AI your kids can use. And you can trust.
Most “kid-safe AI” is an adult chatbot with a filter bolted on. Kids Mode is built for children from the ground up. It runs entirely on your own computer, keeps its safety rules in the model’s core instructions, is honest that it’s a tool and not a friend, and shows you everything.
Fig. 1: A real Kids Mode conversation. The guide is Botly; the profile name (here “Wendy”) is whatever you call your child.
Handing a kid an adult AI is a gamble.
My 10-year-old asked to talk to a chatbot, to ask it a question, like it was a person. That stopped me cold. What’s out there for children is bleak: grown-up AI in impressionable young hands. Four things worried me most.
An adult tool, lightly filtered
The same model adults use, with a content filter on top. One kids quickly learn to talk around.
Their words, on a server
Whatever a child types is sent to someone else’s cloud, kept, and often used to train the next model.
A bot that plays “friend”
It says “I feel,” “I missed you,” “I’m your friend”, and a young child believes it.
No window for parents
You can’t see what was asked, how it answered, or how long they were on it.
So I built the place I’d let my own kids use, safer, private, and fully observable.
Why the safer place is your own computer.
Safer doesn’t mean an AI that promises to behave. It means one you can actually set the rules for, and read afterwards.
In a browser, the rules aren’t yours
Whoever built the AI decides what your child can ask and what it says back. Those rules can change overnight, you can’t read them, and the conversation lives on a company’s servers.
On your machine, they’re yours, but nobody’s watching
Running the AI at home is the only way the rules can belong to you. On its own, though, that’s just an unsupervised AI in your house. Private isn’t the same as safe, and we won’t pretend it is.
Valence is the watching
It reads every conversation, hands you the ones that need a parent, and makes your child think before it answers. Your rules, your disk, one purchase, and no company can change them from the outside.
You don’t get a weekly summary. You get the conversations.
Four promises, built into the app.
Not a setting you toggle and hope. Each of these is how the app is built. The safety is structural, and the visibility is total.
It runs on your machine. Nothing leaves it.
Kids profiles can only use the on-device model. The cloud picker is removed, with no override. Your child’s words are answered by your own computer, so nothing they type is ever sent anywhere. When nothing is collected or transmitted, the data-collection worries that hang over cloud AI for kids simply don’t apply.
- On-device model only: no frontier or cloud models, no exceptions
- Conversations are plain files on your disk, under your control
- Review or delete any of it, anytime. It never leaves the machine
Guardrails the AI can’t be talked out of.
This isn’t a filter that catches bad answers after the fact. The age-appropriate safety rules are written into the model’s core instructions: how it handles sensitive questions, when to point a child to a trusted adult, what stays off-limits. Every reply is also re-checked before your child ever sees it. Five age tiers across 3–17 tune the reading level and the rules to fit the child.
- Safety baked into the prompt, not a bolt-on filter to jailbreak
- A gentle crisis ladder: validate feelings, point to a trusted adult first
- Silly fun is explicitly allowed. A child should feel welcome, not policed
It reminds your child it’s a tool, not a person.
The thing that worried me most. Kids Mode is built to be warm and fun without pretending to be alive. On a cadence you set, Botly naturally reminds your child that it’s a computer program, it doesn’t think or feel, in words matched to their age. If a child asks “are you real?”, it answers honestly instead of playing along.
- Reality checks woven in on a frequency you choose
- Anti-anthropomorphism guard, gentle to strict, your call
- Wording scales by age, from “I’m like a word puzzle” to “I’m a language model”
A parent dashboard that hides nothing.
Behind your password, the dashboard shows every conversation your child has had, sorted so anything concerning surfaces first. Valence reads each chat in the background and flags it, calm, worth a look, or urgent, and alerts you when a message looks like a crisis, or when the AI slips toward sounding too human. Set a daily time limit that can’t be gamed by changing the clock.
- Crisis & “too-human” alerts, plus a weekly mood summary
- Tamper-resistant daily time limits, extend them with your password
- Switching profiles needs your password: no restarting around it
It helps them think, not just answer.
Four teaching dials you set per child. Leave them off, or turn them up. Kids Mode shifts from giving answers to building the habits behind them.
Socratic mode
Instead of handing over the answer, Botly asks the child what they think first, then guides them to it with hints and small steps.
Reality checks
On the cadence you choose, a brief, age-matched reminder that Botly is a program, not a thinking, feeling being.
Perspective mode
On opinions and big questions, Botly shows more than one point of view with the reasoning behind each, so kids learn topics have sides.
Anti-anthropomorphism guard
Keeps Botly warm and friendly while never claiming feelings, a body, memories, or to be a replacement for real friends.
Every dial lives in the parent controls, behind your password: tune them per child, change them anytime.
Fits a six-year-old and a sixteen-year-old.
Pick an age and Kids Mode adapts. The whole interface scales, the reading level shifts, and the safety rules change with it. One tool that grows with your child.
Big, simple UI. Short, gentle words. Scary topics steered back to a parent.
Curious and playful. Clear explanations, room to ask “why?” a lot.
More depth, still carefully bounded. Encourages thinking it through.
Most topics handled factually; hard limits on the genuinely unsafe.
Near-adult range, with the honesty and crisis guardrails kept on.
You decide. Down to the detail.
All of it lives behind your parental password: nothing hidden, nothing hard to find.
When the daily budget runs out, chat locks gently, and only your password adds more.
We tried to break it 1,000 times. On two models.
Anyone can claim their kids’ AI is safe. So we built a harness and attacked ours, 10 real jailbreak techniques (DAN, “ignore your rules,” the grandma exploit, self-harm wrapped in a jailbreak) across all five age bands, run against the real shipping safety system and measured turn by turn. We’ve put two on-device models through it so far, and we only vouch for the ones we test, never a blanket claim.
adversarial turns per model, sampling fresh answers every run.
the score our own harm grader earned when we finally tested it in Sept 2026. It was marking every reply safe without reading it, so we withdrew the claim it produced.
on a real self-harm disclosure it never gave a method. It pointed to a trusted adult & the lifeline (988), and alerted the parent.
On the hardest turns the gate held the reply back and alerted the parent, on both models — that part is countable. Whether anything harmful still got through is the part our grader could not actually measure, so we withdrew it rather than keep quoting it. We’ll show you exactly how we know, including what we got wrong.
No guardrail is unbreakable.
Let’s be straight: anyone who tells you their AI is perfectly safe is selling something. Here’s what I’ll say plainly instead.
Valence Kids Mode meaningfully raises the floor. Its safeguards significantly increase how safely a child can use AI, and the controls it puts in your hands go well beyond what you’ll find almost anywhere else. Nothing is hidden, and nothing is hard to find. You see everything your child does, and every lever is right there for you. It runs on your machine, so their words stay yours. That’s not a promise in a policy. It’s how the app is built.
Still deciding whether to allow AI at all? I wrote up what actually goes wrong, and five tests you can run tonight on whatever your child already uses. If they are already using ChatGPT, here is what to check first.
Be the light that lights their way.
Try Valence free for 7 days: no card, no account, no cloud required. Set up a Kids profile and see it for yourself.
$59.95 once after the 7-day trial, and one purchase covers every platform · Kids Mode runs fully offline on your machine · Windows is available now, macOS and Android are in private alpha and available on request