Skip to content

Get started

Installation

Get OpenWhisper on your computer, grant permissions, and keep it up to date.

On this page

Native packages include Python and every app dependency. On Windows x64 and Linux x86_64 they also include the transcribe.cpp speech runtime for the default engine. Speech models and other runtimes download from inside the app, so you only fetch what you use; the all-in-one installers add a compact English model so you can start without any download.

Download for your platform#

Get the installer from the download section or GitHub Releases.

PlatformRequirementsInstall
WindowsWindows 10/11, 64-bitRun the .exe. It installs for your user without administrator rights and upgrades in place.
macOSApple Silicon, macOS 14+Open the .dmg and drag OpenWhisper into Applications.
Debian / UbuntuDebian 12+ or Ubuntu 22.04+, x86_64Install the .deb with your package manager.
Arch LinuxArch-compatible distribution, x86_64Install the .pkg.tar.zst with Pacman.

Slim or all-in-one

Windows and Linux releases come in two flavors:

  • Slim (OpenWhisper-Setup-<version>.exe, OpenWhisper-<version>-linux-…): downloads your speech model on first use, from guided setup or Downloads.
  • All-in-one (OpenWhisper-AllInOne-Setup-<version>.exe, OpenWhisper-AllInOne-<version>-linux-…): also bundles Parakeet TDT-CTC 110M, a 135 MB English model, so dictation, live preview, uploads, and Meeting Mode work offline right after installing. It's selected automatically until you choose another model, shows an Included badge in Settings, and is kept by Windows in-app updates.

For Linux, replace <version> with the version in your downloaded filename and use the matching command (add AllInOne- after OpenWhisper- for the all-in-one package):

# Debian / Ubuntu
sudo apt install "./OpenWhisper-<version>-linux-amd64.deb"

# Arch
sudo pacman -U "./OpenWhisper-<version>-linux-x86_64.pkg.tar.zst"

Open the app from your applications menu. Linux packages also provide the ow and openwhisper terminal commands; launching either while OpenWhisper is already running brings the existing window forward instead of starting a second copy.

Verify and open the app#

Each release includes SHA256SUMS.txt. Compare your installer's SHA-256 with the entry for that filename before opening an unsigned download. The website also shows installer checksums in Downloads.

# Windows: use the filename you downloaded
Get-FileHash ".\OpenWhisper-Setup-<version>.exe" -Algorithm SHA256
# macOS
shasum -a 256 "OpenWhisper-<version>-macos-arm64.dmg"

# Linux: put SHA256SUMS.txt beside the downloaded package
sha256sum -c SHA256SUMS.txt --ignore-missing

Allow macOS permissions#

In System Settings → Privacy & Security, allow:

  • Microphone to record your voice.
  • Screen & System Audio Recording to capture meeting audio and to take walkthrough screenshots, including Auto Mode.
  • Accessibility to paste a finished transcript into another app automatically.

You can copy and paste transcripts manually without Accessibility. Set up auto-paste… on Settings Home or in Settings → App → General names the app that needs permission and checks every second until access is on. Because macOS builds are ad-hoc signed, an update can withdraw the permission; if auto-paste is on, OpenWhisper notices on the next launch and reopens the setup window. Open Accessibility Settings never resets anything. If the switch is already on but access isn't detected, choose Repair old Accessibility entry…, which removes only OpenWhisper's entry and asks again; then turn the current app on.

Install a speech model#

A brand-new install offers a short guided setup once the main window opens. It tests your microphone, sets your shortcut, and can download the speech model without leaving setup; an all-in-one install shows its model as Included and ready. See Your first transcript.

To manage models yourself, open Settings → Models & storage → Downloads. Download weights from Speech models and the engine's runtime from Components; both tabs also show available updates. Parakeet Unified's transcribe.cpp runtime comes with the Windows and Linux packages. The first time Parakeet or Nemotron needs a runtime, OpenWhisper offers the NVIDIA GPU runtime if your card can use it, with the CPU runtime as the alternative.

Update OpenWhisper#

OpenWhisper checks GitHub for a new release every time it starts and every 4 hours while it runs. When one is available, an Update to x.y.z button appears in the title bar and the tray menu and stays until you update. You can also check yourself: Help → Check for Updates on Windows and Linux, OpenWhisper → Check for Updates on macOS. Turn automatic checks off in Settings → App → General → Updates.

To try releases a few days early, turn on Get beta updates in the same group. You're then offered whichever of the newest beta and the newest regular release has the higher version, verified exactly like any release. Turning it off never downgrades: you keep the beta until a regular release with the same or a higher version is out.

  • Windows: apply the in-app update or follow the setup prompt.
  • macOS: the verified update installs and restarts the app, restoring the previous version if startup fails. If the app cannot replace itself, it opens a verified DMG for replacement in Applications.
  • Linux: install the new package with the same package-manager command as before. If you upgrade while OpenWhisper is running, it offers Restart now once nothing is recording or transcribing.

Run from source#

Linux distributions without a package and contributors can run from source with Git and uv:

git clone https://github.com/Knuckles92/OpenWhisper
cd OpenWhisper
uv sync --locked

uv uses Python 3.12, downloading it if needed, and installs the pinned dependencies into .venv/. Launch with uv run --locked python main.py on Windows or ./scripts/openwhisper on macOS and Linux. The source setup instructions cover system libraries on Linux and the optional ow command, and development with uv covers tests and editor setup.

Intel Macs and macOS versions before 14 use setup with pip instead. An existing pip setup in venv/ keeps working: the launchers look in .venv/ first and fall back to it.

Source checkouts don't check for updates on their own. Update with git pull --ff-only and uv sync --locked; the running app notices the new commit and offers a restart.

Checked against OpenWhisper 2.8.97.App documentation
Still stuck?

Get help from the community, or tell us what went wrong.