Back to the Gyuteiverse
Open source · Flagship 02

Clanker Yap

Local-first push-to-talk dictation for terminals and desktop apps.

Hold the global hotkey, speak, release, and Yap transcribes on-device with whisper.cpp before pasting into the focused app. Terminal-aware paste modes and SQLite-backed history stay close at hand.

Active buildLinux x86_64 AppImage · v0.3
GitHubReleasesDocumentation
Version
0.3.1
License
MIT
Platform
Linux x86_64
Stack
Tauri v2 · Rust · whisper.cpp
Network
None while dictating

Clanker Yap is a small desktop app that turns speech into text wherever your cursor already is. Hold a hotkey, talk, let go, and the transcription is pasted into the focused window: a terminal, an editor, a browser, a chat box.

It exists because dictating prompts to coding agents in a terminal is awkward with most dictation tools. Yap runs Whisper locally, needs no account, and knows when the focused window is a terminal.

How it works

Hold, release, paste

  1. 01

    Hold

    Press and hold the global hotkey in any app. A floating pill shows live microphone levels.

    Clanker Yap recording, with live audio level bars
  2. 02

    Transcribe

    On release, whisper.cpp transcribes the audio on your CPU. Nothing is uploaded.

    Clanker Yap processing a capture
  3. 03

    Paste

    The text is copied and pasted into whatever you had focused. Terminals get the right paste shortcut.

    Clanker Yap ready and awaiting the shortcut

Private by construction. No cloud transcription, no audio upload, no account, and no external service anywhere in the transcription path. The only network use is the optional model download.

Features

What it does

Local transcription

whisper.cpp through the whisper-rs bindings. CPU only; no GPU required.

Push to talk

A global shortcut, defaulting to Ctrl+Shift+V. On Wayland it uses the desktop's GlobalShortcuts portal.

Terminal-aware paste

On Linux it detects the focused application and picks standard or terminal paste (Ctrl+Shift+V / Shift+Insert) automatically, or you can force a mode.

History

Settings, transcription history and a running word count live in a local SQLite database, with copy controls for each entry.

Model choice

Any GGML Whisper model. The app can download ggml-base.en.bin for you; larger and multilingual models can be dropped in.

Quiet by design

A single-instance desktop app with a small window. Configure it once and mostly forget it.

Paste modes

Auto-paste can be switched off to copy to the clipboard only. Otherwise the shortcut style is:

Auto
Chooses terminal or standard paste from the focused app.
Standard
Ctrl+V, as normal applications expect.
Terminal
Ctrl+Shift+V or Shift+Insert, for terminals and Vim.
Clanker Yap paste settings showing auto-paste and the Auto, Standard and Terminal shortcut styles
Models

Pick the speed and accuracy you want

Any GGML Whisper model works. Bigger models are more accurate and slower, and everything runs on the CPU.

Clanker Yap model settings showing the installed ggml-base.en.bin model and its path
ModelFileSizeSpeedNotes
Tinyggml-tiny.en.bin75 MBFastestLowest accuracy
Baseggml-base.en.bin148 MBFastDefault; in-app download
Smallggml-small.en.bin466 MBMediumBalanced
Mediumggml-medium.en.bin1.5 GBSlowHigh accuracy
Largeggml-large.bin2.9 GBSlowestHighest accuracy

English-only models end in .en. Multilingual variants cover 99+ languages. Models live in ~/.local/share/voice-transcribe/.

Install

Get it running

Releases are a Linux x86_64 AppImage. You can also build from source.

Release

Download the AppImage from the latest release, make it executable and run it. On first launch, download a model from inside the app or place one in ~/.local/share/voice-transcribe/.

shell
chmod +x 'Clanker Yap_X.Y.Z_amd64.AppImage'
./'Clanker Yap_X.Y.Z_amd64.AppImage'

On Wayland, the first launch opens your desktop's shortcut dialog. Approve or choose a key for “Hold to dictate”.

From source

Needs the Rust stable toolchain, Node.js and npm, plus WebKitGTK and build tools (package lists for Arch, Debian and Fedora are in the README).

shell
git clone https://github.com/Bynzski/clanker-yap.git
cd clanker-yap
npm install
npm run tauri:dev

Getting started · Troubleshooting · FAQ

Under the hood

A small, layered Rust core

A Tauri v2 app: a thin web UI over four Rust layers.

  1. CommandsTauri commands the UI calls over IPC.
  2. ApplicationOrchestrator, runtime state and use cases (transcribe, paste, model download).
  3. DomainSettings, transcription entity, errors, constants.
  4. InfrastructureAudio capture (cpal, resampling to 16 kHz), whisper-rs, paste injection (enigo), SQLite (rusqlite).
pipeline
hotkey down  → cpal capture → live level bars
hotkey up    → resample to 16 kHz mono
             → whisper-rs (whisper.cpp, CPU)
             → clipboard + paste (standard | terminal)
             → SQLite history

Architecture notes · Development guide

Status

Where it stands

Version 0.3, in active development. Known limits:

  • Release builds are Linux x86_64 AppImages only. Windows and macOS are not supported targets.
  • Wayland is supported through the desktop GlobalShortcuts portal. A desktop without that portal reports a setup error rather than falling back to a shortcut that leaks keystrokes.
  • On KDE/Wayland, auto-paste may trigger a one-time "Remote Control" permission prompt. Turning auto-paste off avoids it.
  • X11 is smoke tested, and shortcuts are configured inside the app there.
  • The default model is English-only. Choose a multilingual model for other languages.
Open source

Read it, run it, change it

MIT licensed. Bug reports, platform testing and pull requests are welcome.

SourceIssuesContributing guideChangelogDocumentation
Tauri v2Rustwhisper.cppOn-deviceSQLiteOpen Source