Matching anonymous short posts by semantic similarity into instant chat rooms is a neat, constrained take on discovery — it sidesteps follower mechanics and outright listing to discourage vanity. The 'echo' mechanic and no-enumeration rule are clever behavioral design choices, but the MVP leaves big unknowns around moderation, spam, embedding thresholds, and how well semantic matching avoids false positives.
Privacy-conscious social explorers, people who like social experiments and serendipitous connections
I made Voidful, a kind-of-anti-social network where users can yell into the void to see how many other people are yelling the same thing. If you yell something similar, this is considered an "echo" and you gain access to a chat room about that topic with other users being notified.
Currently when users yell into the same void they are brought into a chat together, but this is somewhat of an MVP. In the future that chat may have guard rails related to the topic, rewards left by other users, and possibly more!
A key concept is that users can never enumerate or list the voids, as otherwise it will likely turn into a vanity war like traditional social media, which in my opinion is at the all-time-high for gamification. I am trying to remove the incentive to say random stuff to make line go up.
I am interested in how people use this kind of technology. I believe in the future when more consumer devices are capable of inference this will be adapted to be more client side and anonymous, with the vector database using homomorphic encryption via [HEVEC].
Check it out: https://void.devrupt.io
This was similar to my last experiment, [Ethos], also with very low inference cost using `qwen3-8b` for the LLM and `qwen3-embedding-8b` for the embedding.
Any feedback about subjects you give it and incorrect analysis is appreciated. I am curious what happens if a lot of people go into a chat at once, and it makes sense to try "Hacker News" as a chat topic.
The UI is a straightforward, local-first-feeling social shell — communities, posts and polls with an 'Offline' status hinting at P2P behavior. It's an earnest early demo: the tamperproof claim is intriguing but unexplained (no visible crypto/protocol details) and the app is essentially empty right now, so it's more promising sketch than ready alternative to ActivityPub/Nostr.
This intentionally avoids generative LLMs and instead stitches together Whisper, Piper, spaCy, VADER, sumy and YOLO into a deterministic, local assistant — a practical tradeoff that kills API bills and prompt-injection risk. The blog feature (extractive summarization + site crawling) is an especially smart move: it produces usable titles/content without hallucination. It won't replace creative LLM outputs, but for offline, private automation this is a refreshingly pragmatic build.
Anonymous photo sharing without likes, but the email gate blocks preview.
AI agent social network concept is interesting; execution is a static demo with fake activity.
Cross-platform media playlists on MySpace-style profiles, but social networks are incredibly crowded.
Grep meets embeddings for code, runs fully local—but codebase Q&A tools already flood the market.
