Voibe Is Now on Windows — and We Built Our Own Cloud to Do It

For the past six months, the two most-requested Voibe features were the two we kept saying no to: a Windows app and cloud transcription. Not because they were hard to build — because we couldn't ship them the usual way without breaking the promise Voibe was built on: your voice is nobody's data.

TL;DR: Voibe is now live on Mac and Windows (July 2026). Voibe for Windows is a ground-up native app, not an Electron port. Both platforms can now use Voibe's new private cloud: open-source speech models running on servers we control, with zero retention — audio is transcribed, then deleted. No third-party AI provider ever receives a byte of your voice. On-device mode on Apple Silicon Macs is unchanged, and pricing is unchanged: $7.50/month, $59/year, or $149 lifetime, one license across both platforms.

This post covers why we said no for so long, what actually happens to your audio in a typical cloud dictation app, and the three rules our own cloud had to pass before we would ship it — so you can hold us to them.

Here is the whole launch in one table; the rest of the post walks through each row — including the part most launch posts skip, which is what happens to your audio.

We said no because the standard way to ship cloud dictation runs your voice through someone else's AI. Audio leaves your machine, hits a large AI provider's speech API, gets transcribed on that provider's servers, and then lives under a retention policy almost nobody reads. From that point on, the app vendor doesn't control what happens to your voice — the provider's policy does.

Voibe exists because we think that trade is wrong. Our users picked an on-device dictation app precisely because their audio never left the machine. Shipping a cloud that quietly reversed that would have been a bait-and-switch.

But the requests were fair, and they didn't stop. Cloud transcription is genuinely better UX for a lot of people: nothing to download, no models to manage, and it works on hardware that can't run large speech models locally. That last part is the Windows problem in one sentence — on-device Whisper is a guarantee we can make on Apple Silicon, where every Mac from M1 onward has the neural hardware to run Whisper locally. Across the full range of Windows machines, from gaming rigs to five-year-old office laptops, it is not a guarantee we could make honestly.

So the answer stayed no — until we could change the architecture instead of softening the promise.

Typical cloud dictation routes your audio through a third-party AI provider's API, and what happens to it next is governed by that provider's retention policy — not by the app you bought.

The standard pipeline looks like this:

1. You dictate. The app records your voice and sends the audio off your machine.
2. A Big Tech API transcribes it. Most dictation startups don't run their own speech models; they call APIs from large AI providers (OpenAI- and Meta-hosted models are common choices).
3. Text comes back. Usually fast, usually accurate.
4. Your audio is now on someone else's servers. Whether it is deleted immediately, retained for days or weeks of abuse monitoring, or used to improve models depends on policies you have never read — and those policies can change after you subscribe.

None of this is secret; it is disclosed in privacy policies. Our Wispr Flow safety review and the rest of our dictation privacy hub document the pattern tool by tool. The point is not that these companies are careless. The point is structural: once your audio sits on a third party's servers, "deleted" is a promise in a document. It is not something the app vendor — or you — can enforce.

The Zero-Retention Standard is the bar we set before we would ship cloud transcription: three rules, all architectural, none of them a settings toggle.

1. Open-source models only, on infrastructure we control. Voibe's private cloud runs open-source speech models on our own servers. No third-party AI provider is in the audio path — not for transcription, not for cleanup. (Voibe's on-device mode has always run OpenAI's open-source Whisper models locally; the cloud keeps the same open-source principle, on our hardware instead of yours.)
2. Transcribe, then destroy. Your audio exists on our servers only for the seconds transcription takes. The moment your text comes back, the audio is deleted. There is no audio archive to breach, leak, or hand over — you cannot lose what you never kept.
3. Never monetized. Audio and transcripts are never stored, never sold, and never used to train any model — ours or anyone else's.

That architecture is what finally unlocked both launches. It is also why they took longer than they could have: we spent the time building and operating our own inference stack instead of wiring up someone else's API. Privacy by design was the constraint. It turned out to be the unlock.

Voibe for Windows is a ground-up native Windows app — not an Electron wrapper, and not a port of the Mac app. It connects exclusively to the private zero-retention cloud described above, and it does the same job Voibe has always done: press your hotkey, talk, and accurate text lands wherever your cursor is.

• Works in any app — email, docs, Slack, IDEs, browsers. Anywhere you can type, you can dictate.
• Smart Formatting — a bounded cleanup pass that removes filler words and fixes punctuation, capitalization, and numbers. It formats; it never paraphrases or invents content. Off by default.
• Custom Dictionary and Memory shortcuts — teach Voibe your names, jargon, and frequently used snippets.
• Up to 5x faster than typing, per our own user studies.
• 100+ languages, with the 97%+ accuracy Voibe is known for, including technical vocabulary.

The honest catch: there is no offline mode on Windows today. Voibe for Windows is private-cloud only, so it needs an internet connection. If your requirement is strictly offline dictation on a PC, our Windows dictation roundup covers those options honestly — including the free, built-in Windows Voice Access.

Try Voibe for Windows free for 7 days → No card required.

Nothing changes for Mac users who chose Voibe because audio never leaves the machine. On Apple Silicon Macs (M1 through M4), on-device mode still runs Whisper models locally, fully offline, with sub-300ms latency and zero data leaving your Mac. It remains the strictest privacy option we offer, and nothing about this launch moves your dictation to the cloud without you choosing it.

What Mac users gain is choice. Intel Macs (macOS 13+), which cannot run Voibe's on-device models, now have a first-class mode instead of a hardware apology. And Apple Silicon users who prefer not to run models locally can switch modes inside the app — both modes carry the same promise: never stored, never sold, never trained on. New to Voibe? Start with our getting-started guide.

Choose your Voibe mode by hardware and by how strict your privacy requirements are:

• Apple Silicon Mac, strictest privacy or offline work (flights, secure environments, client files): use on-device mode. Audio never leaves the Mac, and no internet is required.
• Apple Silicon Mac, prefer zero setup: either mode works — the private cloud carries the same never-stored, never-trained promise, with nothing to download.
• Intel Mac (macOS 13+): use private cloud mode — on-device mode requires Apple Silicon.
• Windows PC: private cloud is the mode. If you need strictly offline dictation on Windows, Voibe isn't your tool yet.
• Regulated or confidential work (medical, legal): on-device mode on a Mac is the strongest guarantee. Zero retention makes Voibe's cloud unusually clean for a cloud tool, but verify your own compliance obligations first — our HIPAA dictation guide explains what to check.

What this launch means depends on where you're starting from:

• If you're on Windows: this is the first time Voibe's privacy model has been available off the Mac. In our own studies, people dictate up to 5x faster than they type — and independent research points the same direction: Stanford measured speech input at about 3x typing speed (161 versus 53 words per minute) with a 20.4% lower error rate. With Voibe you get that speed in any app, without your voice becoming a Big Tech data point.
• If you're on an Apple Silicon Mac: nothing is taken away. On-device mode is untouched; you now have a second mode for the moments local models don't fit.
• If you're on an Intel Mac: Voibe is now genuinely available to you — private cloud mode was built for exactly your hardware.
• If you work across both platforms: one license covers Mac and Windows, with the same Dictionary, Memory shortcuts, and Smart Formatting on each.

Voibe holds a 4.8/5 rating on Product Hunt. But the claim we most want to be held to is the architectural one: the question to ask any dictation vendor is no longer "cloud or local?" It is "who sees my audio, and how long do they keep it?" Voibe's answer, in either mode: nobody else, and zero seconds after transcription.

The launch

Is Voibe available on Windows? Yes. As of July 2026, Voibe runs on Windows as a ground-up native app with system-wide dictation, Smart Formatting, Memory shortcuts, and a custom Dictionary, powered by Voibe's private zero-retention cloud.

Why did a Windows app take this long? Because Voibe's on-device guarantee depends on Apple Silicon, a Windows app required cloud transcription — and we refused to ship a cloud that routed audio through third-party AI providers. We built our own inference stack first, then shipped Windows.

Privacy and your data

What happens to my audio in cloud mode? Audio is encrypted in transit, transcribed by open-source models on Voibe's own servers, and deleted the moment your text is returned. It is never stored, never sold, and never used to train AI.

Does any third-party AI company process Voibe audio? No. In private cloud mode, audio touches only Voibe-controlled infrastructure running open-source models. In on-device mode on Apple Silicon Macs, audio never leaves the Mac at all.

Is cloud mode private enough for confidential work? Zero retention removes the biggest cloud risk — there is no stored audio to breach or hand over. For regulated fields, on-device mode remains the strictest option, and you should verify your own compliance obligations; our HIPAA guide covers the questions to ask.

Platforms and modes

Does Voibe for Windows work offline? No. Voibe for Windows is private-cloud only and needs an internet connection. Offline, on-device dictation is available on Apple Silicon Macs.

Can Intel Macs use Voibe? Yes. Intel Macs on macOS 13 or later can use private cloud mode. On-device mode requires an Apple Silicon Mac (M1 or later).

Pricing

How much does Voibe cost on Windows? The same as on Mac: $7.50/month, $59/year, or $149 lifetime, with a 7-day free trial and no card required. One license covers both platforms.

Did prices change with this launch? No. The Windows app and cloud transcription were added to the existing plans at no extra cost.

The bottom line on this launch: Voibe now runs on Mac and Windows, and both platforms can use a private cloud that keeps nothing. We took six months longer than the easy path because the easy path — someone else's API, someone else's retention policy — would have turned our privacy promise into a footnote. Instead, it is the architecture.

If the privacy model matters to you beyond any one product, the deeper explainers are here: why offline dictation matters and what your voice data reveals about you.

And if you were one of the people asking for these two features: thank you for pushing, and for waiting. Try Voibe free for 7 days on Mac or Windows →