Why maclisten?

Privacy by default

maclisten sends nothing to a third party unless you explicitly disable on-device recognition:

• Audio is processed by Apple’s native Speech framework.
• With --on-device, recognition happens entirely on your Mac.
• No API keys, no audio uploads, no cloud billing.

Why a CLI instead of a web API?

• No network round-trip — audio stays on the Mac.
• No API keys — uses the Apple ID already on the machine.
• Predictable latency — especially when --on-device is available.
• Pipe-friendly — output JSON lines can drive shell scripts and agents.

Tiny footprint

• Single ~500 KB Swift binary.
• No model download.
• No daemon, no background service, no configuration file.
• Starts instantly and exits cleanly.

Why not whisper.cpp?

whisper.cpp is excellent, but:

• It requires downloading and loading a model.
• Model files are large (hundreds of MB to GB).
• It is not the native macOS ASR path.

maclisten uses the Speech framework that is already installed, optimized, and kept up to date by Apple. The binary is small and starts instantly.

Why not Siri?

macOS provides no public API to invoke Siri from a CLI or to register arbitrary voice commands. maclisten gives you the ASR piece; you can feed the text into your own rules engine or LLM to build a Siri-like command layer.

When maclisten is the right tool

• You need a quick, local transcript of an audio file.
• You want a long-running microphone listener that emits JSON.
• You are building an agent that needs speech input on macOS.
• You want to keep the raw audio alongside the transcript.