Privacy
Written from the deployed system on 20 September 2026. It describes what the code actually does today. It has not been reviewed by a lawyer, and it is not a contract.
The short version. There is no account and no sign-up, so we never ask for a name, an email address, a phone number or a school. We do not sell anything and we show no advertisements. We do not keep your voice recordings or your photographs. We do keep the text of questions and answers, to find the kind of mistakes this page exists to be honest about.
What happens when you ask a question
Your question — typed, or transcribed from your voice — is sent to our server with your class number and the language you are using. The server looks for passages from the supported textbooks two ways: by the words you used, which happens entirely on our own server, and by meaning, which sends the question to OpenAI to be turned into a numeric fingerprint. That second step happens on every question, whether you typed it or spoke it. The passages found and your question then go to Anthropic's Claude model, and every sentence of the reply is checked against a passage before it is spoken.
Corrected 20 September 2026. An earlier version of this page implied the question only left our server when you used your voice or the camera. That was wrong: the meaning-based search sends every question, including a typed one.
Companies that see part of a request
| Company | What reaches them | When |
|---|---|---|
| Anthropic | Your question, the textbook passages we selected, and recent turns of the conversation | Every answer |
| OpenAI | Every question you ask, typed or spoken, so it can be matched against the textbooks by meaning as well as by words. Also your voice recording, when you speak, so it can be turned into text; and a photograph, when you take one, so the page or room can be described. | The question on every turn; the recording only when you speak; the photograph only when you take one |
| Google Cloud | Your voice recording, if OpenAI's transcription is unavailable; the text Suno Dost is about to speak, for the voice | Fallback, and speech output |
| ElevenLabs | The text Suno Dost is about to speak | Speech output, when configured |
| Cloudflare (R2) and Vercel | Hosting and storage — see below | Always |
What we store, and where
On our storage (Cloudflare R2)
- A record of each answer. Time, the first 300 characters of the question, the language, the class number, which chapter answered it, the model name, and the first 500 characters of what was said. No name, no account, no internet address.
- Answers, cached. Keyed by the question and the passages used, so a question a classroom repeats is answered without paying for it again.
- Spoken audio, cached. Keyed by a hash of the sentence, so a sentence already spoken is not re-synthesised.
- Wrong-answer reports. When a student says an answer was wrong, we store the question, what was said, the language and the reason.
An open item, stated plainly: we have not yet configured an automatic deletion schedule for these records. They stay until we delete them. If you want something removed, email us and we will delete it.
On your own device
Your class, your language, which chapter you were in, where you had got to in a lesson, your speed setting and your day-streak are saved in the browser on your own device. They never reach us as a profile. You can clear all of it yourself: open the Study menu in the tutor and choose New learner, which is also what to use before handing a shared school phone to the next student.
What we do not keep
- Voice recordings. Audio is passed straight to the transcription service and discarded. Only the text survives.
- Photographs. A photographed page or room is passed to the description service and discarded. It is never written to disk.
- Faces or identity. Suno Dost is instructed never to identify a person in a photograph, and no face data is stored anywhere.
Your internet address is held in memory for the day, only to count requests against the daily free limit. It is not written to storage and it is not attached to your questions.
Before the microphone or camera opens
Suno Dost tells you, aloud, just before your phone asks for microphone permission, so a silent permission box never appears without warning. The green listening light is on only while it is listening, and it says so. You can always type instead: the text box is at the bottom of the tutor, and typing never opens the microphone.
The camera takes a photo only when you ask for one, and you can cancel. Surroundings description is for standing still and is never walking guidance — the cane stays the safety layer.
Children
Most of our users are school students. That is why there is no account: we ask for nothing that could identify a child, and the only thing resembling personal information we keep is the text of the question itself, which is why we cut it to 300 characters and why we ask students not to type personal details. If a child's question did contain something personal, email us and we will delete that record.
This design aims at India's Digital Personal Data Protection Act by collecting as little as possible rather than by asking for consent. We have not had that assessment reviewed independently, and we will say so here when we have.
Asking us anything, or asking us to delete something
Email shekhar.sattiraju@gmail.com. If your mail app is not set up, the address is shekhar.sattiraju@gmail.com. A person reads it.
Changes
When what the software does changes, this page changes with it, and the date at the top moves. We will not quietly widen what we collect.