Introducing 🚨 The ultra-fast Linux-like OS in the browser, it's a mapping of Linux APIs to the browser APIs. Now run that untrusted code. You might not need a cloud sandbox! Made for Agents and Humans!

This is very experimental right now! It implements a kernel, a shell with interpreter, lexer, parser, pipe, redirection. It has almost no boot time ⚡️ apart from core commands, it maps Node APIs like fs, http over browser network calls, a lot needs to be done here!

I vibe-coded it over the weekend, putting it out to hear feedback and if it solves your human or agent problems! Happy to hear your thoughts! Go & try

hey this is really cool! I've seen similar projects, maybe you can join forces or inspire from 🤙

yes, very inspiring projects

What caught my attention is the focus on Linux-like APIs in the browser, but I'm curious about the security implications when it comes to user permissions - do you consider how this impacts permission management within a single browser tab?

It's sandboxed, so it won't leak. haven't battle-tested that part a lot yet but doable to draw stronger boundaries (with WebWorkers maybe)

Please… I implore you… lead with the “this is cool but I vibe coded this” and “it’s not tested yet” and leave out the “it’s sandboxed so it won’t leak” and the “go run your untrusted code” rhetoric until you’ve done the first part. It’s not “battle testing” anyway… it’s just simply “have you thought of the majority of edge cases and tested those” I

@bygregorr I have tried to write it up everywhere. Sorry if it wasn't clear. Will add more warnings! Thank you

That's a pretty interesting project. Mapping linux commands to web based os is a great idea. I built an OS themed developer portfolio just last weekend as well. Here are the details:-

Looks great!!

Thank you 😄✌🏽

Have you seen this?

Yes, lifo isn't wasm, isn't a container. It's a mapping of Linux commands to Web APIs. So there is no booting

Super interesting direction. If you publish a simple benchmark matrix (boot time, file I/O, network calls, CPU-bound work) against common cloud sandboxes, adoption will likely jump fast for agent teams.

@FarnoodMassoudi That's great idea! will do

We spent decades trying to escape the limitations of local hardware only to realize the final boss of software is just recreating an entire operating system inside a browser tab

Need of the moment, underutilized compute is on users system

If this runs reliably on minimal compute, then it's THE BOMB for product demos and you're a fucking genius.

thanks Vittorio, a lot is possible! let's see

everyday we stray further from God

sandbox without VM?

yes

dammm!!!! crazy

Linux in the browser is kind of wild. This feels like dev tools evolving in real time.

Yes, just scratching the surface, a lot to come

damn so cool

wow, love the idea :) big linux fan, so clearly trying today interesting replacement to sandbox-approach, will explore let's launch you mate on if that helps btw.

thanks man! appreciate it, will launch in the coming days, making it a bit stable as of now.

amazing

JS is a virus. This is a cool project nevertheless.

Haha thanks

This is awesome 👏

Very nice work!

wow

Nice

Something like @ctorresai is building is also top and very interesting to be watched.

@ctorresai Interesting!

@ctorresai Yes. He is a hard working one... Tek is really good.

Everything what I need to running skills for agent.

For AI agents, this is a massive unlock—they can now execute untrusted code in a secure, ephemeral environment without the latency of a remote server. It’s the ultimate bridge between "web app" and "native OS" performance. Agents and humans finally have a playground that is as fast as local but as safe as the cloud.

Hey, really like the idea. it pairs perfectly with ship-safe, my free open-source scanner that checks AI output for security flaws before deployment

This looks great @sanketsahu 👏

Very cool stuff! I love the concepts of ephemeral boxes to run stuff on.

This is wild, a Linux style playground right in the browser so agents can get messy with untrusted code without begging the cloud for permission 😮

You got it right yes

This is so cool, Sanket!

Thanks Anurag! 🙌

The File Explorer design is great!

Vibecoded this part 😅

You could in theory delete most of actual Linux and just run a chrome container (or better) on top of a microkernel when your OS is sophisticated enough

yes, my idea was to inject on the existing runtime so that there is no booting needed

I find this fascinating

Just the starting point, we will have a package manager of wasm on top of this like apt-get

@jmitch This is interesting to have an ai agent running code on the client browser as its sandbox. Only issue is the client needs to be connected but I guess you can easily save the whole state and resume execution? Can wasm execution state be snapshotted?

@jmitch Exploring snapshotting currently, the fs can be easily copied and resumed. Thinking about processes currently.

This is a fascinating approach to sandboxing. How does the performance of mapped Linux APIs compare to traditional virtualization for compute-intensive agent tasks?

Haven't stress tested it but introducing a wasm binary for that would make it the same.

What the fuck is this abomination now

@chirag This is sick. 🚀

so instead of spinning up containers for agent sandboxing you just... run Linux in the browser? zero cold start, no infra costs. okay this is actually clever 👀

Sanket, this looks amazing! You really cooked.

Thanks Bilal 🙌 I'll make ruby work with wasm here! Hopefully