1. VoiceInk’s three draws, and why none of them reaches Windows
VoiceInk is a native macOS dictation app. Its site highlights local AI models that keep your voice on the Mac, open source code, and a buy-once, own-forever price. It requires an Apple-silicon Mac with macOS 14.4 or later (the GitHub README now lists macOS 15.0), recommends 8 GB of RAM, and has added an iOS app. Its GitHub releases contain a single Mac disk image, VoiceInk.dmg, and nothing for Windows.
Beyond plain dictation, VoiceInk has modes that switch settings by app or website, context awareness that reads what is on screen, a personal dictionary for custom words and replacements, keyboard or mouse shortcuts with push-to-talk, and an AI assistant mode. Transcription is local by default. Cloud transcription is optional and uses your own API key with providers such as Groq, OpenAI, Deepgram, and ElevenLabs, and the optional Cloud Enhancement sends the transcribed text, not your voice.
So a Windows replacement should be judged on the same three points, local models, open code, and a one-time price, plus whichever extras you actually used.
2. VoiceInk’s license split: GPLv3 code, paid compiled app
VoiceInk’s terms describe two licenses. The source code is under the GNU GPL v3.0, so you may read it, modify it, and distribute modified versions under the same license. The pre-compiled app from tryvoiceink.com is sold under a separate commercial license, a one-time purchase. The README says experienced developers can build it themselves, but the compiled version adds automatic updates and priority support; the project is not accepting pull requests.
Prices on October 3, 2026: $25 for one Mac, $39 for up to two, and $49 for up to three, each a one-time purchase with lifetime updates and a 14-day money-back guarantee. Devices on the larger plans are managed through a Polar dashboard.
For a Windows user the open code does not solve the problem. It is the source of a native macOS application, and VoiceInk publishes no Windows build. The useful question is which Windows apps follow the same pattern: local first, inspectable, and paid once or not at all.
3. Handy: free, MIT-licensed, and the project IraVoice builds on
Disclosure first: IraVoice is built on Handy, so we have a stake in this recommendation. Handy is still a direct answer to “open-source VoiceInk for Windows”: a free, MIT-licensed app for Windows, macOS, and Linux whose stated aim is one tool with one job, transcribing what you say into a text box on your own computer without sending audio to the cloud.
You hold a configurable shortcut to record and release to stop, or tap to toggle, with hold-only and toggle-only modes too. Silero voice-activity detection filters silence, and you choose Whisper Small, Medium, Turbo, or Large, with GPU acceleration when available, or Parakeet V3, which Handy describes as CPU-optimized with automatic language detection. Its October 3, 2026 release, 0.9.8, includes Windows installers for x64 and ARM64 in both .exe and .msi form, and winget can install it as cjpais.Handy, a package the Handy developers note they do not maintain.
Because Handy keeps its scope narrow on purpose, compare it feature by feature if VoiceInk’s per-app modes, screen context, or assistant were the reason you bought it.
4. VoiceTypr and OpenWhispr: open code with a paid layer
VoiceTypr is the closest Windows match to VoiceInk’s business model: source published under AGPL-3.0 and a lifetime license for the ready-made app. It is built with Tauri, Rust, and React for macOS and Windows; on Windows it needs 64-bit Windows 10 build 19041 or later, or Windows 11, and installs from its site or the Microsoft Store. Local transcription uses Whisper on Windows (its Parakeet models are Apple-silicon only), with an optional Vulkan path that falls back to the CPU. Optional cloud speech engines include Soniox, OpenAI, Groq, Deepgram, and Cohere, and AI formatting can use OpenAI, Anthropic, or Gemini. Its site shows $69 once for two devices, with one- and four-device options, and its product data lists lifetime licenses at $39, $69, and $99. A 3-day trial needs no card, purchases carry a 7-day money-back guarantee, and SmartScreen shows “Unknown Publisher” on the installer because, as VoiceTypr explains, its developer does not yet have a Microsoft code-signing certificate.
OpenWhispr is MIT-licensed and runs on macOS, Windows, and Linux with local Parakeet and Whisper models plus cloud models through your own API keys. Local dictation is unlimited and free on every plan. The Free plan also includes 2,000 words a week of OpenWhispr’s own cloud transcription; Pro, at $80 per user a year billed annually, makes cloud transcription unlimited and adds device sync and an iPhone and iPad app. So the open-source local part is free, and the subscription pays mainly for cloud and sync features.
Both apps, like VoiceInk, can send audio or text to a cloud provider when you pick one, so the privacy boundary is set by the engine each setting uses.
5. Local-first Windows apps for when open source is optional
If the source code matters less than local models and price, three more Windows apps fit. Superwhisper runs natively on Windows 10 and 11 (x64 and ARM64); its Free plan gives unlimited dictation with local Whisper models and up to two modes without AI processing, and Pro, which adds AI modes, context awareness, and cloud models, costs $84.99 a year or $249.99 once for Lifetime. Its modes can switch on automatically by app or website, the nearest thing on Windows to VoiceInk’s per-app modes.
Spokenly for Windows installs from the Microsoft Store on 64-bit Windows 10 and 11, offers local Whisper and Parakeet models and bring-your-own-key cloud engines free with no account, and requires a subscription for Pro features that use Spokenly’s own cloud transcription; Spokenly says its Mac and iOS versions are the most mature. SpeechPulse sells a one-time Windows license for $99 plus tax, with future updates and up to six computers for one person, and runs Whisper offline in 99 languages after a 30-day trial.
6. IraVoice for VoiceInk users who also work on a PC
IraVoice shares VoiceInk’s local-first, pay-once shape without its modes or AI features. On Windows 10 or 11, 64-bit (x64) you hold Right Ctrl, speak, and release, or tap for hands-free. NVIDIA Parakeet runs on the PC: the default Parakeet Unified EN model, about 700 MB, downloads once, and dictation then works offline with audio and text kept on the computer. IraVoice still connects for that download and for license check-ins and update checks to api.iravoice.com, which carry a salted hash of a hardware ID; there is no account and no product telemetry.
The VoiceInk habits that carry over are custom words, replacements and snippets, and corrections: press Ctrl+Alt+Shift+F after a misheard word, fix it, and IraVoice learns that word pair for next time, keeping only the pair. In Windows Terminal, PowerShell, VS Code, Cursor, Windsurf, Zed, Rider, and other developer tools it also spells names from your Git project, so “use effect” becomes useEffect. What does not carry over: per-app modes, screen context, an assistant, user-chosen cloud providers, and Windows on ARM, since IraVoice’s stated Windows requirement is x64.
IraVoice also runs on macOS 13 or later on Apple silicon and, in beta, on Linux. IraVoice Local costs $79 once for one computer, $109 for two, or $149 for three, after a 3-day free trial that needs no email, card, or account. If open source is the requirement, Handy is the direct choice.