Yoggi
Safe AI chat for kids
129 followers
Safe AI chat for kids
129 followers
Yoggi is the AI assistant built for children aged 3–15. Age-adapted answers, live voice chat, image generation, strict content filtering, and parental controls. Free on iOS and Android.
For a young child using voice chat, one question will eventually come: "Yoggi, are you my friend?". How do you handle that kind of question, especially in terms of boundaries?
Congrats on the launch!
@jared_salois Thank you, and great question, it's actually one we thought hard about.
Our approach is simple: Yoggi can be warm and fun, but it's never designed to feel like a replacement for real relationships. So when a child asks "are you my friend?", it won't brush it off, but it will also gently steer the moment back toward the people who actually matter in that child's life: parents, siblings, friends at school... rather than positioning itself as one of them.
Practically, that means two things. It never pretends to remember things long-term or claims feelings it doesn't have (no false sense of continuity or intimacy). And if a child asks what it really is, it always says clearly that it's an AI, in a way that fits their age.
This isn't a scripted answer for that one question, it's a boundary that shapes how it talks in general, every day. For us, keeping kids engaged and having fun with Yoggi is great, but never at the cost of blurring that line with the real people around them.
@lucas_cyber_waffle The no false memory design makes sense. But for younger kids using Yoggi daily, could forgetting become confusing too? “But I told you about my dog yesterday.” How do you handle that moment?
@jared_salois Fair pushback, and yes, that moment does happen.
Within the same day, Yoggi does keep the thread of what was said earlier, so "the dog thing I mentioned an hour ago" works fine. What resets is the day-to-day continuity: past midnight, it's a fresh start. So if a child brings up "yesterday," Yoggi won't fake it or pretend to recall the details, but it also won't just flatly say "I don't remember." It stays warm about it and invites the child to tell it again, treating a repeated story as just as fun as the first time.
Right now that's a deliberate choice, not an oversight. The moment you let an AI companion "remember" things about a child across days, you're one step away from building a long-term profile of that child, and from a young child feeling like a real, ongoing relationship is forming with it. We're not comfortable with either of those.
That said, we are looking at a narrower version of this: not memory of the child, but the ability to pick back up a specific story or a learning topic from the day before, purely to keep something like an ongoing tale or a lesson coherent, not to make Yoggi sound like it "knows" the child better over time.
Honestly, this kind of trade-off, technical and philosophical at once, is what makes building this app so interesting.
the chat safety thread here is thorough, but the image generation feature seems like a separate risk surface nobody's asked about yet. text filtering and image filtering are pretty different problems, what stops a kid from prompting for something inappropriate through a drawing request that would never make it past the text filter as a sentence
@omri_ben_shoham1 You're right to flag it, image and text moderation are genuinely different problems and deserve different answers.
Here's how it works: kids never type a free-form prompt straight into the image generator. When a child asks for a drawing, our chat model first turns that request into a short, factual scene description (just subject, action, setting). That description then gets wrapped server-side in a fixed template we control, not the model, not the child, which enforces a "children's book illustration" style, explicitly excludes anything scary, violent, sad or ambiguous, and blocks realistic or copyrighted people/characters. On top of that, we rely on the image providers' own built-in safety filters (OpenAI and Google both reject unsafe generations before they're returned to us).
So the image path is actually more constrained than the chat path in one sense: there's no direct line from "kid's words" to "image prompt," there's always a moderation-shaped layer in between.
The honest gap: we don't currently run a post-generation visual check on the produced image itself, our safety relies on constraining the input rather than auditing the output. That's a fair thing to want, and it's on our radar.
Appreciate you pushing on this, it's exactly the kind of scrutiny we want on a kids' product.
One thing I'd love to see is a shared family dashboard where parents can review conversation topics their kids explored with Yoggi that week. It would make the parental controls feel less like a black box and help families start real conversations offline based on what their children are curious about.
@erturultaaz8ux Thanks so much for this suggestion, I completely agree that parental controls should help create real conversations, not just monitor activity.
Yoggi actually already includes a weekly email summary of the topics a child explored, as well as a daily recap directly inside the app. The goal is exactly what you described: helping parents understand their child’s interests, questions and concerns, so they can continue those conversations offline.
I’m really glad you mentioned it, as it confirms that this is an important part of the experience. I’d love to hear what information you would personally find most useful in such a dashboard.
Congrats on the launch! Age-adapting the AI's tone from a 3-year-old to a 15-year-old is a genuinely hard problem, most tools I've seen just pick one register and hope it works for everyone.
I'm curious how long a child's conversation transcripts get kept around for the weekly summaries and safety alerts, and who besides the parent can pull up that history once it's stored. That retention detail is the part I'd want spelled out for something built around kids.
@vollos Thank you, and I completely agree that retention should be clearly explained for a product built for children.
At the moment, conversation history is kept for three months. This allows parents to review previous conversations and gives Yoggi enough context to generate daily and weekly recaps. Through the product itself, that history is only available in the PIN-protected parent area; no other user can access it.
I’m also considering giving parents a very simple retention setting. They could keep the current three-month history, choose a shorter period, or opt out of long-term storage entirely. In that last case, conversations would be deleted after the current day, with the clear trade-off that there would be no history or weekly recap available.
I think parents should be able to choose the balance between visibility and data minimisation that feels right for their family, rather than having that decision imposed on them.
The daily memory reset is a thoughtful boundary. For the narrower story or learning continuity you’re considering, how would you prevent those saved threads from gradually becoming a profile of the child anyway?
@adityaharish2002 Fair concern, and it's the real risk with any form of memory.
If we ever add short term continuity (say, picking a bedtime story back up, or remembering what a child is currently learning), the intent is to keep it narrow on purpose: tied to a specific ongoing thread, not general facts about the child, on a short rolling window rather than piling up forever, and kept in plain, readable form rather than an opaque behavioral profile.
It would never feed mood, personality, or social inferences.
Whatever gets stored would also be visible to the parent in plain language, with a one tap way to clear it instantly, no hidden state the parent can't see or reset.
That's a design bar we take seriously, and we won't ship anything like this until we're confident it holds that line, functionality is only worth it if it doesn't cost the child's privacy.
hmmmm i am struggling to see the use case of this for kids. Unless its meant to teach kids how to use AI tech so it prepares them for the future.
@saderod That’s a fair question. Preparing children to use AI responsibly is definitely part of the idea, but it’s not the only use case.
Yoggi is designed as a safer, age-appropriate tool for everyday curiosity: asking questions, getting help understanding something, creating stories or images, practising a language, or simply exploring a topic they are interested in. For younger children, it can also be used entirely through voice.
The main difference is that it adapts its language and guidance to the child’s age, encourages them to think rather than simply giving answers, and keeps parents informed through recaps and safety alerts.
My goal is not to convince every child to use AI, but to offer a more suitable first experience for families who know their children will eventually encounter it anyway.
I’d also be genuinely interested to hear more about your perspective on AI and children. A different point of view could highlight a concern I haven’t considered yet, or even inspire a new feature or a better way for Yoggi to respond and behave.
You listed divorce as a topic a parent can flag, so it is in the content model but not the account model. There is one Parents' Corner, one PIN, one alert address. Split custody turns that into a choice nobody wants: share the PIN, and each parent can read a mood and topics digest covering the child's time in the other house, or keep it, and one parent sees nothing. Either way the child is the sensor. The families most likely to flag divorce are exactly the ones that lands on, so it seems worth deciding on purpose rather than letting the single-account default decide it.
@mamentumapp That’s a very good point, and honestly one I hadn’t considered. I wasn’t planning a split-custody account model, but you’re right that the current single-parent default could create real privacy and access issues. It’s something I need to think through deliberately rather than let the default decide. Thank you for raising it.