How we use AI

Yes, there is AI in here. Here is exactly what it is allowed to do.

Being wary of AI near your child is a reasonable instinct, not a misunderstanding. So rather than reassure you in general terms, this page describes where the boundaries actually sit.

The short version

A language model does the explaining. It does not decide what your child is taught, what counts as correct, what gets remembered about them, or what your report says. Those decisions sit in the system around the model, and the model cannot reach them.

That distinction is the whole design. An AI chat box that answers homework questions is a different product with different failure modes.

What the model does

  • Explains a concept in words suited to your child's age and where they are up to.
  • Offers a hint, a worked first step, or a different angle when they are stuck.
  • Responds warmly and patiently, in a consistent voice.

What the model does not do

  • It does not choose the curriculum. Objectives and their order come from a reviewed structure, not from what the model thinks would be good next.
  • It does not decide what is correct. Right answers are defined by the content, not generated on the fly and trusted.
  • It does not write to your child's record freely. Anything durable goes through the system's own path, so a conversation cannot quietly rewrite what Abbi believes about your child.
  • It does not author your reports. Parent summaries are built from what actually happened in the session.
  • It does not override the safety rules. No amount of persuasion in a conversation changes them, because they do not live in the conversation.
Why it is built this way

Four separations that do the work.

  • Teaching is separate from truth

    The model can change how something is explained. It cannot change what the right answer is, or what your child needs to learn next.

  • Memory is separate from conversation

    What Abbi remembers is not whatever was said. Durable changes go through a governed path, so a persuasive exchange cannot rewrite the record.

  • Voice is separate from level

    How your child is spoken to and what they are working on are set independently — so harder material never means being talked to like an older child, and easier material never means being patronised.

  • Evidence is separate from confidence

    The model being sure is not evidence. Progress moves on what your child demonstrated, which is a different and much harder thing to fake.

The honest limitation

The hardest unsolved problem in a tutor like this is time and silence. When a child stops responding, the system genuinely does not know whether they are thinking hard, distracted, confused, tired, or have simply walked away.

A system that guesses will be confidently wrong about your child, and will then report that guess to you as if it were a fact. Abbi is built to start from unknown, learn what normal looks like for your particular child over time, and combine signals rather than treat one pause as proof of anything. Where it remains unsure, the report says so.

What this means in practice

You will sometimes read a summary that says Abbi is not certain. That is the system working, not failing. The alternative — smooth, confident reporting that outruns the evidence — is the thing we are specifically trying to avoid.

Your data

Your family's learning data is not sold, and it is not used to target advertising at you or your child. A fuller privacy statement is being prepared; until it is published, ask us anything directly at hello@abbi.world and you will get a straight answer.

Questions we have not answered here?

Ask directly. Anything about what Abbi remembers, what it does with it, and who can see it.

Start for free Book a walkthrough