TL;DR: This is a proxy server that routes HTTP requests from a unique subdomain (e.g. http://m7rp3u2ntr9t3xr.browserver.org) to a browser that's visiting browserver.org. The JavaScript in the browser then handles the HTTP response. This is possible thanks to a persistent connection between the browser and browserver.org, via some "websocket-like" protocol.

The title "node.js HTTP server, in your browser" is because the HTTP handler in the client (browser) has a Node-like API, see:

https://github.com/jed/browserver-client#example

Thank you for explaining this much better than I did!

i get a gateway error

Hey all,

I put browserver together in my free time over the last week. It's still a bit of a toy, but I think it's useful to explore ideas around extending webhooks all the way to the browser, to simplify our increasingly complicated web architectures.

It's already late here in Tokyo, but I'll be up for a bit if anyone wants to chat/brainstorm about approaches like this (and also to make sure the server stays up).

I'm slightly disappointed that it's implemented in more than 140 bytes.

Keep on being awesome. :)

Simple statement: stop being so cool.

On a related note, keep at it. I hope you go places.

And where/how did you get the idea?

Things like this make me wonder why I'm not having fun creating neat (and useful!) things.

Hey! I am doing something very similar with PageKite (http://pagekite.net/), although the server does not run in a browser.

I've decided to try to get authors of similar tools together on a mailing list to discuss whether we can agree on a common protocol so our clients/servers (connectors/relays, backends/frontends) can be interoperable. Join us: http://mailman.klaki.net/mailman/listinfo/revprotun

I had this idea, too, the killer application is actually demos and talks, I think. The speaker sets up their browser server, asks the audience to hit a URL, and then demonstrates in real time aspects of the system.

It's also a very neat way to bring WebKit Inspector to bear on server style code. Glad you did it! I presume that you have a wildcard subdomain at the DNS level and then some Node parsing magic to correctly proxy. Why subdomains though? Why not just a tag like browserver.com/29adfija02 ?

I chose hosts over paths for disambiguation because I want the request passed to the proxy to be as close as possible to that passed to the browser.

Also you get same-origin protection, FWIW.

This does sound sort of interesting, but after thinking about it I don't get why this is cool. Can someone explain a practical use case?

How is it different, in practice, from say running a Node.JS on localhost and using localtunnel?

Simple question: how the fuck?

Basically a web browser reaches the Node HTTP server and gets assigned a throwaway hostname on the Node server. The browser and the server have a bidirectional connection (probably WebSocket via Socket.io or something). The throwaway hostname can be queried, via an HTTP API, and will, upon request, query the web browser through the bidirectional connection then will relay the results back.

Yeah, it's not quite a HTTP server, more of an API-driven backend to a HTTP server. Still neat.

mind = blown

I find this title rather misleading. There is no node.js HTTP server in your browser. Meaning, you cannot do anything that is part of Node.js in the browser "server"

The browser is still running its original js engine with none of the Node goodies. It merely receives proxied requests, does some stuff to it, then sends it back to the proxy, which returns it to the client.

    It merely receives proxied requests, does some stuff to it, then sends it back to the proxy, which returns it to the client.

You just described the majority of the web servers that we web developers deal with. Sure, it's a proxy, but the fact that the machine responding is connected via WebSocket and not a local network doesn't make it any less of a server, in my opinion.

The title might be a little too ambitious, but it is partially justified as the HTTP handler in the client (browser) has a Node-like API, see:

https://github.com/jed/browserver-client#example

Nice attempt. But your title is misleading.

You just listen for a URL and forward it to the browser with socket.io (Pusher, Pubnub are some commercial services for this)

If I can change some content in your webpage by just curling localhost, I can judge the title you put :)

So, it's not a new take on POW (plain old webserver): https://addons.mozilla.org/en-us/firefox/addon/pow-plain-old... which apparently have been abandoned, and could use a fork (source at: http://davidkellogg.com/pow/ )?

Apparently there's a newer variation at:

  https://addons.mozilla.org/en-US/firefox/addon/browser-server/

and of course there's Opera's Unity.

I do get that the idea is that it "just" uses javascript (the server appears to be down, so I'm actually guessing) -- but when such projects already exist, I kind of expected more...

I thing they use some kind of native addon (I don't know how FF addon works) to do this.

Unfortunately Chrome's native client(NaCl) does not support TCP yet! So I think something like this impossible in chrome? Any ideas?

POW is presumably written in JavaScript using Mozilla's asynchronous socket APIs. I've written a similar HTTP server before.

Chrome has a JS socket API for extensions as well, but it doesn't currently support TCP listening.

Great. Thanks for the info.

So he still needs to host something that forwards information to the browser? Shucks.

Yes, he needs to have a Node/Socket.io server somewhere.

Indeed, this isn't entirely original, and I know services like this exist. To me, this is what PUSHER's API _should_ have been.

I don't really see why this is special. It's just a websocket connection that happens to show every request made to the server. Your browser is not a server, no matter which way you cut it.

Sure it is. Yes, the web socket connection is passing through requests to the browser-server, but that browser-server is the one responding.

I think there's a ton of unrealized potential in this kind of thinking. Although people will say 'But that was done by X a year ago!' I see that as evidence that this is gaining momentum.

Now, what's the project that will fully take advantage of this kind of thing?

A lot of times it is about coming up with the right terminology and positioning of an idea for it to "take off". This is a really nice approach in that it utilizes concepts/a paradigm we are already comfortable with.

Yeah, agreed. Seems like one of the best things about this is that it positions a (potentially) big idea in a way which makes it easily understandable and fairly everyday.

I mean yes, you never know what it will be used for. It may not be anything by itself, but combined with a dozen other things it may be critical for something clever. That's what I love about the internet - it's a massive jigsaw puzzle.

My hope is an answer to this problem that is so easy that my computer-phobic grandmother can do it: https://xkcd.com/949/

I think the Opera Unity approach is probably the best for this; perhaps such peernets will have their day.

I see the Unity approach being popularise via social networking (https://alicious.com/opera-about-to-change-the-world/) enabling users to retain control over their data and shop for suppliers to present a front-end to unify the data-streams of their associates.

Weinre is a project that, if my understanding of it is correct, leverages the same basic idea. It's used to remotely debug mobile browser issues and also uses Node.

https://github.com/apache/incubator-cordova-weinre

http://www.youtube.com/watch?v=4nL6xey13fE

...I'm looking forward to WebRTC, which should make this obsolete, right?

http://www.webrtc.org/

That said, this is cool!

Webrtc is about realtime communications. This is about exposing webhooks clientside.

It also allows direct client-to-client communications.

not yet arbitrary data. The protocol is also really hairy, I recommend using https://github.com/webRTC/webRTC.io

And back.

Kind of. Servers are 502ing.

Just add a text area and let people share some html. Now that would be cool!

Change some HTML in that text area and refresh automatically in all listeners.

That would be an interesting toy to play for a while.

Anyone know if browserver.org is self hosting (i.e. running on some guys desktop/laptop in a browser window)? Site is down at the moment and I'm wondering if this is why.

Ha, tinco beat you to that joke. The server is the smallest instance that Joyent offers right now... hopefully it'll hang in there during the HN rush.

Would that happen to be one of the free http://no.de instances? Which if I recall correctly come with 128mb ram on their custom solaris servers... I always wondered what kind of load those small configurations using NodeJS while handling something non-trivial could do.

Could you share some traffic/load numbers from getting HN'ed?

Hard to tell right now... I started at Joyent's smallest node.js offering (1GB) and then ramped it up to 8GB when the load hit. The server is crashing a lot, but it's not very easy for me to debug right now.

Let me get back to you once the dust settles, deal?

Wow must of took a real hammering from being on the number spot. Goodluck!

Github has a good readme if this site doesn't stay up: https://github.com/jed/browserver-node

Good point. I'll get to githubbing the browserver.org site itself now.

Okay, it's here:

https://github.com/jed/browserver.org

Ok so I can access it either, but:

Loading the app... If this message doesn't go away within 10 seconds, it means that the server crashed under heavy load. Please refresh mercilessly.

:)

Mind-bending.

If you look at the favicon, there's two of them faviconception

I do not understand what is going on.

And they're down.

It was a good 15 minutes, though.

I bet they closed their window >_>

෴

What sorcery is this!

We put a server in your browser. #inception

Hacker News killed it with fire!

Yes, still down.