Swabble

Why it's ranked here

The design is narrow and practical: local speech recognition feeds ordinary shell commands without a hosted service or bundled speech model. Configurable cooldown, minimum length, and timeout controls reduce accidental hook traffic. Its value is constrained by strict platform requirements and several unfinished service controls.

What's good

Audio remains on the Mac during transcription, apart from macOS obtaining Apple speech assets when needed. Hooks receive both a rendered text argument and protected environment values. Recent transcripts can persist locally for debugging. The same speech pipeline, configuration loader, hook runner, and transcript store are available to Swift applications.

Tradeoffs

It requires macOS 26 and Swift 6.2, so older Macs and other platforms are excluded. Wake detection is only text matching, not a dedicated detector. Preferred microphone settings are saved but ignored during live capture. Background service installation and start, stop, and restart controls remain stubs or placeholders.

How to use it well

Use Swabble for Mac-only voice shortcuts where a shell command is already the useful integration boundary. Keep hooks idempotent because recognizer output can be partial, retried, or corrected. Test commands before leaving the microphone loop running. Choose another tool if you need cross-platform capture, advanced wake detection, or finished service management.

Observed

Primary language

Swift 6.2

Platform support

macOS 26 only

Packaging

Swift Package Manager package with library and executable products

Interfaces

Command-line application and embeddable Swift library

Speech processing

Apple Speech.framework on-device transcription; macOS may download speech assets

Installation surface

Builds from source with Swift Package Manager

Testing structure

Package manifest declares a swabbleTests test target

External package dependency

Commander, declared from version 0.2.2