100% on-device · macOS

Speak it.
Mando types it.

Hold a key and talk. Filler words, false starts, run-ons — gone. Clean, punctuated text lands at your cursor, transcribed and polished entirely on your Mac.

Free & open source · Apple Silicon · no account, no cloud

whisperwoof · listening
Mando listening
You said
✦ polished by Mando
 
Three seconds, start to finish

From mumble to manuscript

No window switching. No copy-paste. The key you press decides where the text goes.

1

Hold Fn

Mando's ears perk up. A small floating indicator shows you're recording. Start talking.

2

Just talk

Ramble. Say "um." Restart your sentence. On-device Whisper hears every word — Chinese and 90+ languages included.

3

Release

A local model cleans it up in ~250ms and drops polished text right at your cursor — or routes it with Fn+T / Fn+N.

Built for people who talk fast

Everything runs on your laptop

Speech-to-text

On-device transcription

Whisper runs locally and is the default — multilingual, including Chinese. Prefer raw speed for English & European? Opt into NVIDIA Parakeet. Nothing is ever uploaded.

On by default

Smart Cleanup

A small bundled local model strips filler, fixes grammar and punctuation, assembles spoken emails, and keeps your voice. Set up in one click; tuned for ~250ms polish.

Hotkey routing

The combo is the intent. Paste at cursor, send to clipboard, save a Markdown note, tag a project — explicit, not magic.

Clipboard timeline

Everything you copy, in a card timeline grouped by day — with the source app and image thumbnails. Click to re-copy.

MCP plugins

Route voice to Todoist, Notion, Slack, your calendar — any Model Context Protocol server works as a plugin.

The whole point

Nothing leaves your Mac.

Speech-to-text and the cleanup model are bundled and run on-device. No account. No telemetry. No cloud dependency. Cloud providers are strictly opt-in, behind your own keys.

  • Bundled whisper.cpp + sherpa-onnx for speech, llama.cpp for polish
  • A privacy lock blocks every cloud call with one toggle
  • Graceful by design — no model? you still get a clean raw transcript
# network activity while dictating
requests ……………… 0
bytes uploaded …… 0
accounts requirednone
# where your voice goes
mic → whisper.cpp → llama.cpp
→ your cursor. that's it.
Mando, sitting

Good dog. Clean text.

Free and open source. Built for Apple Silicon. Your voice never leaves the machine.

macOS, Apple Silicon A microphone ~200 MB free space

Unsigned build — on first launch, right-click the app and choose Open,
or run xattr -dr com.apple.quarantine /Applications/WhisperWoof.app