Skip to content
Jonathan Andreanus
Jonathan Andreanus
All work

Emit

A push-to-talk dictation app for macOS that turns speech into clean text or a structured coding prompt, and falls back to a local whisper.cpp server when cloud providers fail.

Year
2026
Role
Solo project
Stack
Tauri 2, Rust, React, TypeScript, whisper.cpp
Emit's main window in Clean mode, ready to record, with the hint "Hold Right Option anywhere".

Hold Right Option anywhere and speak. Emit has Clean and Raw modes for dictation, and a Dev mode that turns speech into a structured coding prompt. It can type the result into the focused app.

How it works

  • Speech-to-text falls back from 9router to Groq to any OpenAI-compatible endpoint, and then to a local whisper.cpp server. Each provider has a cooldown and a health check.
  • Optional streaming transcription runs over a WebSocket owned by the Rust side, with a bounded audio queue.
  • API keys live in the macOS Keychain.

Status

Version 0.1.0, self-signed and not notarized. The source is private. I fixed a memory leak, which brought memory use down from 642 MB to 40 MB.

Media