I built an open-source real-time AI co-pilot for meetings — runs locally, no subscription
Hey everyone,
I've been building a CLI tool called Meeting Copilot that listens to your system audio, transcribes speech in real time, and streams AI-suggested answers + follow-up questions directly to your terminal. No browser extension, no SaaS subscription — just your machine.
How it works:
- Captures system audio via BlackHole (macOS virtual audio device)
- Voice activity detection (Silero VAD) to know when someone finishes talking
- Local transcription using faster-whisper
- GPT-4o (or Claude) generates a suggested answer and follow-up question, streamed to your terminal
What makes it useful:
- Works on sales calls, job interviews, client demos — anything where you need real-time support
- Optionally pass in context files (.txt, .pdf, .docx) — it auto-selects full-context or RAG mode depending on file size
- Toggle between auto mode (VAD-triggered) and manual mode (press Enter yourself)
- Suppresses filler phrases like "hi" or "thanks" so the LLM isn't called unnecessarily
Stack: Python, faster-whisper, Silero VAD, OpenAI GPT-4o / Anthropic Claude, Qdrant (in-memory), BM25, optional Cohere reranking
All open source, contributions welcome.