Today the most capable AI lives in a few big data centers, behind a few big companies, reachable only through their APIs. For something this important, Neonix Labs thinks that is the wrong default. Knowledge has always been most useful when it is open, forkable, and shared — like open source, like the open web. Neonix Labs' goal is simple: open models and open knowledge that anyone can run, read, fork, verify, and share — on their own hardware, over open protocols, owned by no one. Decentralized by design, reproducible by default, a commons for everyone.
Open, reproducible models small enough to train and run yourself. RitsuGPT — a from-scratch GPT in pure Rust — is the first brick: open weights, an MIT-licensed trainer on GitHub, a reproducible bits-per-byte benchmark, and it runs on your own computer.
Models and knowledge federated over the open standards the web already uses — no central server, no gatekeeper. Neonix Labs builds on Nostr (an early daemon prototype is running), Matrix, ActivityPub and the AT Protocol; libp2p and IPFS for content-addressed storage; ed25519 keys and DIDs for self-sovereign identity. Peer-to-peer, self-hostable, owned by no one.
A self-sustaining commons of open, verifiable AI knowledge. Anyone can contribute, share, fork, and verify — open data, open weights, open provenance. A public good that belongs to everyone and no one: permissionless and censorship-resistant.