Nostr Web Client

2025 Nostr Predictions:

1. Algos will seriously challenge follow lists. Clients will de-emphasize the author to focus the post's merit on the content itself and push algos further.

2. First ad-based, VC-backed Nostr client will appear and seriously challenge incumbent clients on UX quality from the revenue model's deep pockets.

3. The Nostr Android ecosystem will evolve to be the proven app decentralization model for all other platforms. App-to-App communication takes over as a Nostr layer 3. Users are on-boarded to a Store, not to a single client.

4. Local-first clients with shared pool of events will become the only way to create a "good enough" performant experience on Nostr without using centralized caching servers, blurring the lines of online and offline.

5. First AI-native Nostr client attempts to render and react to any Nostr content based on voice/chat commands only. The notion of a "feed" disappears. AI-made content starts to challenge human-made content for top zaps and trending posts.

6. Moderated communities are re-born, grow slowly but consistently over time. Safe spaces that create genuine connections and clean, authentic content become a valueable revenue source for moderators.

7. Nostr libraries continue to have a hard time defining stable abstraction layers for Nostr clients to use. Which will lead to a NIP-like repo especifically designed to enable the development of interoperable abstractions that can overtake the complexity of the Nostr relay network protocol.

Jay 1y ago

As I think about how I want to write my idea for a personal algo-relay, I'm butting up against numbers 1 and 4. I want to calculate my web of trust and generate feeds within that set of events. But I also want it to act like a client in the sky that supports outbox for me based on who's in that subgraph, basically a cache. It'll be interesting to see where that goes.

It's a centralizing idea, so I wouldn't want it to become so huge like primal's cache. It would have to be lightweight and easy enough for a motivated individual to run, though neo4j is more resource intensive than relational dbs. There isn't a good localized implementation of their graph model yet as far as I know. Sorry for rambling about this in your replies.

Reply to this note

Please Login to reply.

Discussion

david 11mo ago

I have a neo4j instance that maintains a graph of about 200k nodes of users and their follow and mute connections. It can calculate personalized PageRank in about 15 seconds and is (as you would expect) better than any non graph method at generating a list of followers (or muters).

I’m thinking my next step is to pair neo4j with strfry and create an ETL pipeline from strfry’s LMDB to neo4j. Then neo4j can calculate reputation scores in real time, which we use to modify the strfry filters and keep out spam.

It seems to me that an ETL pipeline starting from LMDB should be pretty performant, in theory, although I haven’t run across any LMDB to neo4j tools. Relational db to neo4j tools are easier to find, on the other hand.

Any thoughts?