How Hyperfurion VK compares.
Hyperfurion VK is an MIT-licensed voice keyboard for Linux and Windows. It types what you say into whatever app has focus, a terminal included; in a terminal it recognizes, no spoken phrase presses Enter. It ships no speech model: you bring a cloud key (xAI by default) or your own speech server. Here it is next to Talon, Numen, nerd-dictation and Dragon, including the rows where they win.
01At a glance
| Feature | Hyperfurion VK | Talon | Numen | nerd-dictation | Dragon Professional |
|---|---|---|---|---|---|
| Licence and price | MIT, free (a cloud speech provider bills you for what you use) | closed source; free, and Talon+ is paid (Patreon) | AGPL-3.0, free | GPL-2.0 or later, free | commercial, paid |
| Needs no API key or server | no: needs a key (xAI by default; OpenAI, Groq, Deepgram, AssemblyAI) or your own speech server | yes: free speech model | yes: a free Vosk model (about 40 MB) from its install script | yes: a free Vosk model you download once | yes: model included |
| Works fully offline | only with your own OpenAI-compatible speech server (Speaches, for one); rewrites and Kai need a local language model too (llama-server, for one) | yes | yes | yes | yes (the desktop edition) |
| Linux on Wayland | yes, and X11 (types through the kernel’s uinput) | no: X11 only, and it warns on Wayland. Talon 1.0 is planned as the last public Linux release; ongoing Linux support is a paid Talon+ feature | yes, through dotool (also X11 and the text console) | with a flag (dotool, ydotool or wtype); X11’s xdotool by default | no Linux version |
| Windows and macOS | Windows 10/11 tray app; macOS in beta, installed from a source checkout | Windows and macOS | no: Linux only | no: Linux only | Windows 10/11; no macOS |
| Types as you speak | with xAI (the default), or every 2.5 s with your own server at a local address (localhost, a private IPv4 address or a .local name); a word can still change for at least 1.5 s, then stays put. Other providers type when you stop | new in Talon 1.0: “Hum” streaming dictation (GPU-accelerated) | no: each phrase is typed after you pause | yes, by default, retyping words as Vosk revises them | shows a preview, types when you pause |
| Never presses Enter in a terminal | never in a terminal it recognizes, not even on “new line”; a terminal it doesn’t know (PuTTY, an editor’s terminal panel) gets Enter | presses it on “press enter”, “slap” or “paste enter” (you can remove them) | presses Return on “yes” (you can remove it) | never, by design | presses it on “new line” and “press Enter” |
| Spoken corrections | built in: “scratch that”, “correct X to Y”, “cap that”, “spell that” | in the standard community command set: “scratch that”, “select that”, “cap that”, “spell that” | “ditch” erases the last phrase, “change” picks another reading | none (a Python hook can rewrite text) | built in: “correct …”, “select …”, “scratch that”, “cap that” |
| Cursor commands by voice | in editors and terminals; off until you turn it on (setup asks) | extensive | yes: short words for keys and arrows | none | yes |
| Custom commands and scripting | rename or remove phrases, add words and snippets in config; no scripting language | extensive: .talon files and Python | phrase files that press keys and run shell commands | a Python hook that rewrites the text | custom commands and macros |
| Hands-free mouse | none | eye tracking; pop to click, hiss to scroll | noises (blow, hiss) and click and scroll words | none | MouseGrid |
| Rewrite or restyle text with a language model | “VK, make that formal” rewrites what you just said; a style per app is opt-in. Both use your language model (xAI by default, or your own) | not built in | not built in | not built in | not built in |
| Voice assistant | Kai, push-to-talk on Right Ctrl. On by default with your own speech server and language model; with xAI (the default) or another online service, off until you turn it on | none | none | none | none |
| Setup | one installer; setup asks for a speech provider and key, or your server’s address | download and run; most people add the community command set | a distro package (Alpine, AUR, Nix) or build from source | by hand: install Vosk, download a model, run the script | Windows installer |
Checked October 2026 against each project’s own documentation and code. Not a benchmark: nothing here measures recognition accuracy. Hyperfurion VK cells describe v2.4.0 with default settings. Dragon means the desktop edition, Dragon Professional v16; Nuance’s “Anywhere” editions are separate cloud products. Spot something out of date? Open an issue.
02Also worth a look
The table covers four tools, not the whole field. If you want a speech model in the box, or you’re on a Mac or a phone, look at these too.
- Handy: free and open source (MIT) for Windows, macOS and Linux. Runs a Whisper or Parakeet model on your computer, fully offline, and pastes the text when you stop recording. Its Wayland support is limited and needs a helper such as wtype or dotool.
- Windows Voice Access: built into Windows 11 (22H2 and later) and works without an internet connection, with “correct that” and “spell that”. Windows voice typing (Win+H) is built in too, but uses Microsoft’s online speech service.
- Wispr Flow: for Mac, Windows, iPhone and Android, with a writing style per app. Transcription always happens in Wispr’s cloud. Free up to 2,000 words a week on a computer and 1,000 on a phone, then paid.
- Superwhisper: for macOS, Windows, iOS and Android, with offline or cloud models and custom modes that set how it writes. Free and paid plans.
- VoiceInk: open source (GPL-3.0) for macOS 15 and later, with offline models and modes that switch by app or website. The prebuilt app is paid, with a free trial.
03Where each one shines
Choose Hyperfurion VK
You’re on Linux (Wayland or X11) or Windows and want to type by voice in any app, a terminal included, and correct by voice (“correct monday to friday”, “scratch that”). It only acts when you ask: it listens only after you press a hotkey or, once Kai is on, its on-screen button (Kai’s wake word is off until you turn it on), never presses Enter in a terminal it recognizes, and in the chat apps and chat sites it recognizes a spoken line break is Shift+Enter, so the message doesn’t send. You’ll need a speech provider key or your own speech server.
Choose Talon
You want the deepest hands-free control: coding by voice, a scripting language, eye tracking and noise commands, with a shared community command set. It runs on macOS, Windows and Linux under X11, and Talon 1.0 is planned as its last public Linux release.
Choose Numen
You want a small, fully offline voice-control layer for Linux that works on Wayland, X11 and the text console, driven by short spoken words for keys, with a free Vosk model. Dictation is a phrase at a time.
Choose nerd-dictation or Dragon
nerd-dictation for a single-file, offline Python script on Linux that types as you speak and that you shape in Python (Vosk’s output is all lower-case). Dragon if you’re on Windows and want a paid dictation suite with its model in the box, voice correction, custom commands and MouseGrid.
04The short version
If you’re on Linux or Windows, want to type by voice in every app, terminal included, and can bring a speech key or run your own speech server, Hyperfurion VK is built for that. For a speech model in the box, try Handy, Numen or nerd-dictation; for deep hands-free control, Talon; on a Mac or a phone, the apps above.