Jev Voice：用语音控制 macOS 的开源菜单栏应用
Swift/SwiftUI 菜单栏应用，所有类型化决策交给 TypeSafe Jev，DeepSeek 或本地模型只负责生成文本，可语音打开 Devin、发提示词、换模型、搜网页，MIT 开源。
I was building Jev Voice on the side in a public repo when I noticed I have 3 stars and 2 forks and I haven't even posted about it https://chris-wozniczek.github.io/jev-voice-control/ I am building in the meantime a macOS app with Jev to control my mac Just following the Jev trend 😎 Most videos you have seen is just open notes and google search You can see how I can open Devin, send a prompt, change the model, search on the internet, check X all with my voice Still a bit buggy, but it kinda works. Tech stack > Swift / SwiftUI menu-bar app, SwiftPM, macOS 14+ (26 for streaming speech), distributed as a Homebrew cask. > Apple frameworks: SpeechAnalyzer / SFSpeechRecognizer, AVFoundation, Accessibility (AXUIElement), CoreGraphics CGEvent for mouse/keyboard, Vision (OCR), NSWorkspace/AppleScript (osascript) for app and system control, AVSpeechSynthesizer for talk-back, os_log for the debug log. > TypeSafe Jev (jev-latest) for all typed decisions; DeepSeek or a local oMLX model only for generated text and the last-resort fallback planner; embedded Cua driver as a secondary observe/act helper; Chrome DevTools Protocol as a browser fallback. > WhisperKit as the optional local Whisper engine. Building this for fun, fully in Devin Cloud using macOS Cloud VMs - so whenever I have time I just shoot a prompt there, test it locally Full documentation in github, point your agent to it if you want. Usually just reactive building to devin integration email notifications from github pinging me I have PRs to merge Opesource, MIT license