What happens with your child’s data
Last updated 2026-10-13.
What’s collected
Everything your child says to the tutor and everything it says back. If they use the microphone, their speech is converted to text — we do not keep the audio.
Where it lives
A server in Mumbai (AWS, ap-south-1) that we rent and control directly, in a private account. Not a third-party product.
Who can read it
You, and the person running this project. That’s worth being plain about: as administrator, every conversation is readable — we’d rather say so now than have you discover it later. It is only read to fix problems, never for any other reason.
Where it’s processed
Your child’s words go to Anthropic’s API to generate the tutor’s replies. Anthropic keeps what it receives for up to 30 days and then deletes it (longer only if a conversation breaks its usage policy), and does not train models on it. If they use the microphone, their voice audio also passes through two more providers — Deepgram, to turn their speech into text, and Sarvam, both for that and for the tutor’s own voice replies. Anthropic and Deepgram process outside India, so their words — and their voice, in the moment it’s spoken — leave the country, even though everything we store is on the Mumbai server described above. Sarvam is an Indian company, but we have not confirmed that all of its processing stays in India, so treat it as possibly outside too. Deepgram and Sarvam each keep what they receive under their own policies, which we do not control; ask us for the current details.
Checking the tutor
After each session, the conversation is read once more by Anthropic’s model, in a separate request, to check the tutor’s teaching against our rules: for example, that it never handed over an answer and that it praised thinking rather than being right. This grades the tutor, never your child. It does not change what the tutor knows about your child or any score, your child never sees it, and it is kept with the session and deleted with it. It goes to the same provider, under the same terms, as the replies above.
Account emails (verifying your address, resetting a password, and notes about their sessions) are sent through Resend, an email service outside India. It sees your email address and the message, which for a session note includes your child’s first name and a short summary.
What it remembers
Between sessions the tutor keeps what your child said and did, dated and in their own words where possible, the things they are into, and the ideas they worked out themselves. You can read all of it, with where it came from, on the “What Jigyasu knows” page in your dashboard, and remove anything. What you remove is not used again. A pattern is only noted after the same kind of slip in two separate sessions, and it is treated as a guess to test, never as a fact about them. Older notes count for less over time.
Other children
What it learns about your child stays with your family. It is not used to teach any other child. If we ever want to learn from many children together, we will ask you separately first, and you can say no or change your mind later.
Serious moments
If your child asks about something frightening, or the tutor sees real distress, it makes a note for you. A person looks at a serious note before it reaches you, so that you are not alarmed by a mistake. It reaches you within 12 hours at the latest, sooner if it looks urgent. You can always read the full conversation, which is never held back.
What we’ll never do
Show your child’s conversations to anyone else, use any of it commercially, or keep it if you’d rather we didn’t.
Deleting it
Delete your child’s record, or your whole account, yourself from “Account” in your dashboard, or ask us from “Get in touch” or directly and we will do it. Either way it goes with everything in it, permanently. No questions, no exit interview. It leaves the live system immediately. Encrypted backups taken for disaster recovery may still hold a copy, and every backup is deleted automatically within 30 days. If we ever had to restore one before then, we would delete them again before anything else.
The honest warning
It is an AI and it will occasionally say something wrong with total confidence. It’s built to flag its own uncertain claims for exactly this reason, but it isn’t perfect. If your child asks it something heavy — and children do ask about death, illness, and frightening things — it will answer honestly and gently, and flag it so you know to follow up. It will not dodge them, and it will not hide it from you.
Anything here you’d like changed, tell us and we will.