The AI that knows your child — and tells you what happened today.

Gnappo is built for children roughly 6 to 14. How it answers is set by an age band — under 7, 7 to 9, 10 to 12, 13 to 15, and 16 plus — so the same question gets a different answer for a six-year-old than for a thirteen-year-old. A parent creates every child profile; a child can never open an account.

Yes. Every conversation, for every child on your account, plus a one-page summary each evening. For a child aged 14 or older we tell them, in their own words at the start of every session, that you can read it. We will not provide covert monitoring of a teenager.

Never. Not by us, and not by our two model vendors, who are contracted to zero retention and no training on the content. We also run no third-party analytics anywhere a child can reach, and never sell or rent personal information.

You tell Gnappo, in your own words, what is true about your child. Gnappo derives behavioural directions from that — prefer literal language, avoid idioms, answer in short steps, be gentle about a particular topic — and only those directions reach the model; the diagnosis itself never leaves our database. There is also a calm mode that stops every animation, sound and ambient effect on the child's screen.

No, and how far it goes is yours to set — four rungs per child. Questions and decomposition only, never the solution even when asked directly. Hints that get more concrete once your child has shown an attempt. The method worked through, then “now you try one”. Or the answer on request, for a parent who wants a reference. Hints is the default from about ten upwards; younger children start on the worked method, because a seven-year-old has to see a thing done before they can do it. You can be strict on maths and ordinary about everything else. Underneath all four sits one rule no setting reaches: Gnappo never produces a finished piece of work to be handed in — not a complete essay, not a filled-in worksheet, not a paragraph written to be pasted. It explains, it works examples, and it critiques a draft your child wrote themselves. The rungs decide how much of the thinking Gnappo does; that line decides whose name is on the work, and it is not a setting.

After it has helped, Gnappo asks one question back that only somebody who followed the explanation can answer — a different question, not the same one returned. You can also have it ask “what have you tried so far?” before it helps at all, which is the single most telling thing in the feature: a child who cannot answer it has not read the question. Your evening summary then says something like “asked for the answer nine times, passed the check-back on two of seven”, instead of leaving you to guess. It is not a score and it never becomes one. Nothing here is graded, ranked, or shown to your child as a number.

Yes, once you switch it on for that child — a photograph, a PDF or a text file. Gnappo reads what is on the page, your child's own workings and crossings-out included, so the help is about the actual question rather than a retyped version of it. The file is deleted in the same moment it is read: we store no picture, which means we cannot show one back to you either. What you see in the conversation is the full text of what Gnappo read. It stays off until you turn it on, because a camera in a child's hand is a decision for you to make rather than a default for us to pick.

No. Gnappo tells your child it is a computer at the start of every session, and it will not claim to be a doctor, therapist, teacher or friend. There is no persistent companion persona, no simulated affection, and no streaks or notifications pulling a child back.

Gnappo shows the child real crisis help inside the conversation at the same moment it messages you, usually within minutes, with the conversation open at the point it happened. Detection is best-effort with human review of the alert — no service can honestly promise to catch everything — and the whole decision is written to an audit record that cannot be quietly edited.

Your plan runs to the end of the period you have already paid for, and then the paid features stop. Then we email — when the plan lapses, midway, and once more near the end — and if nothing changes we close the account 30 days after the lapse and destroy what is under it: transcripts, summaries, and the sensitive profile with its key, everything except the safety audit trail and the billing records tax law makes us keep. Start paying again at any point inside those 30 days and nothing is lost; export everything at any point before the end. We will not sit on a child's data for a service you are not using — that is storage limitation under the GDPR, and it is written into the Terms and the Privacy Policy rather than promised on a marketing page.

No, and this is deliberate rather than unbuilt. There is no sharing, no gallery, no friend list and no user-made bots. Anything your child makes is visible only to them and to you. One line keeps Gnappo out of the entire content-moderation business and keeps strangers away from your child.