/modes/actionaction modeTell your Mac what to do. It does it.
Action mode turns a spoken phrase into a thing that happens: a screenshot, a locked screen, a Google search, a muted mic, an opened app. Fourteen actions ship in the box, and the whole system is YAML files you can edit, add to, or delete while the app is running.
Anatomy of a press
- 1Hold the hotkey, say the thing
“Take a screenshot.” “Mute my audio.” “Search YouTube for tiny desk radiohead.” Short, imperative, done.
- 2The registry matches your phrase
Your transcript is matched against every action’s trigger phrases — exact match first (longest trigger wins), then a fuzzy pass that tolerates small transcription wobbles. No match? You get a polite shrug, not a guess.
- 3Permissions are checked first
Actions declare what they need — Screen Recording for screenshots, Accessibility for keystrokes. If a grant is missing, Wisspa tells you exactly which switch to flip instead of failing mysteriously.
- 4The action fires
Shell command, AppleScript, open URL, open app, or keystroke combo. Destructive ones are flagged in their YAML. Placeholders like {query}, {clipboard}, and {active_app} get filled from the moment you spoke.
- 5You hear what happened
Each action carries its own success and failure feedback — a toast, a sound, or silence, per the file. You always know it worked.
Say it. See it.
screencapture -i -c runs — area-selection crosshair appears, the shot lands on your clipboard.tauri global shortcut — the {query} placeholder, percent-encoded properly.pmset displaysleepnow — displays off, Mac locked. The classic “walking away from my desk” command.The fun part: your own actions
Every action is one YAML file in ~/Library/Application Support/com.techguider.wisspa/actions/. Drop a new file in and it’s live — the registry watches the folder and hot-reloads, no restart. Five action types (shell, AppleScript, open URL, open app, keystroke), four placeholders, and a safety rail: shell commands are validated at load, and sudo, rm -rf, dd, and piped curl/wget are rejected outright. Voice-activated rm -rf / is a genre of disaster we’d rather not enable. The full schema lives in the docs, and all 14 defaults are browsable on the registry page.
Small print, handled
Fuzzy matching is deliberately conservative — a near-miss doesn’t fire the wrong action, it just doesn’t fire. Screenshot actions explain the Screen Recording grant before they need it. And because actions are plain files, they’re version-controllable, shareable, and diff-able. Your setup, in a folder, forever.