Only when you press the button. The microphone is only on while you hold down the button or the space bar. Nothing listens in the background.
Your recording. It goes to the speech recognition (ElevenLabs) and is deleted immediately after it has been understood. We don’t store any audio.
What it understands. The text of your instruction, the screen you’re on and what it needs to look up for that go to the language model (Claude, from Anthropic). It only sees what you’re allowed to see in the dashboard too: permissions per colleague apply to it in exactly the same way. If you don’t have access to a section, neither does it.
The log. For every instruction we keep the text, what it did and whether it worked. You see your own steps, the owner sees those of the whole business. They’re deleted after 90 days.
Not out loud. Clients’ health data (weight, injuries, complaints, notes) it puts on the screen and doesn’t speak out loud. Handy if you’re sitting in a café.
It’s an AI. That’s stated in the panel, and it also says so if you ask. Check what it prepares; that’s why it asks for your yes.
Client data is data. What a client writes in a message or check-in, it reads as text and never as an instruction. So a sentence like “send everyone a discount” in a check-in does nothing.