Voice Mode
Voice mode lets you talk to the CMS assistant instead of typing. It’s the same assistant and the same actions — only the input is spoken. You speak a request, it’s transcribed, the assistant acts, and you hear a short spoken confirmation.
Starting a voice cycle
There are three ways to start, all doing the same thing:
- Mic button — bottom-right of every screen.
- Keyboard shortcut —
⌘ ;on Mac,Ctrl ;on Windows/Linux. (Ignored while you’re typing in a text field, so you can still type a literal;.) - Wake word — “Hey Alokai”. Off by default; enable it in Settings → Voice.
What the screen tells you
The viewport border changes colour to show the current state:
| State | Border | What’s happening |
|---|---|---|
| Listening | cyan, pulsing | The mic is open and capturing your request. |
| Transcribing | cyan→purple | Your audio is being turned into text. |
| Thinking | purple shimmer | The assistant is working on your request. |
| Speaking | green | It’s confirming a change it made (“Done.”). |
| Confirming | amber, faster | It’s asking you to confirm something — say yes/no. |
Confirmations
For destructive actions the assistant reads the prompt aloud and waits for a spoken yes or no, so you can keep your hands off the keyboard. Your spoken request is also recorded as the consent text in the Activity Log, where voice-driven actions carry a Voice badge.
Privacy
- Your request is transcribed by your own Cloudflare Workers AI binding — it does not go to a third party.
- The wake word, when enabled, uses the browser’s built-in speech recognition, which on Chromium routes audio to Google. This is why the wake word ships off by default. The post-activation request always uses the first-party path above.
Known limitations
- Wake word is Chromium-only — Firefox doesn’t support it, and the toggle is disabled there.
- The first transcription is slow (~3 s cold start); later ones are ~1 s.
- Spoken confirmation voice quality varies between macOS and Windows.