Your hardware, your weights
Models download to disk and run locally. Nothing is streamed from a rented GPU, so nothing stops working when a company changes its mind.
Peer-to-peer model sharing
Keep local AI local.
FrostByte is a peer-to-peer network for AI models. Download weights to your own hardware, share what you build, and stop depending on someone else’s servers.
What that actually means
Share what you build
Publish recipes, not just files. A recipe records the model, the quantisation, and the settings that actually worked — so someone else can reproduce it.
No account to browse
Searching the public catalogue needs no sign-in. An account is only for publishing, syncing, and the API.
Where this is
FrostByte is in public beta. The desktop app is signed and available for Apple silicon today, with Windows and Linux builds following. This site is new — the catalogue and account area are being built out now.
Read the FAQ for what works, what does not, and what we will not claim until it is measured.