Voice dictation on Linux used to be the awkward one. The polished commercial apps (Wispr Flow, superwhisper, Typeless) ship on macOS, Windows, and mobile, and skip Linux entirely. That gap has quietly closed. There are now good native-Linux options, free and open source, plus a couple of commercial ones. You just have to know which is which.
This is an honest map of what works on Linux in 2026. Full disclosure up front: Meander is our product. We've listed the alternatives fairly: strengths, weaknesses, and who each one is for. There's no single "best" here. It depends on what you're optimizing for. We'll go tool by tool, lay them side by side, and finish with an honest recommendation.
TL;DR
- Turnkey, AI cleanup + real translation: Meander. Native Linux app, free tier. (our product)
- Best free, fully-local, open source: Handy. Polished, MIT, runs entirely offline.
- Open source, local or cloud, with sync: OpenWhispr. MIT, the closest open-source take on a commercial workflow.
- Most complete fully-offline GUI: Speech Note (Flathub).
- Minimal / DIY: nerd-dictation, whisper.cpp.
What you still can't use on Linux
First, what's not on this list: the famous cloud apps people usually arrive from still don't run here.
- Wispr Flow: macOS, Windows, iOS, Android. No Linux. (our comparison)
- superwhisper: macOS, Windows, iOS. No Linux or Android. (our comparison)
- Typeless: macOS, Windows, iOS, Android, web. No Linux. (our comparison)
If you came from one of those, everything below runs on Linux.
1. Meander: turnkey, with AI cleanup and translation
Best for: people who want dictation that works in five minutes, with AI cleanup, translation, and the same tool on their phone, without managing models or API keys.
Meander ships native Linux builds (.deb and .AppImage) and treats Linux as a first-class platform. Hold a hotkey, speak, release. It transcribes, cleans up grammar and filler words, and pastes clean text wherever your cursor is. Terminal, editor, browser, anywhere you can type.
- ✅ Push-to-talk → transcribe → AI cleanup → paste at cursor, no API key to configure
- ✅ Translation mode: speak one language, output another. Tuned for Chinese, Japanese, and Korean
- ✅ Same account on Windows and as a full Android keyboard. Settings and history sync across all three
- ✅ Free tier to start
- ✅ Runs in the cloud: fast even on long recordings, and it doesn't tie up your CPU, GPU, or RAM (keep those for ComfyUI, builds, games, whatever else you run)
- ➖ Cloud-based: needs a connection. Audio is processed in real time, not stored
Trade-off: if your hard requirement is fully offline, Meander isn't it. Try Handy or Speech Note below. The flip side: those run on your machine and use its resources (more on that below).
2. Handy: the best free, fully-local option
Best for: people who want polished, private, offline dictation at zero cost, with no account.
Handy takes the same idea as the commercial tools (press a hotkey, speak, get text in any field) but runs everything on your own machine. It's well made (24k+ GitHub stars), open source (MIT), and ships native Linux builds. Transcription uses Whisper (with GPU acceleration) or Parakeet V3 (CPU, with automatic language detection).
- ✅ Free, open source (MIT), fully offline. Your audio never leaves the machine
- ✅ Native Linux, macOS, and Windows. Polished and actively developed
- ✅ Whisper + Parakeet engines: transcribes 99+ languages (speak any language, get text in that language)
- ➖ AI cleanup needs your own API key (OpenAI/Anthropic) or a local Ollama setup. The exception is macOS, which can use on-device Apple Intelligence
- ➖ Desktop only. No Android, no cross-device account or sync
- ➖ No built-in translation. You can translate into other languages through the LLM post-processing (your own key plus a custom prompt), but there's no turnkey translate mode like Meander's. The native Whisper translate only outputs English
Trade-off: if you want private and free, this is the one to try first. If you want AI cleanup with zero setup, an Android keyboard, or translation between languages, that's where the commercial apps come in.
Handy (github.com/cjpais/Handy)
3. OpenWhispr: open source, local or cloud, with sync
Best for: people who want an open-source app that can run fully local or in the cloud, with some of the convenience of a commercial tool.
OpenWhispr blurs the line between the offline tools and the paid ones. Run Whisper locally (Tiny → Turbo, 75 MB–1.6 GB) with no internet, or use cloud models with your own API key. It adds AI cleanup ("clean this up", "draft an email"), 100+ language transcription with auto-detection, and, on its paid tier, sync across your desktops.
- ✅ Free & open source (MIT). Unlimited local processing, fully offline
- ✅ Local or cloud, your choice. Native Linux, macOS, Windows
- ✅ AI cleanup and assistant-style commands. Cross-device sync on the paid tier
- ✅ Also does meeting transcription: auto-detects Zoom/Teams/FaceTime with on-device speaker diarization
- ➖ The free cloud tier is capped (~2,000 words/week). Local stays unlimited
- ➖ Desktop only. No Android
- ➖ Multi-language transcription, but no output-translation between languages
Trade-off: the closest open-source match to a commercial workflow. If you're happy bringing your own keys or managing local models, it's a strong pick.
4. Speech Note: the most complete fully-offline GUI
Best for: people who want free, offline, open-source dictation with a real GUI and lots of engine choice.
Speech Note is a free, open-source (MPL-2.0) Linux app that does speech-to-text, text-to-speech, and translation fully offline. Nothing leaves your machine. It supports multiple engines (whisper.cpp, Faster Whisper, Vosk, Coqui, and more) with downloadable models in 100+ languages, and it types decoded text into any focused window via global shortcuts.
- ✅ Free & open source, on Flathub (plus AUR / openSUSE)
- ✅ Fully offline. Privacy by architecture
- ✅ Multiple engines & models, 100+ languages, plus offline translation
- ➖ You download and manage the models yourself. Accuracy depends on which you pick
- ➖ Raw transcription. No AI cleanup of filler words or grammar
Trade-off: more knobs than a commercial app. That's a feature or a bug depending on who you are.
5. nerd-dictation: minimal and hackable
Best for: command-line users who want a tool they can read, script, and bend.
nerd-dictation is a single Python script (GPL-3.0) built on the offline VOSK engine. No daemon, no UI: you bind "begin" and "end" to hotkeys, and it types what you say. Text processing is a Python config file, so you can rewrite output however you want. Works on X11 and Wayland via input-simulation tools.
- ✅ Free & open source, completely offline
- ✅ Tiny, manual-activation design. Nothing runs in the background
- ✅ Output post-processing is your own Python, infinitely customizable
- ➖ Assembly required: VOSK models, audio utility,
xdotool/ydotool, hotkeys - ➖ VOSK accuracy trails modern Whisper-class models
6. whisper.cpp: the DIY route
Best for: people who want maximum offline accuracy and don't mind building the pipeline.
whisper.cpp isn't a dictation app. It's a fast C/C++ port of OpenAI's Whisper that many of the tools above build on. Linux users wire it into dictation setups with a mic-capture script and an input simulator. Accuracy is the best you can get offline. Everything else is on you.
- ✅ The offline accuracy ceiling, free & open source
- ➖ No app, no hotkeys, no UI. You build all of it
How they compare
Two things separate these tools. One is how much you manage yourself: a turnkey app, versus models and API keys you wire up. The other is where the work happens.
Fully-offline tools keep your audio on your machine, but your machine pays for it. Local speech-to-text, an LLM cleanup pass, and (if you want it read back) TTS all want RAM and GPU. An LLM plus a decent voice model can easily ask for 6 GB or more, and even good TTS alone wants a couple of gigs. While they run, they compete with everything else you'd rather point that GPU at: ComfyUI, model training, builds, games. And they're slower than the cloud unless you drop to tiny models, which costs you accuracy, noticeably so on longer recordings.
Cloud tools (like Meander) make the opposite trade: your machine stays free, long audio transcribes fast, and you get large high-accuracy models plus LLM cleanup without spending a byte of your own VRAM. The cost is needing a connection. Neither side is "better." Pick the one whose cost you'd rather pay. (And only Meander puts your phone in the loop too.)
| Meander | OpenWhispr | Handy | Speech Note | |
|---|---|---|---|---|
| Native Linux | ||||
| Runs fully offline (private) | cloud¹ | local mode | ||
| Keeps your CPU/GPU/RAM free | cloud | only in cloud mode | uses them | uses them |
| Fast on long recordings | depends on model/hardware | depends on model/hardware | depends on model/hardware | |
| AI cleanup, no setup | built-in | ⚠️ bring your own key² | ⚠️ bring your own key³ | raw text |
| Multilingual transcription (speak X → text X) | 100+ | 99+ | 100+ | |
| Translate into another language (speak X → output Y) | built-in, zero-setup, CJK-tuned | ⚠️ via your own LLM | ⚠️ via your own LLM⁵ | offline, separate step |
| One account across Linux + Windows | paid tier | per-machine | ||
| Android (full keyboard / IME) | ||||
| Settings & history sync across devices | incl. Android | paid, desktop | ||
| Bundled AI assistants | ||||
| Free | free tier | free tier⁴ | fully free | |
| Open source | MIT | MIT | MPL-2.0 |
¹ Meander streams audio for real-time processing; it doesn't store recordings. ² OpenWhispr local models run offline with no key; cloud models use your own API key. ³ Handy uses your own OpenAI/Anthropic key or a local Ollama setup. On macOS it can use on-device Apple Intelligence with no key. ⁴ OpenWhispr's free tier is unlimited locally; its cloud tier is capped (~2,000 words/week). ⁵ Handy can translate into other languages only through its LLM post-processing: your own API key (or local Ollama / Mac Apple Intelligence) plus a prompt you write. There's no built-in translate-to-target mode, and its native Whisper translate outputs English only. (Same goes for OpenWhispr's LLM step.)
Features, platforms, and pricing change. Check each project's own site for current details.
The honest recommendation. There's a real choice here, not a single winner. If privacy and zero cost are non-negotiable, and you're happy to spend your own CPU, GPU, and RAM on it, go offline. Handy is the best place to start, Speech Note if you want more engines and offline translation, or OpenWhispr if you want something between local and cloud.
But for most people, the simpler path is the other one. Running everything locally means your machine carries the load: RAM and GPU you'd probably rather use elsewhere, and slower results on anything longer than a sentence. Letting the cloud do the work skips all of that. You get the speed, your machine stays free, and there's nothing to set up. That's the bet Meander makes: fast even on long recordings, with AI cleanup and real translation built in, and the same account on Windows and Android. There's a free tier, so you can see if it fits before paying.
Related: The best local voice-to-text models in 2026 · Voice typing on Linux Wayland