Voice-to-text for developers

Your AI deserves
better context.

Hold a key. Speak. Your words appear wherever your cursor is — in Claude Code, Cursor, or any text field.

Download for Mac Download for Windows
Claude Code — ~/project/koedesk
❯
claude-sonnet-4-6 · 0 tokens Bypass Permissions: ON esc to interrupt
· Transcribing ·
✓

How it works

Fn
or
Alt
mac / win

01

Hold

Hold a configurable key. A floating pill bar appears with a live waveform.

02

Speak

Speak naturally. A live waveform confirms the mic is listening.

|
typed at cursor

03

Done

Transcription appears at your cursor, in any app.

Supported models

ElevenLabs
Scribe v2
The world's most accurate STT model. 2.3% WER — #1 on every major benchmark. This is what koedesk uses by default.
2.3%
Word Error Rate
36
Languages
#1
Global benchmark leader
Filler words (um, uh) are automatically cleaned at the STT level — no post-processing needed. Keyterms support lets you feed domain-specific vocabulary (product names, codebases, proper nouns) for even higher accuracy.

Plans

Free

$0

5 minutes per day
ElevenLabs Scribe v2
No credit card required

Pro

$15/mo

Unlimited transcription
ElevenLabs Scribe v2
Cancel anytime

Get started

Features

Push-to-talk

Hold to record, release to transcribe. Instant response, no delay.

Voice Activity Detection

Silero VAD filters silence before sending — faster, cheaper, cleaner.

36 languages

Auto-detected from OS locale. No configuration needed.

Custom dictionary

Add specialized terms, project names, or proper nouns to improve accuracy.

macOS + Windows — built on Tauri

Apple Silicon and Windows x64. Uses the OS's own WebView — no bundled Chromium, fraction of the memory footprint, near-instant launch.