Offline-first AI
Parakeet TDT and Whisper run locally. Your voice never leaves the machine — a guarantee, not a setting.
RAPTOR LABS / SPEECH-TO-TEXT DAEMON
SPEAK. IT TYPES.
Hold a key, say the words, release — your speech lands as text at the cursor, in any app. Transcribed on your machine. No cloud. No clipboard residue. No subscription.
$ curl -fsSL https://typemic.com/downloads/typemic-1.0.0.tar.gz | tar xz
$ cd typemic-1.0.0 && ./scripts/install-deps.sh
$ ./bin/typemic setup
01
Three motions.
Three motions.
Zero friction.
-
HOLD
Press your shortcut
A global hotkey — works everywhere, even fullscreen. Toggle, push-to-talk, or auto tap/hold hybrid.
-
SPEAK
Say the words
A minimal on-screen waveform confirms it's listening. Local models transcribe in real time on CPU.
-
RELEASE
Text appears
Injected straight at the cursor — terminals, browsers, IDEs, chat. Your clipboard is restored untouched.
02
Built like an instrument,
Built like an instrument,
not an app.
Six STT backends
Local: Parakeet, Whisper.cpp, faster-whisper. Cloud (your keys): OpenAI, Groq, Gemini, ElevenLabs.
Five recording modes
Toggle · push-to-talk · auto · continuous · long-form. Dictate a sentence or a chapter.
Wayland-native OSD
GTK4 layer-shell overlay with a live oscilloscope waveform. Looks like it shipped with your compositor.
Self-healing
Survives USB mic hotplug, PipeWire restarts, suspend/resume. It recovers; you keep talking.
Yours to tune
Word overrides, filler-word filtering, symbol replacement, per-language shortcuts — one JSON file.
03
TypeMic for iPhone & Android
The same engine, as a keyboard. Dictate into any app on your phone — on-device, private, fast. Early-access members get it first, free.
- Custom keyboard with live waveform key
- On-device transcription — works in airplane mode
- Free for early-access members
04
Found a rough edge?
TypeMic is in active development. Reports from real machines shape every release.