A desktop AI companion built with Electron and Live2D Web SDK.
- Transparent desktop window
- Live2D character display
- Basic interactions (click response)
- System tray integration
- Draggable window
npm install
npm start- Development mode stores the agent workspace in
./agentWorkspace. - You can override it with
CELCAT_WORKSPACE_DIR. - Packaged installs fall back to
Documents/CelCat/agentWorkspaceunless overridden.
- Local skills can be discovered automatically from the default Codex and Agents skill directories.
- You can override skill scan roots with
CELCAT_SKILL_DIRS. - External MCP servers can be injected with
CELCAT_MCP_SERVERS_JSONorCELCAT_MCP_CONFIG_PATH.
npm run dist:win- Installer output:
release/ - Installer type: NSIS (
setup.exe)
VOLCENGINE_REALTIME_PROVIDER_MODE=dialogueCurrent legacyopenspeech realtime dialoguetransport.VOLCENGINE_REALTIME_PROVIDER_MODE=voiceChatOfficial-route migration mode. The current implementation keeps a compatibility transport while the app is being moved towardStartVoiceChat + Function Calling + MCP + Memory.
See docs/volcengineVoiceChatMigration.md for the migration plan.
- Main Process: Window management, system tray
- Renderer Process: Live2D display, user interactions
- Live2D Manager: Model loading and animation
- Real Live2D model integration
- AI conversation system
- Voice recognition
- System integration features