For some reason the opening settings page made me think this would be someone who just told claude to make a monkey ball style game.. maybe from seeing too much of that on HN. forgive me for that, this is awesome.
As far as i can tell it's not even an emulator or a decompilation running in emscripten or anything like that, they remade the game in TypeScript. love stuff like this https://github.com/sndrec/WebMonkeyBall
I keep saying JS JIT + WebGL/WebGPU is fast enough for these kind of games, no need for the WebAssembly toolchains that are still a pain to use years later.
See PlayCanvas.
The whole GX code reminds me of the Gamecube API from the same name.
I don't see much of a reason to keep a copy of node_modules on the git repository considering they can be reinstalled for deployments and it is generally bad form.
sndrec (the author):
Thank you for this - I'm newbie at webdev so I wasn't sure what was and wasn't needed. I'll merge this soon.
In 2006 the iPhone was announced without an App Store and Apple’s party line was to just build/use web apps.
Fast forward to 2008 and the App Store is launched along with Super Monkey Ball – a day one app – the perfect game to demonstrate the power of a true native app that could _never_ be achieved on the web.
Fast forward to 2026 and after 15 years, the browser vendors have not yet provided anything remotely similar to RenderDoc, only SpectorJS survives, barely usable.
Also this game remains the exception, Infinity Blade was released in 2010 to show what iPhone could do with OpenGL ES 3.0, the base for WebGL 2.0, and yet most Web games are basically Flash like remakes.
I'm really at a loss to explain why there aren't more web games of this quality. It's totally feasible to make these, and yet they are so rare. I've ported a couple of games myself (https://thelongestyard.link/q3a-demo/, https://thelongestyard.link/cave-story/), but there ought to be hundreds like this.
I was in the market looking for some fun iOS games, things that I could play casually, pick up in a moment, load quickly, and not be burdened by the ridiculousness of modern gameplay and incentive mechanics. To my surprise, it was very hard. I couldn’t find anything. This is exactly what I’m looking for.
Apple has had some good chips, but is not a high priority for game developers. If you are on ARM, than the emulation performance hit will heavily limit what kind of games are playable.
The gyro permission request doesn't work on iOS since it's not tied to user input.
If you're feeling brave, you can paste this into your phone's javascript console to add a button that requests permission.
var b=document.createElement('button');
b.textContent='Gyro';
b.style='position:fixed;z-index:999';
b.onclick=()=>{DeviceOrientationEvent.requestPermission();b.remove()};
document.body.appendChild(b);
The GTA Vice City in browser was also really impressive, but it seems it has been taken down. How much of an advantage has AI got on decompilation projects? Complex assembly seems to be still done to some degree by hand these days (see - ffmpeg), and I wonder how big of a training set you could provide. I have wondered if it was possible to take the re3/reVC code and the assembly and use it for training data to get GTA San Andreas on macOS.
GTA Vice City and San Andreas were released on iOS more than a decade ago. I tried installing the mobile version on my Mac with Apple Silicon. It launches fine (if I remember right), it just needs an update for the controls to work, since it was made for touch. I haven’t tried hooking a gamepad to the Mac, maybe that would solve it.
It seems like Rockstar could make a relatively minor update to officially support macOS and sell a lot of additional copies. At this point, they could simply not support Intel Macs and I don’t think anyone would mind.
The experience of re3 and reVC were dramatically better than the new remastered versions, or the iOS sandbox version (which has no clean keyboard binds).
Well this was a fun way to see that Firefox on Linux finally fixed the shader cache being broken (at least for NixOS). This is great.
Though I gotta say, I am a little disappointed that there are no monkeys inside the balls. It's just a big ball, at least for me on Firefox and Chrome on NixOS.
Monkey Ball (without the Super iirc) was an arcade game initially. With a banana-shaped joystick and everything. Then SMB added some extra modes and came out as a release title for the Nintendo GameCube. It was probably intended as kind of a low-budget thing, but ended up being recognized as one of the best games for the system, especially early on.
If you haven't seen Smiling Friends, you're in for a treat. Zach Hadel is a genius.
The mix of 2D animation, 3D animation, claymation, stop motion, live action rotoscoping, and comping in guest animators like Joel Haver and David Post amazing. You know they appreciate the art form.
I've seen the multiple techniques becoming more popular (Wabie shorts), it really showcases a dominance of your craft when you can use multiple techniques instead of overrelying on a simple one. Great comedic/expressive technique as well.
I imagine how it would be with software, you have a whole ass huge website, but out of the blue you download a .jar for Nokia and you have to run it in a nokia or a very niche VM,(Or just in a JVM). Maybe to get a 6 digit verification code so you can log in to an account.
I tried to put on a movie while I was home for the holidays and my brother instantly complained that the drone shot made him motion sick. Was weird to me to hear that a stationary screen could upend someone's vestibular senses.
The author commented on their ko-fi: "there isn't much to say that would require a big writeup - a lot of the code is already reversed, and anything that's missing can be yoinked from ghidra decomp output and cleaned up, so it's just a matter of transpiling to a different language. plus much of the game's proprietary formats are thoroughly documented by the modding community. time consuming but quite easy if you're just patient haha"
I was a professional game dev working on 3D engines and the scale of this task is significant for a single person.
I started a task yesterday looking into decompiling Elite PC version which runs on a 256Kb XT. My goal is to replace the math and rendering with more modern algorithms which should be able to improve the framerate. Weirdly Gemini isn't bad at writing 16bit x86 CGA code. It had me set up with a dev environment in 2 mins using DOSBox.
Embarrassingly, I only ever knew this game as Neverball because there was a period when I would only play open source games and this, Xmoto, and tux racer.
So. this code was most likely generated with AI trained on community decompilation efforts, possibly without their knowledge. I know that the community has not yet reverse engineered custom model skinning for the game, so it does not appear here because it wasn't in the training data. Why would somebody who has supposedly already implemented billboard object support, or as the code calls it "flipbook objects", couldn't just stick a similar animated billboard texture inside the ball? probably because they have no clue how the code actually works or is structured.
It's genuinely impressive that generative AI has advanced to the point where this was possible, but it also feels like this was built backwards, extremely niche mechanics in the game are rendered nearly perfectly, where the base elements of the game had this been built from the ground up are implented wrongly or completely absent.
not trying to be negative, friend. I think clearly the author of this put a ton of effort into this project to make it functional, and I am acknowledging that. But, there are no disclosures about this anywhere from the author when so much of the work has been done by AI. I find this especially concerning when you realize the author is soliciting money for this project on X and other places.
For some reason the opening settings page made me think this would be someone who just told claude to make a monkey ball style game.. maybe from seeing too much of that on HN. forgive me for that, this is awesome.
As far as i can tell it's not even an emulator or a decompilation running in emscripten or anything like that, they remade the game in TypeScript. love stuff like this https://github.com/sndrec/WebMonkeyBall
The website credits include roles for "decompilation" and "porting". So I guess it was decompiled from the original binary and ported to TS.
Ah, this clarifies the GX references I mentioned on another comment.
I keep saying JS JIT + WebGL/WebGPU is fast enough for these kind of games, no need for the WebAssembly toolchains that are still a pain to use years later.
See PlayCanvas.
The whole GX code reminds me of the Gamecube API from the same name.
uh that code looks like claude to me
Pull Request: chore: remove node_modules
I don't see much of a reason to keep a copy of node_modules on the git repository considering they can be reinstalled for deployments and it is generally bad form.
sndrec (the author):
Thank you for this - I'm newbie at webdev so I wasn't sure what was and wasn't needed. I'll merge this soon.
Haha, almost certainly Claude
6 replies →
If anything, it seems that the author used GPT 5.2 (-codex) in Codex, which is actually far more capable at such work than Opus 4.5 in Claude Code.
Can you tell from the pixels?
7 replies →
In 2006 the iPhone was announced without an App Store and Apple’s party line was to just build/use web apps.
Fast forward to 2008 and the App Store is launched along with Super Monkey Ball – a day one app – the perfect game to demonstrate the power of a true native app that could _never_ be achieved on the web.
Fast forward to 2026 and after 15 years, the browser vendors have not yet provided anything remotely similar to RenderDoc, only SpectorJS survives, barely usable.
Also this game remains the exception, Infinity Blade was released in 2010 to show what iPhone could do with OpenGL ES 3.0, the base for WebGL 2.0, and yet most Web games are basically Flash like remakes.
https://en.wikipedia.org/wiki/Infinity_Blade
"Infinity Blade: iPhone Trailer", 15 years ago
https://www.youtube.com/watch?v=JDvPIhCd8N4
An easier answer to this is "why build web applications when you can build phone applications and monetize them?"
4 replies →
I'm really at a loss to explain why there aren't more web games of this quality. It's totally feasible to make these, and yet they are so rare. I've ported a couple of games myself (https://thelongestyard.link/q3a-demo/, https://thelongestyard.link/cave-story/), but there ought to be hundreds like this.
What was your process? Decompilation and porting to typescript? Crosscompilation to wasm?
I was in the market looking for some fun iOS games, things that I could play casually, pick up in a moment, load quickly, and not be burdened by the ridiculousness of modern gameplay and incentive mechanics. To my surprise, it was very hard. I couldn’t find anything. This is exactly what I’m looking for.
You may like hnarcade.com You’ll find a bunch of these types of games there
Apple has had some good chips, but is not a high priority for game developers. If you are on ARM, than the emulation performance hit will heavily limit what kind of games are playable.
https://github.com/86Box/86Box/releases
https://github.com/Moonif/MacBox/releases
Note, Steam will also natively run on many Apple ARM systems now, but again the compatibility of game titles will be sparse. Have fun =3
Looks and feels great, but is missing the monkey in the ball? :(
Super Ball!
Ball
The gyro permission request doesn't work on iOS since it's not tied to user input. If you're feeling brave, you can paste this into your phone's javascript console to add a button that requests permission.
var b=document.createElement('button'); b.textContent='Gyro'; b.style='position:fixed;z-index:999'; b.onclick=()=>{DeviceOrientationEvent.requestPermission();b.remove()}; document.body.appendChild(b);
The GTA Vice City in browser was also really impressive, but it seems it has been taken down. How much of an advantage has AI got on decompilation projects? Complex assembly seems to be still done to some degree by hand these days (see - ffmpeg), and I wonder how big of a training set you could provide. I have wondered if it was possible to take the re3/reVC code and the assembly and use it for training data to get GTA San Andreas on macOS.
GTA Vice City and San Andreas were released on iOS more than a decade ago. I tried installing the mobile version on my Mac with Apple Silicon. It launches fine (if I remember right), it just needs an update for the controls to work, since it was made for touch. I haven’t tried hooking a gamepad to the Mac, maybe that would solve it.
It seems like Rockstar could make a relatively minor update to officially support macOS and sell a lot of additional copies. At this point, they could simply not support Intel Macs and I don’t think anyone would mind.
The experience of re3 and reVC were dramatically better than the new remastered versions, or the iOS sandbox version (which has no clean keyboard binds).
You’re supposed to fork and/or save it.
I feel like it’s more sensitive than the original but this is a solid job.
I think it's because the Game Cube had a proportional joystick, and using a keyboard is 100%
For me, it's not really the same without the monkey yelling when you fall off the level. (example: https://www.youtube.com/watch?v=rIs7bCOCQj0 )
Adjust the input falloff thing and it is actually usable on mobile
Neverball is a similar game that's been open source for ages. It has a web based version too: https://play.neverball.org/
Seeing the translation from the decomp to ts is pretty interesting. Makes me wonder how one would actually write it these days
I don't want to spam comments, but I was absolutely hooked on SMB 1 and 2 for Gamecube and Wii.
This is incredible. Well done.
Well this was a fun way to see that Firefox on Linux finally fixed the shader cache being broken (at least for NixOS). This is great.
Though I gotta say, I am a little disappointed that there are no monkeys inside the balls. It's just a big ball, at least for me on Firefox and Chrome on NixOS.
[flagged]
Monkey Ball (without the Super iirc) was an arcade game initially. With a banana-shaped joystick and everything. Then SMB added some extra modes and came out as a release title for the Nintendo GameCube. It was probably intended as kind of a low-budget thing, but ended up being recognized as one of the best games for the system, especially early on.
This is immediately what my mind went to.
If you haven't seen Smiling Friends, you're in for a treat. Zach Hadel is a genius.
The mix of 2D animation, 3D animation, claymation, stop motion, live action rotoscoping, and comping in guest animators like Joel Haver and David Post amazing. You know they appreciate the art form.
You've probably already seen the gif of this scene: https://www.youtube.com/watch?v=5zxL77g1em4
I've seen the multiple techniques becoming more popular (Wabie shorts), it really showcases a dominance of your craft when you can use multiple techniques instead of overrelying on a simple one. Great comedic/expressive technique as well.
I imagine how it would be with software, you have a whole ass huge website, but out of the blue you download a .jar for Nokia and you have to run it in a nokia or a very niche VM,(Or just in a JVM). Maybe to get a 6 digit verification code so you can log in to an account.
This is essentially a port, it was done by transpiling a decompilation.
how does something like this work so well but scroll-based animations on mobile still choppy?
I tried to put on a movie while I was home for the holidays and my brother instantly complained that the drone shot made him motion sick. Was weird to me to hear that a stationary screen could upend someone's vestibular senses.
Seeing this, I understand.
Is there any info how this was done?
The author commented on their ko-fi: "there isn't much to say that would require a big writeup - a lot of the code is already reversed, and anything that's missing can be yoinked from ghidra decomp output and cleaned up, so it's just a matter of transpiling to a different language. plus much of the game's proprietary formats are thoroughly documented by the modding community. time consuming but quite easy if you're just patient haha"
I was a professional game dev working on 3D engines and the scale of this task is significant for a single person.
I started a task yesterday looking into decompiling Elite PC version which runs on a 256Kb XT. My goal is to replace the math and rendering with more modern algorithms which should be able to improve the framerate. Weirdly Gemini isn't bad at writing 16bit x86 CGA code. It had me set up with a dev environment in 2 mins using DOSBox.
1 reply →
Another poster commented that it was likely done with Claude.
Embarrassingly, I only ever knew this game as Neverball because there was a period when I would only play open source games and this, Xmoto, and tux racer.
So. this code was most likely generated with AI trained on community decompilation efforts, possibly without their knowledge. I know that the community has not yet reverse engineered custom model skinning for the game, so it does not appear here because it wasn't in the training data. Why would somebody who has supposedly already implemented billboard object support, or as the code calls it "flipbook objects", couldn't just stick a similar animated billboard texture inside the ball? probably because they have no clue how the code actually works or is structured.
It's genuinely impressive that generative AI has advanced to the point where this was possible, but it also feels like this was built backwards, extremely niche mechanics in the game are rendered nearly perfectly, where the base elements of the game had this been built from the ground up are implented wrongly or completely absent.
So. Certainly didn't expect this much toxicity and negative attitude this early in the morning.. I think that's enough HN for me today.
not trying to be negative, friend. I think clearly the author of this put a ton of effort into this project to make it functional, and I am acknowledging that. But, there are no disclosures about this anywhere from the author when so much of the work has been done by AI. I find this especially concerning when you realize the author is soliciting money for this project on X and other places.
I know HN is riding the LLM hype, but this codebase has no LLM in it.
Are gyro controls broken?
This would be prefect for iPhone gyro controls, but I’m not getting it to work.
Edit: never mind, the permissions are broken:
https://news.ycombinator.com/item?id=46791545
Cool. There's this from 13 years ago
https://experiments.withgoogle.com/world-wide-maze
(only a video is still up)
Oh man, this nerd sniped me. This brings me back to my childhood :)
It almost works on my phone but glitches out. Pixel 7, Chrome browser
Really blown away at how well this works on mobile. Awesome stuff
Man, this really takes me back. I loved that game!
I completed stage 10 and it shot me into space but the timer kept going and then counted it as a failure and restarted the map.
skill issue
This is awesome. Monkey Target was my favorite part - I hope that makes it in one day.
WHERE THE MONKE IN THE BALL?!
And the "Aaaah!" when you fell.
Thanks, got to stage 10 without messing with any of the settings on iPhone!
Is there a chance Super Monkey Ball will finally see a Dreamcast version?
Wouldn’t that be wild? I don’t know that the Naomi version runs on it yet, but that would probably be a decent place to start.
At least as a reference implementation for most of the engine!
forgot how much fun this game is. really takes me back
Welp, there goes my productivity for the next week
Hahahah no wayyy
I miss the "woop woop woop woop" noise you get when you move though, and it feels a little fast somehow?
Absolute cinema!
so good!
We're only 2 years away from "Claude, Make GTA VI!" /s
Looks fun but keyboard doesn't seem great for this, it feels like it needs an analog stick. Note I've never played the original.
Perf wise it seems bang on.
Amazing!