Show HN: Safely run multiple applications on the page using Web Workers

I think silos are created in the Web by the security model of document sandboxes.

It's not possible to share a connective session between two applications without involving remote hosts. All traffic has to go out to a server to reach other tabs, and it only happens if the two hosts have channels to each other. As a result, the functional connectivity of web applications is determined by the providing services.

Put another way, I can't tell the the facebook interface to download all of my posts and hand them over to my google docs interface. I might tell the google docs server to use a protocol like OAuth to get permission for a remote session, so the docs server could download the posts from the facebook server-- but that scenario is only happening if both parties agree to and implement the channel, which is cumbersome. So as of right now, even if the facebook interface can easily grab a list of my posts, they're staying in the interface.

And it's actually a little worse than that. Even with a perfectly connective world of hosts, where every service has a channel to the other service you need, the arrangement is difficult to trust. Exactly what am I granting permission to by connecting services? How could I know? It's only vaguely apparent in the permissions interfaces, and the decisions are rarely case-by-case. "Will read all your mail and post to your feed (and communicate with its third-party host)." Great, I really want to roll the dice on that.

Again, this is because there are no permeable barriers between browser tabs for information to pass through, nor are there the tools to easily configure and safely mediate that barrier if it existed. And since we rely on domain-based security, there aren't any good tools for mediating remote connections either.

So how do we solve it? I think we need to update the browser environment to support more fine-grained connectivity and security policies, which is the purpose of LinkAP. Once you get a sandbox that can inter-communicate with other sandboxes (workers) it's a matter of setting up a strong protocol for routing messages, configuring endpoints, and enforcing security.

There are probably a lot of ways you can do it, but the choice I made was to use HTTP so that local applications can behave as invisible proxies to remotes. That allows programs to be gatekeepers to their services, usually with some permissioning tools from the environment. HTTP also allows links/forms/ajax to work as before on the in-browser apps. There are a lot more details (some still developing) but I'll leave that for another time.

The design doc gets into it more: https://github.com/pfraze/link-ap/wiki/Design-Document