Neurex distributes, mirrors and serves multi-terabyte model weights to a closed circle of researchers and builders — at line-rate, anywhere on earth.
Access is gated. You'll need a referral code from an existing member.
Every checkpoint is integrity-checked, deduplicated and pushed to the edge before you ever click download. No throttling, no queues.
Pull 70B+ weights at full pipe speed over resumable, chunked transfer. Saturate your uplink, not ours.
Anycast edges on five continents keep the bytes close. Whichever region you're in, the model already is too.
Signed manifests, per-member tokens and end-to-end encryption. Nothing leaves the circle.
Block-level dedup across fine-tunes means a new LoRA costs megabytes, not gigabytes.
safetensors, GGUF, MLX — fetched, verified and unpacked with a single neurex pull.
Each weight ships with its lineage, license and reproducible hash. Know exactly what you're running.
Neurex isn't a hub you wait in line for. It's a distribution backbone engineered to move enormous models the moment they drop.
Neurex grows by referral. Members vouch for members — which keeps the catalog clean, the network fast and the community sharp.
Enter it below to join the waitlist for the next onboarding cohort. We open a small batch of seats every few weeks.