There's a specific sound a laptop makes when a sixteen-month-old finds the keyboard. It's not typing. It's percussion.
A very committed percussionist. And Daniel's household policy is to divert rather than confiscate, so the keyboard is exactly the thing Ezra gets drawn to, because it's the thing that responds.
Daniel's written in about this. Ezra's sixteen months old, endlessly curious, and the family's approach is to hand him things he can use instead of taking away things he can't. Daniel found a Linux tool that locks the keyboard so a baby can bash away safely, and that got him thinking bigger. He's got an old laptop lying around, and he's imagining dedicating it to Ezra. His name for it is EzraBuntu. Custom Linux setup, baby-friendly interactive apps, and maybe a better teething-meltdown fallback than YouTube Kids, which is currently the last resort when the teeth get brutal.
He's asking whether anyone has actually built a custom OS for very young children, what would go on it, and at what age any of it makes sense.
Here's the fun part. The hardest technical piece of his idea already exists, and it's been sitting in Debian stable for over a decade.
That's what I love about this prompt. Daniel's instinct, divert rather than restrict, has a direct software analogue. The dedicated session. Instead of childproofing your machine, you give the child a machine where nothing is off-limits. The machine itself becomes the safe thing. That's a design philosophy, not just an install.
And it's not a greenfield problem. The baby-keyboard category has multiple maintained tools in it right now. The hard part is curation and age-appropriateness, not the lock.
There's a real tension hiding in there too. Because when you go looking for a custom kid's OS, you find a graveyard. And when you go looking at what pediatric guidance says about a sixteen-month-old and a screen, you find it's not exactly cheering. So the episode writes itself: what exists technically, why the kid-distro genre died, what's age-appropriate at sixteen months versus three years, and whether any of this beats the YouTube Kids fallback.
Start with the tools, because the distinction Daniel needs is between two things that look identical from across the room.
Right, the category splits in two. Input lockers and baby mashers. An input locker is parent-side. It freezes the keyboard, sometimes the mouse too, so the parent can keep working or wipe the thing down or let a toddler watch a video without the video getting murdered by a stray palm on the spacebar. The screen doesn't change. It's a lock on your own machine.
And the baby masher is the one where the screen becomes the toy.
It grabs the input and replaces the display with something reactive. Keypress, color, picture, sound. That's what Daniel actually wants, and I'd bet the tool he found is from the first category, because those are the ones that show up first when you search "toddler-proof keyboard Linux." Hands Off is a good example of the locker type. It grabs input at the kernel level, using an evdev ioctl called EVIOCGRAB, which means no other application sees a single event while it's on. Works identically under X11 and Wayland because the grab happens below both of them. You switch it on with the numpad star and minus keys pressed together, and you unlock it by typing the word quit, which can't leak into any running app because nothing else is receiving keystrokes.
The numpad pair because they sit next to each other and nobody presses them by accident. Somebody thought about that.
Somebody with a toddler. Catlock does the same job for Wayland with Ctrl Alt K, and it lists its users honestly: a cat who walks over your keyboard, a young child, or, quote, an annoying person in your house who likes to prank you.
The annoying person is the underserved demographic.
Always has been. But none of those give Ezra anything to do. They just protect the parent's session. The masher is where it gets interesting, and there's really only one serious entry: BamBam.
Which has been in Debian for over a decade, which for a baby toy is a geological age.
Python and Pygame, GPLv3, and the author wrote it because Mac OS had a lovely program called AlphaBaby and he couldn't find a Linux equivalent. So he learned Python to build it. Current upstream is 1.4.2, and Debian stable carries 1.4.1, with a tracker note about the newer version that was updated literally two days ago. Somebody is still minding it. Install is one line, apt install bambam.
So the whole bottom layer of EzraBuntu is one package.
One package, plus one architectural decision. Here's the load-bearing insight, and it comes straight from BamBam's own README. Running it as an app inside your normal desktop is, in their words, not one hundred percent bulletproof, depending on your environment. Take care when leaving your child unattended. Instead, they recommend running it as a dedicated graphical session. You copy one desktop file into the X11 sessions directory, and then BamBam appears in the little gear menu at the login screen alongside your normal desktop. You pick BamBam, and the session is BamBam. The README puts it plainly: only the game is launched, and the user is logged out as soon as the game quits. Thanks to this, a child is not able to cause any damage even if he or she somehow manages to quit the game.
The machine has one door, and the door is the toy.
That's the entire security model, and it's a good one. There's no desktop underneath to escape to. Ezra can't reach your files because there is no your files. There's a picture of the design philosophy I was describing, right? Nothing is off-limits because nothing else exists.
And quitting is typing the word quit, which appears in the corner.
In English locales, yes. Which a sixteen-month-old is not going to manage. Though knowing Ezra's track record with hardware, I'd give it a year and a half before he's found some other exit. And there's curation built in, which matters for exactly Daniel's situation. Flags for sound blacklisting and image blacklisting, glob patterns, so if any of the bundled sounds or images are too loud or too startling for a particular kid, you can strip them out. There's even a deterministic mode so the same key always makes the same sound. Some children want randomness, some want the world to keep its promises.
Ezra strikes me as a randomness guy.
Most sixteen-month-olds are. There's also an extensions system, no programming needed, just media files and a YAML file. The one that ships plays recordings of American English letter pronunciations when you hit the corresponding key. So the thing quietly becomes a pre-literacy toy, key, letter, sound.
Teething meltdown, colorful noise, incidental phonics. That's a decent product for something you install in one line.
Now the harder question. Has anyone actually built the distro Daniel is imagining?
You mentioned a graveyard. I'm guessing there are headstones.
Several, and they're informative. DoudouLinux is the closest thing to a literal EzraBuntu. It targeted ages two to twelve, pitched itself as an environment as easy to use as a gaming console, kids can learn and discover without Dad and Mum always watching. It ran as an autorun CD. Last image updates were 2013. Last blog post, April 2015. Qimo was Ubuntu-based, ages three and up, retired in January 2016 with a graceful farewell note saying the need it filled was, quote, now done in much more effective and inventive ways. Edubuntu wound down from 2016. Sugar, from One Laptop per Child, is the survivor, but it targets literacy-emergent children, kids who are about to read, not toddlers.
So EzraBuntu would be resurrecting a dead genre.
You'd be reviving a category, not joining one. But the strange part is that the components never died. BamBam, GCompris, Tux Paint, all packaged, all current. The distros went away and the apps underneath stayed in Debian the whole time.
That's an odd shape for a collapse. If the need vanished, the apps should have vanished with it.
Which is the argument worth having. The best evidence is a father named Jimmy Sjölund, who actually tried this with his two-year-old daughter, DoudouLinux on a USB stick, and wrote about what happened. She'd been using an iPad at preschool. She tried a few games and then hit the wall. His words: she didn't understand the laptop's dinosaur method of swiping a touchpad and clicking a button to make things to happen.
Dinosaur is a brutal word for a touchpad.
Accurate, though. He also found the apps clunky, ugly, and uninspiring next to the App Store. And his conclusion years later was that the whole premise was wrong. There's no Windows or macOS for kids, he said, so why would we need separate distributions in the Linux world? On other platforms it's all about the applications and games. For children it is all about the apps and how good they are.
So his version is: the genre died because the premise was wrong, kids need apps, not operating systems.
That's his position, and it's defensible. Here's the counter I'd offer, and I want your read on it, because you're closer to the design question than I am. Nobody with design taste ever showed up. The genre didn't lose to better products, it lost to the first platform that paid actual designers. If someone with genuine craft built a kid Linux in 2026, is Sjölund's argument still true, or was it only true of the 2013 crop?
I think it's unfalsifiable until somebody tries, and Daniel has accidentally volunteered.
EzraBuntu, the live experiment.
One household, one subject, no control group. Science will be furious.
Science has survived worse. Here's what I'd actually put on it, and the age ladder matters, because this is where Daniel's "maybe he needs to be a little older" is doing real work. At sixteen months, the honest list is short. Cause and effect only. Keypress makes something happen. BamBam is the whole tier. Alongside it, Gamine, which is a mouse-based toy for very little kids, and KTuberling, the Potato Guy game, drag stickers onto a scene, but both of those assume pointing, so they're lap games with a parent steering. Around two, GCompris opens up. It's the flagship, ages two to ten, around a hundred and ninety activities now, current release 26.2 from September. Current. Actively maintained under KDE, fifty-plus languages. Around three, Tux Paint, which DoudouLinux's own guidance recommended from age three. Around four or five, Tux Typing, Tux of Math Command, the GCompris reading and arithmetic categories.
The evolution lever is the session, not the distro.
That's it exactly. Version one, BamBam is the only session on the login screen. Nothing to choose between. Version two, add a kiosk session with a few giant launcher icons. Version three, a real child account on a real desktop. The laptop grows up with him, and each stage is a config change, not a reinstall.
And the honest description of version one is one app.
One app. Because of the form-factor problem, which I think is the most underrated obstacle in the whole idea. A sixteen-month-old's motor skills map onto touch and whole-hand slapping. BamBam asks only for keypresses, so it works now. Everything past it assumes pointing, and pointing at a precise target is a skill that arrives later. Sjölund's daughter, iPad-fluent, could not map touch onto trackpad-and-click. She was two. Ezra's younger than that.
So the laptop is a two-year-old's toy that a sixteen-month-old can only bash.
For now. And honestly, bashing might be the point. He gets a keyboard that responds to exactly the input he's capable of, percussion, and the machine treats it as valid. That's the diversion philosophy working at the hardware level.
And that gets us to the question Daniel actually asked. Better than YouTube Kids?
And here the guidance cuts both ways, so I'll lay it out without flinching. The older AAP line was about the strictest thing in pediatrics: essentially no screens other than video chat for eighteen months and younger. Ezra's sixteen months. On that arithmetic, the laptop loses before the comparison even starts, and so does YouTube Kids.
The old rules don't have a winner in this fight.
Neither contestant passes the first round. But the AAP moved away from time-specific limits this year, toward what they call the five Cs. Child, Content, Calm, Crowding Out, Communication. Individualized rather than a number on a clock. And under that framing, the picture changes. A pediatrician named Tiffany Munzer, explaining the policy, named the mechanism directly: features like autoplay, endless scrolling and targeted ads aren't built with children's well-being in mind. Engagement-based design competing for attention and crowding out sleep, play and family.
BamBam has no autoplay. No ads. No scroll.
No anything. It's a box that lights up when you hit it. And crucially it's co-played, an adult on the lap, which is the Communication C. By the five Cs, a BamBam session may well score better than high-quality YouTube Kids watched alone, while still being screen time. Better on design, not exempt from limits. And I have to be honest, there's no head-to-head evidence comparing a curated Linux setup against YouTube Kids on any outcome. Nobody's studied it. The argument has to be made on design grounds, and the design grounds are actually pretty strong.
Medieval Pottage Facts. Many monasteries had an almonry, a gate where the poor were given bread and food. The gate stayed open. The poor still had to come to it.
The almonry model is roughly what the dedicated session is, now that I think about the architecture. One gate, one thing given out.
And no menu behind it. Which conveniently is EzraBuntu's whole design brief.
There's one more thing worth pulling up from the research, though, and it's the pediatric framing around what screens displace rather than what they contain. The AAP's current language on infants and toddlers is blunt, that screens carry more risks than benefits in those years, and that heavier viewing habits young predict more problematic media habits later. So even the friendly framework isn't a permission slip. The fair reading is that EzraBuntu is a better shape of screen for the moments when the family was going to use a screen anyway, and teething at midnight is that moment. It's not a reason to add a screen.
Which is Daniel's own framing, to be fair. He called YouTube Kids the last resort, not the curriculum.
He did. And swapping the last resort for something with no autoplay in it is a meaningful upgrade on the margin, because the autoplay is the whole trap. You hand over a five-minute video and the platform hands back forty more. BamBam hands over a keypress and the keypress ends when Ezra ends it. Nothing in it wants him.
That's a rare property in software for children.
Almost unheard of. AlphaBaby had it, BamBam inherited it, and everything commercial abandoned it, because a toy that doesn't want anything from the child can't be monetized through the child.
So the honest summary of the build is one package, one session, some blacklist flags, and a parent on the lap.
That's version one. The thing I keep noticing is that the total work is an afternoon, which is the opposite of what "custom OS" implies. Daniel has built entire voice pipelines that took longer than EzraBuntu would. We did an episode on him building custom speech recognition from scratch, and that instinct is the same one here, assembling from existing components rather than accepting the off-the-shelf product. Except this time the off-the-shelf component happens to be good.
Rare, in his builds.
He'd be the first to admit it. Usually by the third weekend.
The numpad star and minus, though. That's the part I keep coming back to.
What about it?
Herman, those keys are a centimeter apart. You're describing a lock you arm with the two most bashable keys on the keyboard. Ezra finds that chord in a week.
That's the parent-side tool, though, the lock for your own machine. And actually the activation chord is star and minus together, pressed as a pair. You're right that it's a fat-fingerable pair by design, because the design goal was that a parent can hit it fast one-handed. The security isn't the chord, it's that Ezra doesn't know typing quit unlocks it.
Yet.
Yet. Which is the entire toddler security model, everywhere, forever. Everything is safe until the toddler learns the thing.
Star and minus aren't adjacent on every numpad, Herman. On the one at the desk here there's a slash between them. I just checked.
Fair. The README's claim is about the layout they tested.
My sister's grandson does that. He's two now. Every second Sunday he comes over and works the keyboard at my desk like it owes him money. I unplugged it eventually. He was delighted, kept at it anyway, which tells you something about the whole category. I looked at getting him his own machine, a secondhand ThinkPad at the shop on Jaffa Road, hundred and forty shekels. Didn't buy it. The one key he always went for, the right shift, I popped it off and put it in the drawer. He never touched the others. Total concentration on one key, that boy. Typing quit to get out of something is no use to me anyway, these claws hit three letters at once.
You removed the one key he wanted and the siege ended?
He slapped where it had been for a while. Then the drawer was more interesting, because drawers open. I should've thought of that first.
The drawer is never the answer with a toddler. The drawer is the prize.
The single-key fixation is actually the interesting part, because that's real. At that age it's often not the variety they want, it's the one reliable cause. Which means EzraBuntu version one is honestly over-engineered. One key that reliably does something might be the whole product.
Daniel's still going to build the full thing.
Obviously. It's Daniel. There'll be a Git repo by Thursday.
So where does that leave the form-factor question, though. Because that's the one I'd actually put to him. The laptop might be a two-year-old's toy, and Ezra's sixteen months. Is the honest advice to wait, or is bashing itself the point?
I think the bashing is the point now, and the pointing arrives later on the same machine. The keyboard is the part of a laptop a baby can already use, which is a lucky accident of the form factor. A tablet would be better designed for him today, but a tablet is a consumption device designed to be wanted. The laptop is a tool that happens to have a bongos attachment.
And it's an old spare, so the cost of finding out is zero.
Which is the best possible price for a live experiment on whether kid Linux died because Linux is wrong for children or because nobody with taste showed up. One household, one subject.
Herman, you promised to be honest about the under-eighteens. This whole episode is a screen for a sixteen-month-old.
I noticed, yes. Being honest about it is the job. The honest version is that the guidance for his age is restrictive, full stop, and Daniel's not proposing more screen, he's proposing a better-shaped version of the screen that was already happening. The five Cs don't bless it. They just rank it.
That I can live with. The open question I'd leave him with is whether EzraBuntu v1 is one app on a laptop or one key on a keyboard. Hilbert's grandnephew would say the key.
And the bigger one underneath it, whether a genre that died for lack of design taste can be revived by one parent with an afternoon and apt. If it can, the graveyard was never about Linux.
Our producer is Hilbert Flumingtop, who keeps a shift key in a drawer for reasons that made sense at the time.
If this was your kind of episode, go back for episode two, Local STT For AMD GPU Owners; and episode eight, Building Your Own Whisper. This has been My Weird Prompts. You can find us at my weird prompts dot com.
We'll be back soon. See you then.
See you.