Pirate Face describes itself as a decentralized, peer‑to‑peer layer for sovereign AI. According to the source, every open model from Hugging Face is mirrored as a magnet link that carries the model’s official SHA‑256 checksum, turning the weights into a torrent that can be downloaded and verified anywhere. The source explains that while a model remains on Hugging Face, downloads pull the bytes directly from that host, preserving speed and integrity. If Hugging Face later removes the model, the built‑in web‑seed (the HTTPS URL from Hugging Face) disappears and the download automatically falls back to the global swarm of peers that continue to seed the torrent. At that point the model is marked “Rescued” and stays reachable as long as at least one peer retains a copy. The source notes that this mechanism eliminates any single point of failure and can serve users far from Hugging Face’s infrastructure faster than a direct download. Users can point existing pipelines at Pirate Face by setting the environment variable HF_ENDPOINT=https://pirateface.co; the source says this requires zero code changes and preserves the same API paths. Account creation is optional: browsing, downloading, and seeding can be done without an account, but registering allows users to claim a handle at pirateface.co/yourname, verify a matching Hugging Face identity to earn a creator badge, and eventually access planned community benefits such as free compute credits and exclusive model releases. The source also outlines a points system that rewards verified handles for rescued models, referrals, and initial handle claims, though points do not automatically convert to credits. Direct model publishing on Pirate Face is planned but not yet live; today a model must first exist on Hugging Face before it can be submitted as a torrent. All files retain their original Hugging Face SHA‑256 hash, ensuring that any copy obtained through the network matches the exact weights distributed by the original host.
