Hi HN, I'm Alex. I built inkoscribe because I needed a transcription tool I could actually trust with sensitive meetings, interviews, and 1:1s — without uploading audio to anyone's cloud.

It's a macOS CLI that does real-time speech-to-text fully on-device:

  brew tap uohzxela/inkoscribe
  brew install inkoscribe
  inkoscribe -s sys   # captures system audio (meetings, calls, videos)
  inkoscribe -s mic   # captures your microphone

Requires macOS 13+ on Apple Silicon (M1/M2/M3+).

A few things I cared about while building it:

- Zero network calls. Your audio never leaves the machine.

- Real-time, word-by-word output — not batch.

- No accounts, no config, no Electron. Just a binary you brew install.

It's my daily driver for work calls. Happy to answer questions about the model choice, latency tradeoffs, capturing system audio on macOS (ScreenCaptureKit was… an adventure), or anything else.

Feedback and bug reports very welcome.