-file Python script for offline speech to text with the VOSK API. You download a Vosk model, bind nerd-dictation begin and nerd-dictation end (and cancel) to keys yourself, and choose an input tool: xdotool by default, ydotool or dotool for X11, Wayland, and TTYs, or wtype for Wayland. Its configuration file is Python, so you can rewrite text before it is typed. It suits people who want to assemble and tune every piece.
5. IraVoice on Linux: a beta AppImage with local Parakeet
IraVoice 1.2.2 supports Linux as a beta. One command installs it, curl -fsSL https://iravoice.com/install.sh | bash: an AppImage that runs on any 64-bit distribution from 2022 or later (Ubuntu 22.04, Debian 12, Fedora 36 and newer, Arch) and updates itself. There is also a .deb (sudo apt install ./IraVoice-*.deb, which pulls in libgtk-layer-shell0 and libopenblas0) and an .rpm (sudo dnf install ./IraVoice-*.rpm); both install the program to /usr/bin/iravoice. There is no Flatpak, Snap, or AUR package.
Speech recognition runs on the computer with NVIDIA Parakeet once the default Parakeet Unified EN model, about 700 MB, has downloaded. Hold or tap Ctrl+Space to dictate. IraVoice types the text with a helper it chooses automatically: wtype on wlroots compositors such as Sway and Hyprland, kwtype on KDE Plasma Wayland, dotool or ydotool on any Wayland session including GNOME (both need permission to /dev/uinput), or xdotool on X11; you may need to install one. On Wayland, where the global shortcut may not reach the app, bind a desktop shortcut to iravoice --toggle-transcription.
The beta has real limits: Wayland support is limited, the cancel shortcut is not offered on Linux, and code-name spelling is not available because the beta cannot detect the focused app. Dictation itself works in any window, including a terminal running SSH. IraVoice Local costs $79 once for one computer, $109 for two, or $149 for three, after a 3-day free trial with no account; the license checks in weekly with api.iravoice.com and allows 30 days offline.
6. If your Linux lives inside WSL, a VM, or SSH
Many people searching for Wispr Flow on Linux sit at a Windows or Mac desktop and reach Linux through WSL, a virtual machine, or SSH. In that setup the dictation app runs on the host. Wispr’s terminal article lists WSL windows, Linux VMs, and remote SSH sessions as places where Flow transcribes but cannot paste directly: you press Paste last transcript (Alt+Shift+Z on Windows, Ctrl+Cmd+V on Mac) or paste by hand.
IraVoice on Windows and Mac types into the local terminal window, so text lands in WSL inside Windows Terminal, in tmux, or in an SSH session the same way it lands anywhere else, while the microphone and recognition stay on the local computer. Code-name spelling uses Git projects on the local computer only, not the remote host, and is not promised inside WSL.