At Protocol: Building the Social Internet | Hacker News Reader

At Protocol: Building the Social Internet

atproto.com

94 pointsresiros41 comments

Someone posted this article[0] earlier today and I thought it was a really good primer on AT-stuff

I've been playing with my own PDS and AppView for a side project, and it's really fun and interesting.

[0] - https://overreacted.io/a-social-filesystem/

That article should be the official ATProto docs at this point. The current docs are hilariously self-referential and uses terminology that is either way too broad and overloaded for the component they are supposed to describe, or so far misaligned with the standardized computer science concepts that it seems intentionally opaque.

It reads like the devs think they invented a whole new world of computing and not just a little protocol.

I didn't understand what this was until I did some more digging around.

Apparently, it's a decentralized way to interact with social media. A protocol created by Bluesky which allows social networks to communicate directly. This is similar to how different email clients like Gmail and Outlook can send messages to each other.

Yes, but to nitpick slightly: the focus of this protocol is to let people use their own data with social media sites. It might not matter much while the site is active, but separating data ownership from the site makes moving platforms in the future much, much more feasible. Data like posts, friends lists, block lists, likes, comments, etc.

Obviously, this was informed by people's experiences with Twitter, especially the early power users who built strong communities and then felt trapped.

though technically your data lives within Bluesky's servers (kind of like your money lives in a bank) so in some ways it's still the bank's money though it's allotted to you

Actually your data does not live in Bluesky’s servers if you don’t want it to.

Mine doesn’t.

At least another 20-30,000 people’s doesn’t either.

What keeps bluesky from simply catching the data?

I often wonder if people have forgotten that you can send information through the internet without HTTP(S)

Since others starting intercepting and eavesdrop on non-authenticated traffic, it got a lot less comfy to do so though. It's not like people are adding encryption to stuff for the fun of it.

it's not impossible to implement encryption elsewhere, gemini pretty famously requires it

Of course not, the context here is ATProto which literally stands for "Authenticated Transfer Protocol", and it uses other protocols than just http/tls, so of course there are other ways :)

it seems the parent comment isn't referring directly to encryption / TLS, but building everything on top of HTTP / HTTPS, which seems to be the default abstraction these days to build on top of.

I agree, it is always HTTP nowadays. The Librecast project [0] is doing interesting R&D around multicast and social networking, and has been / is being funded by multiple NLnet grants over the years. I think they have plans to use ActivityPub to demonstrate their work. Current active project is LibreCast Studio, a collaboration environment. The linked video on the page is a very interesting watch.

[0] https://librecast.net/about.html

I often wonder if people have forgotten that you can use TLS through the internet without HTTP

I assume for anything else someone will run into firewall/proxy issues.

That's such a non-issue, though. And using HTTP doesn't guarantee you won't run into firewall/proxy issues.

If you ever want to implement a protocol in the browser it has to be on top of HTTP, WebSocket, WebRTC, WebTransport, etc.

You can, but every non web protocol would need the user to install a binary. This is gated by Microsoft, Google and Apple today.

In addition to the middlebox problem, most (not all, but most) of the things that send information over the Internet that aren't HTTP-shaped (incl. HTTP/2 and HTTP/3) are worse than the best HTTP-shaped things. This makes sense: HTTP-shaped things are where all the energy is directed.

Those of you old enough to remember etherpeg can now see an ATproto version:

https://bsky.land

I think it would be interesting if the file servers had read/write/list/delete permissions on files. For both groups and users.

It would mean the public stuff could see your files but private projects could exist. Eg. Maybe I don't want my At Protocol version of Figma making all my drawings public. If they could be shared in a group (anyone in a list in that folder or whatever).

Maybe this is coming, but would interest me way more about using applications on the atmosphere.

The ability to make information private fundamentally conflicts with how ATProto is designed. All records have to be sent to all Relays and AppView nodes on the network to provide a "global view" of the network. So there's no way to keep records private without locking out some user's servers from viewing them, and since AppViews are centralized indexing services, they won't function without being able to see the entire network.

Yeah, apps wouldn't be able to only listen to the firehose.

There are some proposals for private files. However, I'm outside the AtProto world so not sure what exactly the suggested implementations are. I just hope they give enough control.

I think the technology could potentially be used for way more than microblogging. I would love to use webapps that store the data on my devices and share it with specific people. The data and access under my control.

[1]: https://dholms.leaflet.pub/3mhj6bcqats2o

What you linked is by Bluesky’s own staff - it’s him laying out the design plans for the protocol’s private data implementation.

He’s doing it to engage publicly with the community, since many of us also build on the protocol.

> Sync is pull-based. Applications are responsible for staying in sync with all member PDSes. PDSes assist by sending lightweight write notifications to prompt pulls when new data is written.

It looks like this basically just reinvents ActivityPub (local servers can pull or push to remote servers). So it defeats all of the "benefits" you get from Bluesky's firehose-based approach anyway, except for the fact that Bluesky assumes you're going to be using their AppView and they will always have access to your private data.

What would happen if you returned different content depending on who was asking?

It would fail the merkle tree validation

The summary of what this is about is:

"Atproto is a big-world open social protocol. Users publish JSON records into repositories. The changestreams of those records then sync across the network to drive applications."

It's too bad that information isn't on the front page. You have click "GET STARTED" and scroll down

Is it really, at the end of the day, a server that serves your data over JSON (aka a social media post) and offers feeds (ala an RSS model) of the changes of that data and provides services for discovering the schema of that JSON (aka its lexicon)?

Yes

Try reading the actual ATProto docs they are a million times worse

This is really cool. It has retro vibes of the era when the Internet was still free from the big five domination.

The W3C equivalent PDS efforts are in the LWS WG.[1][2] And of course other W3C standards like DIDs, VCs, and ActivityStreams/ActivityPub. W3C is still missing standards around data replication and a global indexing / feed layer.

[1] https://www.w3.org/groups/wg/lws/

[2] https://www.w3.org/TR/lws-ucs/

Previous discussion in 2022: https://news.ycombinator.com/item?id=33252108

How common is it to host your own PDS? It's not really clear to me what the advantages/disadvantage are.

It is not common, but it is becoming more common: https://arewedecentralizedyet.online/

The advantage of Bluesky over Twitter and Mastodon is that it's centralized with no actual plan to decentralize.

This is false. Every single people of the atproto stack can and has now been hosted and deployed by others and is operating at less get than the scale of a small community.

Blog post about this new site from February might shed some light: https://atproto.com/blog/new-site-2026

I could see this turning into a more modern and sane usenet replacement

I love that it embraces standoff properties for rich text markup.

Reply on news.ycombinator.com