SlashVibe

macOS Menu Bar Speech + Memory Workspace

活动标签：#小红书黑客松巅峰赛 | GitHub Topic: redhackathon

Minimal macOS push-to-talk speech-to-text app scaffolded for the architecture described in the product and hardware research docs.

What is implemented

• Native macOS menu bar app shell using SwiftUI MenuBarExtra
• Clean seams between hardware input, audio capture, transcription, and UI orchestration
• Global hold-space push-to-talk with short-tap space passthrough
• Mac microphone capture via AVFoundation
• Deepgram live transcription over WebSocket
• Keychain-backed Deepgram API key storage
• Auto-insert of final transcript into the currently focused chat or text field
• Future-ready interfaces for USB HID and USB Audio replacements
• Coordinator tests for the session state machine
• Deepgram message decoding tests
• File-backed audio source test for integration-style coverage

Project structure

• Sources/SpeechBarDomain Shared protocols, events, audio descriptors, and state enums.
• Sources/SpeechBarApplication VoiceSessionCoordinator and the app-level orchestration logic.
• Sources/SpeechBarInfrastructure On-screen push-to-talk source, global space-key capture, mac microphone capture, Deepgram client, Keychain, and transcript delivery implementations.
• Sources/SpeechBarApp Menu bar app shell and the UI.

API key handling

The app never hardcodes a Deepgram API key in source or plist files.

• Launch the app.
• Paste your Deepgram key into the secure field.
• Save it to Keychain.

The key shared in chat should be treated as exposed. Rotate it before any real use.

Build locally

This workspace currently has Swift command line tools but not a full Xcode installation.

Build the executable

cd /Users/lixingting/Desktop/StartUp/Code
swift build

Build a local .app bundle wrapper

cd /Users/lixingting/Desktop/StartUp/Code
./Scripts/build_app_bundle.sh
open ./dist/SlashVibe.app

Build a release zip for another Mac

cd /Users/lixingting/Desktop/StartUp/Code
./Scripts/package_release.sh

That creates a distributable zip in release/.

Permissions for the new workflow

To use global hold-space and automatic transcript insertion into another app:

• Allow microphone access
• Allow Accessibility access
• If macOS asks, allow Input Monitoring for the app so it can watch the space key globally

Run tests

cd /Users/lixingting/Desktop/StartUp/Code
swift test

Notes

• Default Deepgram model is nova-2 with language=zh-CN.
• Audio is normalized to 16kHz / mono / linear16 before streaming.
• KeepAlive messages are sent every 4 seconds.
• Short press on Space still types a normal space. Long press on Space starts voice capture.
• Replacing the on-screen button with a future USB HID input should only require a new HardwareEventSource.

Distribution and GitHub

• Install and sharing guide: INSTALL.md
• GitHub upload guide: GITHUB_SETUP.md

---

小红书黑客松巅峰赛 · GitHub Topic redhackathon