A tour your visitors
can talk to.
The biggest shift in interpretation isn’t the audio — it’s that visitors now expect to ask. Convo’s conversational layer lets a visitor pause the tour at any stop and put a question to the guide in voice, text, or by holding their phone up to the object. The answer comes back grounded in the curator’s source materials, in the visitor’s language. When the guide can’t ground an answer, it says so.
Book a demoA tour your visitors
can talk to.
A visitor scans the QR at the wall card, the tour opens in their browser — no download, no account. At any stop they can tap a button and ask the guide a question: why is his hand carved like that, what is this material, is this the artist’s late work, can you say that last part again, slower. The tour pauses, the guide answers, and the visitor chooses to keep going or to ask something else. The shift from listening to asking is small in description and large in feeling.
Grounded in your
reference materials.
When a visitor asks, the guide searches the reference set the curator uploaded — wall cards, catalog entries, exhibition essays, scholarly articles — and composes an answer from them, in the same voice as the rest of the tour. It does not improvise. It does not reach to the open internet. It does not pull what it learned in pre-training and frame it as your institution’s answer. The boundary is the institution’s canon.
Voice, text, and image —
in every language.
Voice taps the microphone and answers aloud; we shipped it on OpenAI’s Realtime API over WebRTC, so the latency feels like conversation, not a command line. Text is the unobtrusive mode for quiet galleries. Image lets a visitor hold the camera up to an object — useful for archaeological fragments, geology, or installations where the relationship between object and card isn’t one-to-one. All three modes share the same source set, the same refusal behavior, the same ten languages.
Declines when it can’t
ground an answer.
When the reference set doesn’t contain a basis for the answer, the guide declines and tells the visitor so. A confident invention from a museum tour is worse than an honest miss. The platform also filters obvious abuse before it reaches a response. The procurement-grade trust argument is simple: the guide is bounded by the same canon the curators built it on, and new canon ships when you ship it. Nothing else.
Every question becomes
editorial signal.
Every visitor interaction is logged in the admin. A curator can open a tour, go to any stop, and read the actual conversations that happened there — question, answer, time, language, and the source passages the model leaned on. Declined questions are particularly useful: they tell you what visitors wanted to know that the canon didn’t cover. Sometimes the fix is to upload one more essay; sometimes it’s to rewrite a stop. The Q&A log becomes a feedback channel back into the institution’s own interpretation.