No. Speech recognition runs entirely on-device using Apple's engine. Only the recognized text is sent out for LLM cleanup — never the audio. A fully offline mode (local model via Ollama) and bring-your-own-key are also available.
Yes — that's the core use case. Say 「我用 Claude Code 写了个 component」 and it comes out exactly like that: no mode switching, and technical terms survive recognition.
Built-in dictation transcribes literally — fillers, false starts and recognition errors included. ListenIn adds a cleanup pass that removes filler words, repairs sound-alike recognition errors, and formats the result the way you would have typed it.
2,000 cleaned words per week, no account required. Pro ($12/month or $96/year) removes the cap.
Yes. A dedicated mode turns spoken rambling into a structured prompt — goal, numbered steps, constraints — and can send it straight into your terminal or editor.
macOS 14.4 or later on Apple Silicon. The newest recognition engine (macOS 26) is used automatically when available.
Yes. Correct a word once and it goes into your personal dictionary; future recognitions and cleanups use it.