Skip to content

WhisPaste turns Whisper into a desktop app: speak, get text instantly.

WhisPaste is an open-source desktop app, free on GitHub, that turns your voice into text anywhere on your computer. Press a hotkey, speak, and the words appear right where your cursor is — no typing, no account needed, no cloud unless you choose one.

Get WhisPaste, open source and free on GitHub

Works offline

With the local engines, audio and text stay on your machine — no cloud, no account.

No Python setup

One installer, whisper.cpp bundled in — pick a model, done.

Runs without a GPU

CPU-only works out of the box; GPU (Vulkan/Metal) makes it faster.

How to use Whisper as a desktop tool with WhisPaste

One hotkey, your voice, text at the cursor — no app switch in between.

  1. Install WhisPaste and pick a Whisper model

    Download WhisPaste from the Microsoft Store, the Mac App Store or GitHub. On first launch the setup assistant recommends an engine and a quality level based on your language and graphics memory (on Apple Silicon: unified memory). You can change it any time.

  2. Let the model download once

    WhisPaste fetches the chosen whisper.cpp model file in the background and verifies it. The download resumes after interrupted connections, so a flaky network does not force you to start over. After that, transcription runs fully offline.

  3. Press your hotkey and speak

    Place the cursor wherever you want the text, hold your configured hotkey, and speak. whisper.cpp transcribes the audio on your machine and the transcript appears at the cursor: no upload, no browser tab, no copy-paste step.

WhisPaste in everyday use: dictated notes turn into text instantly, right on your own device.

Technical details

Setup: one installer, no Python environment

WhisPaste ships whisper.cpp (the C++ port of OpenAI Whisper) as a bundled library inside the download. After install, you choose a quality level from the cards (Quick & Compact, Balanced, Best Quality), and the app downloads it once, and from then on transcription runs fully offline on your machine. No virtualenv, no pip, no CUDA toolkit — the GPU acceleration ships inside the download (Vulkan on Windows and Linux, Metal on Apple Silicon). Recent NVIDIA cards can add an optional CUDA pack from the settings.

Performance: the CPU is enough, a GPU makes it faster

whisper.cpp runs on the plain CPU on every supported machine (Windows 10+, macOS 12 Monterey+ on Apple Silicon, Linux), so an 8 GB laptop without a dedicated GPU is enough for the compact model. With GPU acceleration (Vulkan on Windows and Linux, Metal on Apple Silicon) the same transcript comes back noticeably faster: in our benchmark on Apple Silicon with the compact model, Metal was about 2.6 times as fast as the CPU alone. Apple Silicon shares memory between CPU and GPU, so 8 GB already covers the balanced model. No usable GPU at all? WhisPaste's second on-device engine, Parakeet, uses a GPU when one is available and was several times faster than the compact Whisper model on the same processor (our benchmark), at the cost of language range (about 25 languages instead of 99).

Compact model, same audio, measured with whisper.cpp on Apple Silicon (M5). Details on the benchmark page.

Local vs cloud: same Whisper, different trade-offs

Local Whisper keeps audio on your machine, costs nothing per minute, and works without internet, at the price of a one-time model download and slightly more RAM. Cloud providers like OpenAI or Deepgram trade those resources for raw speed and the largest models. WhisPaste lets you pick the provider in Settings: stay local by default, switch to a cloud provider when you need maximum speed on a long recording.

FAQ

Before you get started

What is a "Whisper desktop app"?
It's a desktop application that runs OpenAI's Whisper speech-recognition model directly on your computer — not in a browser tab, not in the cloud. WhisPaste is exactly that: whisper.cpp bundled into a native app for Windows, macOS and Linux.
How does WhisPaste turn Whisper into a desktop tool?
WhisPaste bundles whisper.cpp straight into the app, so there is no Python environment to set up. On first launch you pick a model size and the app downloads it once. After that: press a hotkey anywhere on your system, speak, and the transcript is pasted right at your cursor.
Does the Whisper desktop app work offline?
Yes. Once the app and a model are installed, transcription runs locally with no cloud connection, no account and no internet. Usage statistics, crash reports and the update check are on by default and use the network, but transcription does not depend on them, and all can be switched off.
Does it run on older or non-NVIDIA GPUs?
Yes, as long as the driver supports Vulkan (Windows, Linux). One build covers NVIDIA, AMD and Intel. Without a usable driver WhisPaste uses the CPU. If a GPU crashes twice while loading a model, WhisPaste switches to the CPU, tells you, and you can re-enable GPU acceleration yourself.
Is WhisPaste free?
Yes, on GitHub: WhisPaste is open source under the MIT licence, with no account, no subscription and no paid tier. The Microsoft Store and Mac App Store versions are paid, support development and have the same core features; the Mac App Store build leaves out two macOS-internal fallbacks (AppleScript-based pasting and the permission repair tool).
Is it safe to use?
The full source code is public on GitHub and anyone can review it. Transcription runs locally by default, with no account. Usage statistics and crash reports are designed to leave out your audio and text, and you can switch both off.
Where does the Whisper transcript end up?
Right where your cursor is. WhisPaste runs Whisper as a system-wide voice-input layer, so the transcript lands directly in the editor, chat window, terminal or form field you are already working in — no upload step, no queue, no copying out of a separate transcript window. That is what turns Whisper from a speech-recognition model into an everyday desktop input method.