I've also done a full network replica (all the data indexed in a postgres) on a raspberry pi (with an 8tb nvme attached via a hat). Its really not expensive to do . And if I wanted to drop data older than say 3 months, it would be even cheaper still.
Case in point! This is an often mentioned statement on which the argument that "atproto is no decentralized" largely hinges. There are honest atproto digs out there but that is not one.
> There are honest atproto digs out there but that is not one.
Which one? The expense, or not being decentralized? The latter remains valid because the majority of the userbase chooses ("only" by default ofc) to coordinate through a single operator. Network effects mean that you either play by their rules or you aren't allowed in the garden.
It's good to learn that a full mirror is so cheap but I think the criticism still holds to the extent that it's a high enough price that unless something changes it will continue to discourage the network from ever becoming truly federated. Compare to activitypub where you can stand up a fully self sufficient node on more or less anything that's capable of networking. The obvious downside being that the network is more fragmented and often less reliable overall (ex nodes are regularly flaky or go missing entirely, no single unified view of the network, etc etc all the perfectly valid complaints about AP).
I think AP, AT, and nostr all get certain things right but all have major downsides baked into their designs. Note that I don't mean this comment to be negative, merely to respond to your remark that the dig in question is somehow invalid.
The default pds packaging takes care of SSL, but thats not a requirement, just something we try to make easy for users.
Also at:// URIs are of the form at://DID/..., and your human readable handle is bound to your DID through DNS TXT records _atproto.roshangeorge.dev, but applications all know to render that as just roshangeorge.dev. That DID points to a document that specifies where your server lives, so the HTTPS/WSS routes can live wherever you want them to.
Also likes/replies/etc on your posts go in their authors repos not yours, your intuition is correct there.
I tend to agree with this, thats why theres canonical URIs for atproto posts that are agnostic to the hosting provider. The format is at://DID/collection/recordKey.
These are used throughout the system, for example if you look at the output of a feed generator, its just a list of these URIs
It's possible, and we've talked about it, but it would take a good amount of work (as would any 'correct' solution), so we have punted on properly solving DMs for a bit. I know Matthew has been very interested in working with us to figure something out here
On the Matrix side, we’d definitely like to help make this happen. Decentralised e2ee is Hard, and after a lot of iteration we’ve got Matrix into a good place these days (plus there’s a new wave of React Native Matrix dev happening atm). Plus we’re not shy about doing stuff like switching IDs on Matrix to DIDs, or even supporting ATProto as a federation mechanism and client/server API if that’s what it takes. https://bsky.app/profile/patak.dev/post/3lbbxbekjw22q gives more context.
What would suck is if bluesky ended up with an entirely incompatible (but functionally identical) thing, so that all the existing Matrix clients, bridges, servers etc couldn’t be leveraged.
> My assumption for this being proprietary is so that it cannot be gamed. It could be a darker reason, I suppose.
As the guy developing said feed, this is indeed the reason. I have a hard enough time as it is with people gaming the specific parameters of the feed, having it fully open would be a huge tax on my time.
You could take the code from there and get to a workable "Discover" feed relatively quickly if you wanted to. Also several of the feeds I have in that repo are among the most popular feeds in the network, especially Quiet Posters (infrequent.go).
Algorithmic curation is, I believe, a key to making (para-)social media enjoyable and Bluesky is the only (semi?) open project I’m aware of investing in this type of stuff. It not being open is a bummer. Is there any good reading material on this topic?
I believe in a world in which people can choose and tune their algorithms, a diversity here should make them harder to game, as well as a highly-weighted „not interested“ button.
It’s surely not the plan to keep this proprietary forever, right?
yeah it works fine on bare metal, you'll just have to do a bit more set up work yourself (https terminating and such). The installer script should be instructive in how to run it but you'll have to figure out the BSD specific stuff
