Feature
Say it, and it happens
Voice is here because speaking is the fastest way to hand over a job with several parts in it.
Wake word, or a key
Say the wake word and Caelia starts listening, or hold a key if you would rather it never listened on its own. Both are settings, and the microphone is off when you have not asked for it.
Talk the way you talk
You do not have to phrase things as commands. Sentences with an afterthought in the middle work, because what reaches the model is your sentence and not a keyword match.
It answers out loud
Replies are spoken as well as written, with a voice and a personality you pick in settings. Turn the voice off and it stays text.
What voice is actually for
Dictation is not the point. The point is that the thing you say out loud becomes work that gets done: open this, find that, put those two together, tell me when it is finished. Speech is the fastest way to hand over a task with several parts in it, which is why it is here.
Voice notes over WhatsApp
The same listening works from your phone. Send Caelia a voice note on WhatsApp and it acts on it, which is the version of this most people end up using - you are away from the desk and the machine is not.
What it needs
A microphone, and an internet connection for the model that understands you. The audio goes to the AI provider you connected, the same as your typed messages do; nothing about your speech is stored by us. What leaves your computer is written out in full on the privacy page.
Is the microphone always on?
Only while wake-word listening is switched on, or while you are holding the key. It is your choice which of the two, and off is also a choice.
Does it work in accented English?
The recognition is done by whichever AI provider you connected, so it is as good as theirs. The major providers handle South Asian English well in practice.
Can I use it without voice at all?
Yes. Every single thing Caelia does can be typed. Voice is an input, not a requirement.
Stop typing instructions.
One payment, and your PC starts listening when you want it to.