wtf is meetingcopilot?

MeetingCopilot is a desktop app for Windows and Apple silicon Macs that listens to a meeting or interview in real time and shows a live, scrolling transcript of what the other person is saying, along with suggested answers you can read aloud. On Windows it captures system audio directly with no extra setup, and on macOS it needs a virtual audio device such as BlackHole to route sound in. Your own voice is picked up through a separate microphone channel so the two speakers stay separate in the transcript. Speech is converted to text using one of four swappable engines: a local FunASR model that runs for free and keeps everything on your machine, a local Whisper model as an offline backup, or two cloud options through Alibaba Cloud or MiMo. Partial subtitles appear while a person is still talking. The app automatically detects switches between Chinese and English mid conversation, and you can flip the answer language with one click. The whole interface, including menus and dialogs, is also available in both English and Chinese. Suggested answers come from an AI language model that you connect using your own API key, with DeepSeek recommended. You can import your resume and a job description ahead of time, and the app writes answers meant to be read aloud, following patterns like stating a conclusion first or using a structured approach for behavioral questions, while staying grounded in what your resume actually says rather than inventing experience. A running summary of the conversation, covering questions asked and claims you have made, updates in the background so a long interview stays consistent. The app also includes a feature to select a region of the screen and ask a vision model about it, and it can hide its own window from screen sharing and recording tools, though the README notes this protection cannot be guaranteed on macOS against newer screen capture systems. Running it requires Node.js 20 or newer, and the default local speech recognition needs a Python 3.10 or 3.11 environment with the funasr and torch libraries installed.