My VoiceInk review after five and a half months of use: VoiceInk* is a dictation app for the Mac that turns speech into text at a keyboard shortcut and pastes it into whatever text field your cursor is in. On my machine, transcription runs on a local model. Since 30 April 2026 I've spoken 2,794 dictations with 135,388 words, and on my Mac the text is there after a median of 0.71 seconds of processing. According to the pricing page, the app costs a one-time $25 for one Mac, $39 for two and $49 for three (as of October 2026), with no subscription. My main argument is privacy: with a local model, the recording doesn't leave the Mac, as long as cloud models and AI enhancement through cloud providers are switched off. The limits: VoiceInk is Mac-only, plus an iPhone app in beta, and the vendor states the system requirements differently in two places.
I use it, among other things, to dictate tasks to Claude and to my AI employees, on 137 of the 168 days since I installed it. My reasons, in my own words from October: "Local, local model, super fast, a one-time price, and the updates are great." If you're looking for a lab test: this isn't one, and it isn't a comparison with other apps. It's what VoiceInk's own statistics show on my MacBook Pro. I bought the app myself; my only tie to the developer is the affiliate link.
* Affiliate link: if you buy through this link, I earn a commission. Your price stays the same.
My VoiceInk review in numbers
All values come from VoiceInk's local statistics on my Mac, read in October 2026. According to the docs, the dashboard keeps its own session metrics for this, without the dictated text.
That works out to about 48 words per dictation and roughly 5,600 words per week on average. Usage grew over the months, according to the dashboard (snapshot 30 September 2026): April 131 words (from the 30th), May 19,089, June 14,854, July 23,186, August 27,915, September 39,477.
One number I leave out on purpose: the estimated time saved. According to the docs, it assumes a typing speed of 40 words per minute, which is an assumption, not a measurement of my typing.
How fast the text shows up on my Mac
The median processing time on my machine is 0.71 seconds, and 90 percent of dictations finish in under 1.5 seconds. The dashboard average is 0.84 seconds and the "Speed Factor" is around 27, meaning the audio is processed at roughly 27 times real-time speed. That's all I mean by "super fast".
These numbers apply to my setup: a MacBook Pro with an Apple M3 Max and 64 GB of memory, Whisper Large v3 Turbo in its quantized version, language fixed to German, recordings of mostly 15 seconds. A different Mac, a different model or long recordings will look different. VoiceInk's own docs say that speed and accuracy depend on your microphone, language, recording and hardware.
Local model means the recording stays on the Mac
For me, this is the main reason for VoiceInk: with a local model, the recording doesn't leave the Mac, and there's no external service provider for the transcription, so no data processing agreement and no transfer to a third country for you to check. For a business in the EU, that makes using it considerably simpler under data protection law, but only as long as cloud models and AI enhancement through cloud providers are switched off. How that looks with a cloud service is covered in VoiceInk vs Wispr Flow: Local or Cloud.
VoiceInk separates three settings, and the docs advise checking all three:
- Transcription: a local model, or a cloud provider with your own API key. With cloud, the audio goes to the selected service.
- AI enhancement: sends text, vocabulary and any enabled context to the selected provider. Locally, this works with VoiceInk Refine or a local Ollama.
- Dictionary Auto Learn: sends changed text plus a little surrounding text to its own review provider.
I don't use enhancement. The raw text goes straight to Claude or my AI employees, and they cope fine with filler words. What happens to the text after that depends on the tool you dictate into: a prompt to Claude goes to Anthropic as text, however it was produced. How I handle that is in Dictating Prompts to Claude and ChatGPT.
Two things stay your job even locally: the history with transcripts and audio is stored on the Mac, and you set retention in the history settings. And no tool settles your legal basis or access control for you. I'm a developer, not a lawyer, and this isn't legal advice. How I weigh local against cloud in general is in Local AI and Privacy: Cloud or Server?.
VoiceInk pricing: pay once instead of a subscription
According to the vendor, that includes "Lifetime of updates", a 14-day money-back guarantee and a 7-day trial during onboarding. If you connect cloud providers inside VoiceInk, you pay for them separately: "Your VoiceInk license does not include provider charges or credits."
For comparison: Wispr Flow Pro costs $144 a year billed annually, but it also has a free tier with 2,000 words per week on desktop, runs on Windows, iPhone and Android, and comes with team features. At my roughly 5,600 words per week, I'd be above the free limit. The three-year calculation is in VoiceInk vs Wispr Flow: Local or Cloud, and more apps without a subscription, with their prices, in Wispr Flow Alternatives Without Subscription.
VoiceInk setup, step by step
These steps follow VoiceInk's own docs:
- Install: download from tryvoiceink.com, move it to Applications, launch it and follow onboarding. If macOS blocks the first launch, approve the app under System Settings → Privacy & Security.
- Permissions: Microphone for recording, Accessibility for pasting the text. Screen Recording is only needed for the optional screen context.
- Download a model: under AI Models → Local, then select it in your Mode under Transcription → Model. Onboarding starts with Parakeet V3.
- Set the language: in the Mode under Language. The docs advise checking the language picker before downloading, to see whether the model supports your language.
- Shortcut: under Settings → Shortcuts, choose Toggle (press once to start, once to stop), Push to Talk (hold), Hybrid or Double Tap. Fn or a middle or extra mouse button works too.
- Dictionary: add names and technical terms under Dictionary → Vocabulary, and fixed substitutions under Word Replacements, in the docs' example "voice ink" to "VoiceInk".
- Check the Mode: the starter Mode "Dictation" pastes without AI enhancement and is the default. You can bind further Modes to apps or websites.
- License: start the trial, or later paste your key under VoiceInk Pro. Activation needs a network connection, according to the docs.
My setup is one of many: Whisper Large v3 Turbo (quantized), about 547 MB on disk, language fixed to German, no enhancement. According to the docs, Whisper processes after you stop and shows no text while you speak. If you want to watch the text appear, you need a model with real-time transcription.
Choosing a local model and language
For German, the docs list several local models: Parakeet V3 for "English and supported European languages", Nemotron Latin with German, Cohere Transcribe with German, and the multilingual Whisper models. For English-only dictation, the docs point to Parakeet V2 or Parakeet Unified. On Intel Macs, the app recommends cloud transcription because local models don't run reliably there.
My language is fixed to German. For names, product names and technical terms there's the dictionary: according to the docs, Vocabulary helps the model recognise a term, while Word Replacements always swap a known mishearing for the same text. My dictations for this very article show why that matters: I got "Voice Inc." instead of VoiceInk, "Schoolstats" instead of SkoolStats and "Whisperflow" instead of Wispr Flow. Claude still understood what I meant. For text that's meant to stay as written, names like these belong in the dictionary. If you switch between languages a lot, check the automatic detection some models use, according to the docs.
Updates and open source
In the six months to October 2026, the developer published eleven stable versions (1.74 to 2.21) and four pre-releases on GitHub, with 641 commits. Version 2.0 (16 July) introduced Modes and a German interface, 2.11 the local enhancement VoiceInk Refine, 2.20 the Auto Learn dictionary, and 2.21 (28 September) double-tap shortcuts. I'm on 2.21.
The source code is on GitHub under GPL-3.0. According to the README, the compiled app adds automatic updates and support over a self-built one. "Free because it's open source" therefore only holds if you compile it yourself. And "Lifetime of updates" is a vendor statement: the release rate so far is documented, but what a solo developer ships in years to come, nobody can promise.
Limits worth knowing
- Mac only: there's no VoiceInk for Windows. The iPhone app is, according to the vendor, "Free during beta. One-time purchase after.", and per the docs a Mac license doesn't cover iPhones.
- Conflicting requirements: the pricing page and home page say "Apple Silicon Macs" and macOS 14.4, the installation docs and README say macOS 15. Check your version before buying and use the trial.
- Memory: the FAQ recommends 8 GB, VoiceInk Refine needs at least 16 GB.
- Stability: on 1 October 2026 a diagnostic pre-release came out to investigate failures and hangs in Parakeet transcription. I don't use Parakeet, so I can't speak to that from experience.
- Local isn't automatic: enhancement, Auto Learn and cloud models send data out when you switch them on.
What I use VoiceInk for
Among other things, I dictate tasks to Claude and to my AI employees, including the tasks for this article. The raw text goes out without enhancement. How to dictate prompts so Claude can work with them is in Dictating Prompts to Claude and ChatGPT.
That Whisper Large v3 Turbo runs smoothly on my MacBook also has to do with the hardware. What Apple Silicon means for local models, and what my M3 Max manages with a large language model, is in Local AI on a Mac: Mini as AI Server.
Frequently asked questions
Is VoiceInk free?
Not as a ready-made app. According to the pricing page, VoiceInk costs a one-time $25 for one Mac (as of October 2026), and you can try it for seven days first. The source code is on GitHub under GPL-3.0; if you compile it yourself you pay nothing, but according to the README you give up automatic updates and support. For most people, the trial is the sensible start.
Does VoiceInk work with languages other than English?
Yes. Several local models cover other languages: Parakeet V3 handles English and supported European languages, Nemotron Latin and Cohere Transcribe include German, and Whisper is multilingual. The interface has been available in German since version 2.0. I dictate in German with Whisper Large v3 Turbo. Check your Mode's language picker before downloading a model.
Does VoiceInk work offline?
Transcription does, once a local model is downloaded. Offline doesn't automatically cover everything, though: AI enhancement, Auto Learn and cloud transcription can send data to providers if you switch them on. According to the docs, license activation needs a network connection. For a workflow that stays entirely on the Mac, check all three settings.
Is there VoiceInk for Windows or iPhone?
Not for Windows. For iPhone and iPad there's a separate app, free during the beta according to the vendor and a one-time purchase after that. The Mac license doesn't cover the iPhone. If you need Windows, my article on Wispr Flow alternatives without a subscription lists options.
Is there a VoiceInk discount code?
Not through my link; your price stays the same. The "50% Off" item in the menu of VoiceInk's website leads to the Creator Bonus programme: if you post publicly about VoiceInk, the vendor refunds part of the price from 1,000 views and the full price from 10,000. For students, the FAQ mentions a discount on request to the developer.
Where to go from here
Start the 7-day trial, download a model for your language and dictate everything for a week that you'd otherwise type to Claude or into emails. Then look at the dashboard to see how many words came together, and decide. You can buy through VoiceInk*.
If you want to see how I hand dictated tasks to my AI employees, my community Claude Practitioners is German-speaking, and that's where you can ask questions about setup and ways of working: Community.
* Affiliate link: if you buy through this link, I earn a commission. Your price stays the same.
