In practice: Having arbitrary files doesn't help much, as any protocol aiming to be interoperable will have to agree on a standard anyway. Because the homeserver is so flexible it also lacks a querying language for getting the posts that follow a certain standard (whatever is their kind:1, for example) efficiently directly by clients. Of course one can be bolted on, but they don't seem to care about this. They have chosen instead to solve it by having an aggregation layer on top: a central server that connects to each home server, inefficiently, and indexes them. After that "clients" connect directly to this aggregator server, blindly trust it (because there are no signatures) and live happily in fully centralized environment (this part is exactly like Bluesky). The possibility of having keys always offline is good, but in practice it doesn't make much difference. Most users have the same likelihood to lose the keys anyway. Claims of maximum decentralization and infallible discovery using the BitTorrent DHT remain to be proven: DHTs are beautiful in practice but I'm not convinced they scale in the real world. Relying on people serving your content against their will or without even knowing they're doing that can't be a reliable solution. The BitTorrent DHT numbers are also suspicious, they're way too big, I would guess this is either fake, a lot of home users that hurt the network more than help because they can't accept connections, or ran by the Mossad. Another weird part of the protocol is that because browsers can't connect to the DHT directly they rely on these gateway servers -- now, if this is it, why can't these gateways just store the keys anyway and be the discovery layer? DHT not necessary. (I already see @Nuh complaining the DHT offers ultimate censorship-resistance that the gateways can't, ok). After all is said and done, I think Nostr could benefit from a "generic filesystem" abstraction at some point, but this can easily be done later using normal servers.