#Install Doubao Say

Doubao Say supports macOS and Linux. The Mac client is a native Swift menu-bar app; the Linux GTK4 client supports Wayland, native X11, and optional Omarchy integration.

#macOS

Requires macOS 13 or later. Check About This Mac: choose Apple Silicon for M-series chips, or Intel for an Intel processor. The Mac app does not require Python, GTK, or PipeWire.

The backup packages above are version 1.8.1, Developer ID signed and Apple notarized. Other versions are on GitHub Releases. Choose the DMG matching your processor.

Open the DMG, drag Doubao Say to Applications, then open it from Applications. Follow the normal system prompts; do not disable Gatekeeper or remove quarantine attributes.

#First run on Mac

1. In Recognition, sign in to Doubao or enter your own API key for Volcengine Seed ASR 2.0 or Deepgram Nova-3 (US English).
2. In Setup, allow microphone access and use Enable global shortcuts to grant Accessibility permission. Complete any system authentication in the system dialog.
3. Choose a microphone, finish the three-second local check and five-second speech test, then complete setup. The local check does not upload audio; the speech test calls your selected provider and only displays the result inside the app.
4. Focus a text field and press the default Control+Option+Space, or record a shortcut in Shortcuts. Tap to start and tap again to finish, or hold to talk and release to finish.

Setup shows the current check status and can be run again. Settings save automatically. Closing Settings leaves the app running in the menu bar.

#Mac input, polishing, and updates

General offers clipboard paste or Accessibility-based direct input. If a target does not support input, retained text can be copied manually. Pressing another key during dictation finishes and inserts the text without performing that key's original action. Double-tapping the trigger sends Enter only while idle.

Optional polishing can use your OpenAI-compatible service or a ChatGPT plan connection. Recognition and polishing contact the selected providers; this is not offline recognition. Provider accounts, limits, and billing apply separately.

Check Updates and choose to download and install. The app checks the hash, developer signature, notarization, and app identity before installing and restarting. It will not install a Linux archive as a Mac update. See the Mac usage guide (Chinese) for details.

#Linux

Doubao Say is a Linux desktop application for Omarchy / Hyprland and native X11. It requires Python 3.11 or newer, plus system components including GTK4, WebKitGTK, and PipeWire. On Arch / Omarchy, the installer checks for missing packages and asks before installing them.

#Download a Linux offline bundle

Visit GitHub Releases. Choose an app (standalone) or plugin (Omarchy plugin) bundle that matches your CPU architecture and Python version. Keep the archive and SHA256SUMS in the same directory, verify the archive, then extract it and run:

sha256sum -c SHA256SUMS --ignore-missing
./install.sh --check
./install.sh

The bundle includes matching Python wheels. System libraries such as GTK, WebKit, and PipeWire are not bundled.

#Install the Linux standalone app from source

Run these commands in your Arch / Omarchy desktop session:

git clone https://github.com/quanru/doubao-say.git
cd doubao-say
./install.sh
./start.sh

Keep the source directory: the desktop launcher points to it. Other distributions need equivalent system dependencies installed separately.

#Install the Omarchy plugin

The plugin and standalone app use the same voice engine. Choose one installation path. If Omarchy asks to enable the plugin while adding it, decline until installation and first-run setup are complete.

omarchy plugin add https://github.com/quanru/doubao-say.git
cd "${XDG_CONFIG_HOME:-$HOME/.config}/omarchy/plugins/md.lifeos.doubao-say"
./install.sh
./start.sh

After first-run setup, exit the foreground app and run omarchy plugin enable md.lifeos.doubao-say.

#First run on Linux

1. Sign in with a Doubao web account, or configure your own API key for Volcengine Seed ASR 2.0 or Deepgram Nova-3 (English) in Settings.
2. Choose a PipeWire microphone and run the three-second local test.
3. Choose or record a voice shortcut, then use Try speaking to check transcription.

For full dependency, upgrade, uninstall, and troubleshooting details, read the project installation guide. Which install to pick, and which desktops can paste automatically, is in Linux and Wayland voice input.