← For Parents & Educators

How Our AI Works — and What It Will Never Do

We built the AI in Spark & Anvil apps to help your child think, never to do the thinking for them — and never to behave like a chatbot friend. Here is a plain account of the only kinds of AI a child actually meets, and the guarantees behind them. For what is kept on this device, see See what this app knows →.

What our AI actually does

These are the only kinds of AI a child ever meets in a Spark & Anvil app. Each one is tied to the activity on screen — a hint, a question, a character to teach, or a labelled AI-literacy lesson.

Socratic tutorA hint phrased as a question instead of an answer — for example, “What do you notice about the two shapes?” It withholds the answer on purpose so your child does the thinking.
Teach-the-Cast noviceA character your child TEACHES says what it is still figuring out — “So a bigger bottom number means a bigger piece?” — for your child to correct. Your child is the teacher.
Confidence-Ledger coachA gentle, proactive nudge after two misses in a row — “Let’s slow down on temperature for a moment.” It never gives the answer.
Game-review coachAfter a saved game ends, a calm question about a turning point — “What was better here?” — asked between games, never during play, and only if your child opens the review.
Jagged-AI helperIn the AI-literacy activities, a deliberately fallible “AI helper” answer your child decides whether to trust — clearly labelled as the lesson, so children learn WHEN to check a machine.
Legend-or-Record panelA pretend info-bot mixes real facts with popular myths; your child judges each claim against the record — teaching that a confident machine can still be wrong.

What our AI will never do

✗ Never chats freely with your childThere is no open chat box. Every AI line is a hint, a question, or a labelled lesson tied to the activity on screen — never an open-ended conversation.
✗ Never hands over the answerThe tutor is built to withhold the answer — it asks a question or gives a faded hint. A separate, plain (non-AI) engine is what checks whether an answer is right, so the AI is never the judge of correctness.
✗ Never keeps or sends your child’s dataLearning progress lives on your child’s own device. Where an activity needs a server (for example, generating a study pack), the request is processed and thrown away — nothing about your child is stored on our side or used to train anything.
✗ Never pretends to be a friend or companionThe AI is a helper tied to a task or a character your child teaches — never a persona that asks to be your child’s friend. (This is exactly the “AI companion for kids” pattern that new laws are moving to ban.)
✗ Never connects your child to a strangerThere are no accounts and no public matchmaking. Playing together uses a private, code-shared room with preset messages only — no free-text chat, no way for a stranger to reach your child.
✗ Never shows ads or sells anythingNo ads, no tracking pixels, no data sold. The only thing on the page is the learning activity.

Our promise

Everything here lives on this device only. Nothing is kept beyond your child’s learning progress, and nothing leaves the device.

See what’s on THIS device → What each kind of app keeps