Speak instead of
typing.

Yes. The complete app is available from GitHub under the MIT License at no charge. The paid Store versions help fund development and receive updates through the Store. Apart from that, the app is identical to the GitHub download.

Yes. The built-in speech engines run on your computer, so local transcription does not need an internet connection. Connecting a cloud provider is optional.

Accuracy depends on the model you choose. The compact model works well for everyday voice input. Larger models usually produce cleaner transcripts but need more RAM and processing time. You can switch models at any time.

You need a 64-bit processor, 8 GB of RAM, 1–2 GB of free disk space and one of these systems: Windows 10, macOS 11 Big Sur or later on Apple Silicon, or Ubuntu Linux. A GPU is optional; WhisPaste can use the CPU, and Parakeet is designed for it. For faster Whisper transcription, we recommend 16 GB of RAM and a dedicated GPU with 2–4 GB of VRAM (Vulkan on Windows and Linux, Metal on Apple Silicon). Approximate VRAM use is 900 MB for compact, 1.5 GB for balanced and 2.6 GB for premium (turbo). On Apple Silicon, unified memory is shared by the CPU and GPU.

Whisper supports 99 languages, including English, German, Japanese, Arabic and Hindi. Parakeet supports about 25 and transcribes much faster on a CPU. You can choose a language and engine in Settings or let WhisPaste detect the language automatically.

Whisper (whisper.cpp) supports 99 languages and can use a GPU. Parakeet (NVIDIA Parakeet TDT via sherpa-onnx) supports about 25 languages but is several times faster than Whisper on the same CPU. Both work offline. Choose Whisper for language coverage and maximum accuracy, or Parakeet for fast results without a dedicated GPU. You can switch at any time in Settings → Speech-to-Text.

Voice Snippets are reusable text blocks. After speaking, say your trigger word — or press the snippet hotkey at any time — and a searchable picker opens right at your cursor. Interactive snippets take this further: templates with fields that WhisPaste walks through with you, one field at a time, then pastes the finished text in one piece. There is also a clipboard side panel: hover the edge of your screen to open your recent clipboard history and drag any item straight into your document.

Yes, with Smart Mode: an optional post-processing step with three presets. Cleanup removes filler words and fixes punctuation, Concise shortens the text without losing facts, and Translate converts it into one of seven validated target languages (German, English, Spanish, French, Portuguese, Mandarin, Russian). It runs on a local AI model on macOS and Windows, so the text never leaves your device. Attach a preset to your main hotkey, or give the dedicated Smart-Mode hotkey its own preset and target language — presets can also be applied to history entries afterwards.

With a local engine, the recording stays on your computer. If you select a cloud provider, WhisPaste sends it directly from your device to that provider and does not store it. Anonymous usage statistics and crash reports never include audio or text, and you can turn them off in Settings → Privacy.

Yes. WhisPaste provides ready-to-run downloads for Windows, macOS and Linux. Linux packages are available as AppImage and .deb files, and you can also build the app from source. You will find every option on the Download page.

Microsoft SmartScreen warns about new software that hasn't built a download reputation yet. The warning does not mean the app is unsafe. The WhisPaste source code and release workflow are public on GitHub. Click “More info” → “Run anyway” to proceed.

macOS Gatekeeper blocks unrecognised apps downloaded outside the App Store. This only applies to the free GitHub DMG, which is not yet notarized by Apple. The Mac App Store version is notarized and opens without a warning. For the DMG, right-click the app → “Open” and confirm in the dialog.