Pirate Face Rescues LLM Models from Deletion

Overview

Pirate Face is a decentralized, peer-to-peer infrastructure designed to provide permanence for open-source AI. It transforms open models-LLMs, images, audio, and datasets-into checksum-verified torrents that cannot be permanently removed. By distributing models across a global swarm rather than relying on a single host, Pirate Face eliminates single points of failure and ensures that once a model exists, it survives even if any individual server shuts down.

Every open model is mirrored from Hugging Face as a torrent and held in a peer-to-peer swarm. This means the real weights remain accessible regardless of whether the original Hugging Face instance remains online. The system is built around the principle that the data lives wherever it's stored, not under the control of any single entity.

What Is Pirate Face?

What is Pirate Face?
A decentralized, peer-to-peer layer for sovereign AI. Open models from Hugging Face become checksum-verified torrents, held by a global swarm so they survive any single host taking them down. The primary goal is permanence for open-source AI.

Do I need an account to use Pirate Face?
No. Users can browse, download, and seed models without creating an account. However, an account provides additional functionality such as managing claimed handles and tracking contributions.

How It Works

Torrent-Based Distribution

Each model is distributed as a magnet link (torrent) that includes a built-in web-seed-a plain HTTPS URL pointing directly to the model file on Hugging Face. This design ensures that models can download successfully even when there are zero peers available, because the web-seed serves as a guaranteed source during the initial download phase.

When the original Hugging Face instance is removed, the web-seed becomes unreachable and the download automatically falls back to the peer-to-peer swarm. This fallback mechanism is central to the resilience of the system: the same bytes are delivered both directly from Hugging Face while it exists and through the distributed swarm afterward.

Checksum Verification

To guarantee integrity, every weight file carries its official Hugging Face SHA-256 hash-a unique fingerprint of the exact bytes. When users download a model, they verify that the downloaded hash matches the official one. This prevents tampering and ensures that the files being accessed are the genuine weights, never modified copies.

Account Management

Claiming Handles

Users can reserve a public handle by visiting Reserve pirateface.co/yourname. This creates a public profile for the account. While reservation alone does not confer a creator badge, it allows the network to spread awareness of the handle and supports community engagement.

Creator Badge Verification

To earn the creator badge associated with a Hugging Face username or organization, verification is required. The user must verify the matching Hugging Face account to prove ownership. Without verification, the handle remains unreserved and does not carry the badge, though it can still be used for browsing, downloading, and seeding.

Points and Rewards System

Pirate Face implements a points-based system to recognize participation across the community:

• Welcome points: One qualifying handle per account.
• Referral points: 50 points for a new account's first handle, up to 10 credited referrals per referring account (self-referrals and extra handles do not count).
• Rescued models: 25 points per distinct rescued model, credited to its first verified handle.
• Seeding rewards: Planned feature; currently inactive pending attributable seeding evidence.

Points do not automatically convert to compute credits, and reserving additional handles does not increase benefits. There is no token economy.

Model Submission

If you wish to contribute your own model to the swarm, follow these steps:

1. Ensure the model has a surviving copy on Hugging Face.
2. Upload the model to Hugging Face (free tier suffices).
3. Submit a magnet link along with:
   • Source evidence
   • File checksums
   • License information
4. Restrict access to MIT and Apache-2.0 licenses, plus the approved Kimi-K3 exception.

Only submissions that pass review (with matching Hugging Face LFS hashes at the pinned revision) will be accepted. Models not yet present on Hugging Face must first be uploaded there before submission.

Technical Architecture

Web-Seed Concept

A web-seed (defined by BitTorrent specification BEP-19) is a simple HTTPS URL embedded within a torrent file. In Pirate Face, this is the model's file on Hugging Face. Because it is included directly in the torrent metadata, the model can begin downloading immediately-even with zero peers-since the URL acts as a reliable source.

Drop-In API

Existing pipelines can integrate seamlessly by setting the environment variable:

export HF_ENDPOINT=https://pirateface.co

This redirects model lookups through Pirate Face while maintaining compatibility with the original Hugging Face interface. The change requires zero code modifications.

Security and Integrity

Both the model weights and the handle itself are protected against impersonation:

• Weights: Cannot be faked. Every file is checked against its official Hugging Face SHA-256, ensuring the exact bytes are retrieved.
• Handles: Anyone can reserve a name (displaying it as "reserved"), but only proof of matching Hugging Face identity earns the .pirateface.co badge.

Even if a squatter attempts to lock a handle, the real owner can reclaim it by verifying their Hugging Face identity. This dual-layer protection-cryptographic verification of content and identity confirmation for handles-ensures the integrity of the entire system.

Summary

Pirate Face represents a significant step toward permanent, decentralized hosting of open-source AI models. By combining torrent-based distribution, cryptographic verification, and a community-driven governance model, it addresses the core vulnerability of centralized platforms: the risk of arbitrary removal. With its focus on resilience, transparency, and user empowerment, Pirate Face aims to make open AI models truly permanent and independent of any single host.