yappatron

just yap.

voice dictation for mac with OpenAI Realtime, local mode, always-on listening, and configurable push-to-talk.

macOSopenai realtimecloud + localfree
github get started

features

instant streaming text

OpenAI Realtime turns your voice into text as you speak. It feels less like dictation software and more like a live voice keyboard.

built for vibe coding

Talk to your AI. Describe what you want. A humane end-of-utterance pause sends the thought when you actually stop.

modern cloud stt

OpenAI Realtime with gpt-realtime-whisper is the default fast path. Deepgram Nova-3 is still available when you want that backend.

knows who's talking

On the Deepgram backend, enroll your voice once and Yappatron can label two-person conversations with hybrid cloud + local voiceprint matching.

local or cloud — your call

Switch between OpenAI Realtime, Deepgram cloud STT, and fully local Nemotron from the menu bar. Your words, your rules.

  1. start talking no button to press. just speak naturally wherever your cursor is.
  2. watch it stream text appears character by character as you talk, with live corrections.
  3. pause to send stop talking and it auto-presses enter. perfect for chatting with claude.
  4. keep going it's always listening. start your next thought whenever you're ready.

install

git clone https://github.com/alosec/yappatron
cd yappatron
./scripts/run-dev.sh

first run

On first launch, macOS will prompt for Microphone and Accessibility permissions. Grant both, then restart the script. If dictation isn't working, go to System Settings → Privacy & Security → Accessibility, remove yappatron with the minus button, then re-add it with plus.

requirements

  • macOS 14+ (Sonoma)
  • Apple Silicon (M1/M2/M3/M4)
  • Microphone permission
  • Accessibility permission
  • Internet + API key for cloud STT

philosophy

The best interface is no interface.

Voice dictation should be invisible. Think, speak, words appear. No buttons, no modes, no cognitive load.