Boilerplate of whisper.cpp + chatGPT + Electron. Goal is low latency.

Could be extended in many ways :

• add voice output with elevenlabs/Bark
• get text selection as input
• get file selection as input
• use OpenAI functions to execute "tools"
• save "memories" into Prisma db

e.g. : "convert this mov to wav please"

• Expose OPENAI_API_KEY env variable in your shell
• Copy .env.sample to .env
• create bin folder in root directory and add required binaries :
  • whisper -> main program from whisper.cpp
  • sox
  • at least one ggml whisper model (I use quantized french fine-tunes from here)
• Mac M1 binaries at your own risk
• Expect many bugs and hacks
• pnpm i
• pnpm dev