Processed locally
Local dictation sends no audio to a transcription server after the one-time model download.
Local mode · Free · MIT open source formerly HushType
Lamitype| Your voice stays on your Mac.
In Local mode, Lamitype turns speech into text on your Mac. After the one-time model download, dictation works offline, with no account or app telemetry.
Download free for Mac
macOS 15+ · Apple Silicon
No saved recordings · No app telemetry · No word limits
Real Lamitype capture. Hold Right Option, speak an English coding prompt, then release to place the transcript at the cursor.
Local privacy, with clear boundaries
Your audio can stay on your Mac.
Local speech recognition runs on your Mac after the model download. The app is open source, and the ordinary dictation path does not save audio recordings.
Discarded after transcription
Audio is held in memory while Lamitype transcribes, then discarded instead of saved as a file.
Open to inspection
The app contains no telemetry and its MIT-licensed source is public.
Useful where you type
Lamitype pastes at the cursor in most editable Mac apps. If insertion is blocked, the transcript remains on the clipboard.
Destination apps have their own policies. Once text is pasted or sent elsewhere, that app handles it under its own privacy policy.
Cloud features are optional. They send audio to the provider you choose only when you enable them and supply the needed credentials.
How it works
Three steps from thought to text.
Download and open
Install the free app. Setup walks through microphone and Accessibility permissions.
Hold Right Option
Focus an editable text field, hold the shortcut, and speak naturally.
Release to paste
The transcript appears at the cursor. If an app blocks insertion, it remains on the clipboard.
Optional · Text Polish
Double-tap to polish a short selection.
Select a short passage in an editable app and double-tap Right Option. Text Polish attempts an in-place correction, then shows a diff of the result.
Manual polish
- Select a short passage.
- Double-tap Right Option.
- Review what changed in the result diff.
Limits stay visible
Review the changes
The small on-device model can miss issues or over-edit. Automatic polishing is off by default, and the raw transcript is inserted unchanged if it errors or times out.
Local model footprint
Model size comparison
Model weights at the precisions shown, in MB.
≤ 2.1 GB maximum observed RAM. For the whole Lamitype process with the local model loaded, including the MLX idle cache. Unload the model from the menu with one click to release its weights and cached buffers; a small amount of app memory remains.
FAQ
Before you download.
Is Lamitype really free?
Yes. Lamitype is free under the MIT license, with no account, subscription, app telemetry, or word limits. Optional cloud recognition uses your own provider API key, so provider charges may apply.
Does Local mode need an internet connection?
The one-time model download uses the network. After that, local speech recognition works offline. Update checks connect to GitHub only after you choose Check for Updates and confirm. Optional cloud features use the selected provider when enabled.
Does Lamitype save audio recordings?
Local dictation holds each recording in memory while it transcribes, then discards it. Lamitype does not save ordinary dictation audio as files. The transcript is placed on the macOS clipboard, which can persist or sync according to your Mac settings.
How much memory does local dictation use?
Lamitype’s default local model has about 708 MB of model weights. The initial download may include additional files. Maximum observed RAM for the whole Lamitype process is ≤ 2.1 GB with the local model loaded, including the MLX idle cache. Unloading the model frees its weights and cached buffers.
Which speech recognition model does Lamitype use?
Local mode runs Qwen3-ASR 0.6B as a 4-bit MLX quantization, about 708 MB of weights, executed in-process by native Swift on Apple Silicon with no Python runtime or local server. Qwen3-ASR handles mixed Chinese and English natively, and OpenCC guarantees Traditional Chinese output. You can also use your own key for OpenAI or Gemini cloud recognition.
Where can I dictate?
Lamitype pastes the transcript at your cursor in most editable Mac apps. If an app blocks direct insertion, the transcript remains on the clipboard so you can paste it yourself.
Which Macs are supported?
Lamitype supports Apple Silicon Macs with M series chips running macOS 15 or later. Intel Macs are not supported because the local engine uses Apple's MLX framework.
What is Text Polish?
On macOS 26 or later, Text Polish can use Apple Intelligence to attempt an in-place correction of a short selection, then show a diff of the result. It can miss issues or over-edit.
What is the relationship between Lamitype and HushType?
They are the same product. HushType was renamed Lamitype in 2026. The developer, codebase, and version history continue on GitHub.
Turn your voice into text with minimal friction, while keeping privacy first.
Free, open source, and no account needed. If it’s not for you, just move it to the Trash. There’s no subscription to cancel.