Pirate Face turns AI models into torrents that never die

Pirate Face wants to turn open AI models into torrents that "live forever." The pitch is simple: language models, image and audio generators, and datasets become magnet links with no single owner and no single point of failure. If a model vanishes from one host, a swarm of peers keeps it alive.

How it works

Every model is mirrored from Hugging Face while also being a torrent with a built-in web-seed — a plain HTTPS link to the model file, per the BitTorrent spec (BEP-19). As long as the model is up on Hugging Face, the bytes come straight from there: same speed, same weights. The day Hugging Face removes it, the web-seed dies and the download falls back to the peer-to-peer swarm. Such a model is marked "Rescued" — still reachable, kept alive by whoever is seeding.

Guarding against fakes

The biggest fear with mirrored models is tampered weights. Pirate Face attaches each file's official Hugging Face SHA-256: no matter where you download from, the hash still has to match. That guarantees you get the real weights, never a doctored copy.

A drop-in API is planned too. Set HF_ENDPOINT=https://pirateface.co and your existing pipeline resolves models through Pirate Face — from Hugging Face while it's up, from the swarm the moment it isn't. Same paths, same API, zero code changes.

Limits and plans

For now the service only mirrors what already lives on Hugging Face: to add your own model, you first have to upload it there. Only MIT and Apache-2.0 licenses are accepted, plus a single Kimi-K3 exception. Direct publishing on Pirate Face itself is promised "soon."

No account is required to use it — you can browse, download and seed without one. Registration merely reserves a public handle, while linking a Hugging Face account earns a creator badge and helps prevent impersonation. There is a certain irony here: a project built on pirate bravado about independence from centralized hosting still leans, for now, on the very platform it promises to outlive — a contradiction the authors acknowledge openly and say they intend to resolve.

Source: Hacker News