Speak anywhere. It types for you.
A tiny macOS menu-bar tool that turns your voice into clean text and types it
straight into whatever app is in front of you.
typeformic is a press-to-talk dictation utility. Hit a global hotkey, say a sentence, and it transcribes your speech on-device, cleans it up with a language model, and types the result at your cursor — in any text field, in any app. No window to switch to, no copy-paste.
It was built for fast, hands-on text entry: chat, code comments, commit messages, notes — anywhere you can type, you can talk instead.
- 🎙️ On-device transcription — Apple's
SpeechAnalyzer/SpeechTranscriber(macOS 26). Audio never leaves your Mac for speech-to-text. - 🤚 Auto-stop on silence — a lightweight VAD ends the utterance ~1.5s after you stop talking. Nothing to press to finish.
- ✨ LLM cleanup — strips filler words, fixes punctuation, and restores mis-transcribed technical terms (e.g. 派森 → Python, 吉特哈勃 → GitHub).
- 🔌 Bring your own model — switch between the on-device model, any OpenAI-compatible endpoint, or the Anthropic API. Base URL, key, model name, and the cleanup prompt are all configurable, with a one-click connection test.
- ⌨️ Types anywhere — output is injected as real keystrokes via
CGEvent, so it lands in the frontmost app exactly where your cursor is. The pasteboard is never touched. - 🌐 Multi-language — pick your dictation language; models download on first use.
- 🫧 Stays out of the way — a small wake pill fades in at the bottom of the screen while you talk and disappears when you're done. A stats panel tracks dictations / characters / corrections.
⌃⌥M ─▶ mic ─▶ SpeechTranscriber ─▶ VAD auto-stop ─▶ LLM cleanup ─▶ CGEvent keystrokes ─▶ frontmost app
- A global Carbon hotkey wakes the pill and starts the audio engine.
- Microphone buffers stream into
SpeechAnalyzerfor live transcription. - Per-buffer RMS energy drives a voice-activity detector that auto-finalizes after a trailing silence.
- The raw transcript is cleaned by the selected engine (on-device / OpenAI / Anthropic).
- The cleaned text is typed into the frontmost app via synthesized Unicode key events.
- macOS 26 (Tahoe) or later, Apple Silicon
- Xcode 26+
- Apple Intelligence enabled (for the on-device cleanup model). Remote APIs work without it.
git clone git@github.com:uk0/typeformic.git
cd typeformic
open MicMix.xcodeproj # the Xcode target / scheme is "MicMix"
# ⌘R to build & runThe repository is named typeformic; the Xcode app target is
MicMix. Build and run theMicMixscheme. The app lives in the menu bar (mic icon) — there is no Dock icon.
On first use macOS asks for three permissions — all are required for the tool to work end-to-end:
| Permission | Why | Where |
|---|---|---|
| Microphone | capture your voice | prompted on first dictation |
| Speech Recognition | transcribe audio | prompted on first dictation |
| Accessibility | type into other apps | System Settings → Privacy & Security → Accessibility (grant manually) |
The app is not sandboxed — system-wide keystroke injection is incompatible with the App Sandbox — and ships with Hardened Runtime plus the audio-input entitlement.
Menu bar → Settings… (⌘,):
- Dictation → Language — speech-recognition locale (Auto, or pick one).
- Cleanup Engine — On-device (Apple) or Remote API.
- Model API — provider (OpenAI-compatible / Anthropic), base URL, API key, model name, and a Test Connection button.
- Cleanup Prompt — the instructions used to tidy the transcript; editable, with reset-to-default.
API keys are stored locally in UserDefaults.
| Action | Shortcut |
|---|---|
| Start / stop dictation | ⌃⌥M (Control-Option-M) |
| Show / hide the pill | menu bar → Show / Hide Widget |
| Open settings | ⌘, |
| Statistics | menu bar → Statistics… |
Put your cursor in any text field, press ⌃⌥M, speak, and pause — the cleaned text appears.