Docs / Voice
Voice calls
Visitors can call your lens and talk to it in real time, with the same grounding rules as text chat.
A lens on a paid plan can take live voice calls. A visitor taps Call on your lens page and talks to your lens the way they would talk to you: they speak, it answers out loud, and they can interrupt it mid-sentence. Speech recognition and speech output both run through ElevenLabs, so a call sounds the same in every browser and never depends on the device’s own speech engine.
By default the lens speaks with a clean platform voice. You can replace that with a clone of your own voice, or with a ready-made voice you pick from a list. See Cloning your voice.
How a call starts
Tapping Call asks the browser for microphone access, because a call cannot listen without it. Once the visitor allows it, the call screen opens: your name and photo at the top, and a soft gradient glow along the bottom that moves with the conversation. It swells while the visitor speaks, pulses while the lens thinks, and flows while it answers.
If the visitor blocks the microphone, or their browser remembers an old block and skips the prompt, a short message explains what happened instead of dropping them onto a call screen that cannot hear them. They can unblock the microphone in their browser settings and tap Call again.
Talking in real time
Calls are real conversations, not recordings played back and forth. The lens starts answering about a second after the visitor finishes a sentence, and it keeps its replies short and spoken. Visitors can interrupt naturally: if they start talking over an answer, the lens stops and listens, just like a person would.
The call listens hands-free, and background noise or room chatter will not trip it. A live transcript of both sides appears above the controls, so nothing said on the call is lost.
Same grounding rules as chat
A call follows exactly the same rules as text chat: answers come only from the material you have added, nothing else. If a question falls outside what you have shared, the lens says so out loud, in the same voice, instead of going quiet or guessing. See Facts and citations for how grounding works.
If the voice itself is briefly unavailable, the answer still appears in the on-screen transcript, so the call never dead-ends.
Every call also shows the AI badge in its header, and the audio carries a machine-readable AI marking. See AI disclosure.
When a call cannot start
If a call cannot start, the message says why. When a lens has used up its monthly call allowance, visitors see that clearly and are pointed to the text chat, which keeps working. Rapid repeated call attempts are asked to wait a minute. Only genuine connection problems show a “try again” message.
Availability
Voice calls are included on paid plans. What your plan includes, and your monthly call allowance, is shown in the app.
Try this
- Open your lens’s public page and tap Call.
- Ask a question, then interrupt the answer halfway. The lens stops and listens.
- Ask something you know is not in your material, and hear it decline honestly instead of guessing.