One orb. Four personalities.
Scroll to meet them — each rewrites what you say differently.
Every word, exactly as spoken — no rewrite, no delay.
Cleans up grammar and filler, keeps your own words.
Turns a ramble into crisp notes and a dash list.
Rewrites into a polished message, ready to send.
Prompt Claude out loud.
The fastest way to write a prompt is to say it. Outspoken floats over Claude Code, and your words land in the composer — punctuated, translated if you want, ready to send.
Welcome to Claude CodeIt types wherever your cursor is.
Refactor this without breaking the tests
Your thoughts move faster than your fingers.
Average typing ~40 WPM vs. speaking ~150 WPM.
English WER on the Hugging Face Open ASR Leaderboard, shown as accuracy. Apple's built-in engine is OS-managed and isn't on that leaderboard.
Every engine, running on your silicon.
Apple Intelligence and Qwen3 rewrite every personality on-device today, with Llama and Gemma landing next. Apple, Parakeet, and Whisper handle transcription — swap anytime.
On-device, no download — the default backend.
Rewrites every orb personality, fully on-device via MLX.
Open foundation models, landing as an MLX option.
Lightweight open weights, landing as an MLX option.
No download · live captions · wide locale support.
~1 GB · 25 languages · near-instant transcription.
~1.6 GB · 99 languages · the most accurate engine.
The whole journey fits in your Mac.
No account. No cloud. No audio uploaded — ever. The models run on the Neural Engine, so it works the same in airplane mode.