Most tutoring treats the symptom.
Agentdio finds the cause.
A reasoning-first AI tutor for primary, middle and high school. It doesn't just mark the answer wrong — it works out why your child got it wrong, traces the gap to its root, and teaches that instead.
No product yet · No charge · One email when there's something real to try
¾ + ⅙ = 4⁄10
A conventional system marks this wrong and serves another fraction question.
Illustrative of intended behaviour · Agentdio is in development
Why parents are stuck
Two bad options, and the same blind spot in both.
Private tutoring and cram schools sit at opposite ends of cost and attention. Neither of them actually diagnoses anything.
Private tutors — expensive, and quality is a lottery
You buy one person's attention by the hour, with no way to judge their diagnostic skill until you've spent months finding out. Scheduling is rigid. When they leave, everything they learned about your child leaves with them.
Cram schools — built for the average student
The pace is set for a class, not a child. Students who are behind fall further behind in silence; students who are ahead sit through material they already own. Reports measure attendance and scores, not understanding.
Neither one tells you why
A score of 62% tells you there is a problem. It doesn't tell you whether it's a concept gap, an arithmetic slip, a misread question, or a chapter from two years ago that never set. Without that, extra hours are just extra hours.
The premise
An inference engine, not a chatbot.
Plenty of tools will generate another practice question or explain a worked solution. Agentdio is being built around the step those tools skip.
Every answer a child gives — right or wrong — is evidence of what they believe. Read enough of it carefully and the mistake stops looking like a mistake and starts looking like a consequence. That is the part we are working on, and the part we will say the least about until it works.
We'll show you the rest when it's ready.
Where it fits
Compared with what you're using now.
An honest comparison, including the place where a human still wins.
| Agentdio | Private tutor | Cram school | |
|---|---|---|---|
| Pace | Set by the individual child | Set by the child | Set by the class |
| Availability | Any time, any day | Booked hours | Fixed timetable |
| Root-cause diagnosis | The core design goal | Depends entirely on the tutor | Rarely attempted |
| Consistency | Identical every session | Varies by day and by person | Varies by teacher |
| Memory of your child | Cumulative, never resets | Lost when the tutor changes | Effectively none |
| Visibility for parents | Which concept, why, what next | Verbal feedback | Scores and attendance |
| Cost structure | Subscription, not hourly | High, per hour | Moderate, per term |
| Human encouragement | Limited — the honest gap | Strong | Peer group helps |
By stage
The same engine, three different jobs.
Catch the gap before it compounds
Most secondary maths anxiety traces back to a handful of primary concepts that never properly set — place value, fractions, the meaning of the equals sign. Found early, they take days to fix. Found at fifteen, they take months.
Survive the jump in abstraction
This is where arithmetic becomes algebra and many capable children quietly stall. The failure is usually conceptual rather than effortful, which is exactly what extra practice hours fail to fix.
Spend revision time where it pays
With exams fixed and hours finite, the question is no longer what to study but what to study first. Agentdio is being built to rank weak topics by how much each one is actually costing.
Trust & safety
A product for children has to earn this first.
These commitments are being designed in from the start, not added after launch.
Parents can see everything
Every session, every inference, every change to the learning plan is visible to you in plain language. No black box — and nothing your child does with Agentdio is hidden from you.
Minimal data, clearly held
We intend to collect only what the tutoring itself requires, retain it for a stated period, and let you export or delete it on request — consistent with UK and EU rules on children's data.
It teaches — it doesn't do the homework
Agentdio is being designed to refuse to simply hand over answers. The point is that the child can do it next time without us.
It supplements teachers, it doesn't replace them
Classroom teachers do things no model can: notice, encourage, and know a child as a person. Agentdio is aimed at the diagnostic gap they don't have time to close for thirty students at once.
Who we are
We started at the dinner table.
We're a small group of developers. What brought us together wasn't an interest in education technology — it was watching what homework does to a household.
Homework becomes negotiation. Negotiation becomes argument. A parent who only wants to help ends up policing, and a child who is stuck learns to hide it. Nobody in that room is the problem, and no number of extra tutoring hours has ever fixed it.
Most of that friction comes from not knowing. Not knowing why the answer was wrong — so the only lever left is pressure.
Agentdio is what we're building to take the not-knowing away. We'd rather say little about it until it works.
Early access
We'll tell you when there's something real to try.
Agentdio is in development. If you'd like to be among the first families to test it — or you teach, tutor, or build in this space and want to talk — leave your email.
Opens your mail app addressed to [email protected] · Nothing is stored on this page
Questions
Straight answers.
Can I use it today?
No. Agentdio is in development and nothing is available to try yet. This page exists so interested parents and educators can be told first — and so we can hear what you'd actually want from it.
How is this different from asking a general AI chatbot for help?
A general chatbot answers the question in front of it. It has no model of your child, no record of what they got wrong last month, and no reason to withhold the answer when giving it away is the easier reply. Agentdio is being built around persistent diagnosis of one specific learner.
Which subjects will it cover?
Maths first, because misconceptions there are the most clearly traceable and the consequences of leaving them compound fastest. Sciences are the intended second step. We'd rather do one subject properly than four badly.
What will it cost?
Not decided. The intention is a subscription priced well below private tutoring, but we won't quote a number we can't yet stand behind.
Will it make mistakes?
Yes. Any system that infers will sometimes infer wrongly. That is precisely why parents see the reasoning rather than just the conclusion, and why the plan is re-tested rather than assumed correct.
Who is building it?
A small independent team of developers, several of us parents. We're keeping the details quiet while the work is unfinished — but the inbox is real, and it's one of us reading it, not a support queue.