huggingsocial

Find, discuss and download AI models. Peer to peer.

No company owns it, hosts it, or can shut it down.

How it works ↓

1Anyone creates a hub: a mutable record only their keys can update · gossiped to the rest of the p2p network

2A publisher gossips a signed listing to the network · hubs listening pick it up

3Hubs spread the listing and its discussion peer to peer · over a p2p pubsub network

4Peers download and seed the weights · over BitTorrent, keeping every model alive

Why now

Almost every open model lives on Huggingface, and in August 2026 Nvidia agreed to buy it. When one entity controls where models are stored, the next thing that gets controlled is which models are visible, which are removed, and which conversations are allowed. Models have already vanished, removed by the platform (GPT-4chan) and wiped by corporate takedown (Meta's DMCA against 403 LLaMA repositories).

How it works

Discovery

The index of models is held by hubs controlled by cryptographic keys. Nobody can delete a hub or moderate a hub except its owner. Each hub sets its own rules and anti-spam challenge. Anybody can create a hub without needing permission or to pay anything

Provenance

Every listing is signed by whoever published it and carries the checksums of every file. Listings include multiple URLs for different ways of downloading, a bittorrent swarm, IPFS, a HuggingFace URL.

Discourse

Model cards, benchmarks, "does this quant run on my GPU", fine-tune requests and license debates happen in the same hubs, with moderation set by each hub, not by huggingsocial.co

HuggingSocial hosts nothing. We build the client and tooling. Anyone can run a hub, an index, or a seedbox. There is no central server or domain to block, which is what makes the network censorship-resistant: it keeps working where huggingface.co doesn't.

What we're building

A grassroots project

There is no company behind HuggingSocial. Every line is free software under GPL-3.0: the clients, the tooling, this website. If you would like to help in development or designing this, or can run a hub or a seedbox, tick a box in the or come say hi on GitHub.

FAQ

When can I use it?

We're building the MVP: the hs CLI, the local HF_ENDPOINT node, and a first hub mirroring popular models. Sign up and we'll email you when there's something to run.

How is this different from a torrent index?

A torrent index gives you files. HuggingSocial gives you a signed listing (who published it, which version, which checksums) plus the discussion around it, in a community that can't be taken down by a hosting provider. Torrents are just one of the ways the bytes arrive.

Is this just a Hugging Face mirror?

No. Mirror sites re-serve Hugging Face's catalog; HuggingSocial is a place to publish. Anyone can post a model to a hub directly without Hugging Face account, or needing any permission from anybody.

Why is a social layer needed?

Because the model ecosystem lives on platforms that remove things. Hugging Face already takes down models and datasets at its own discretion, and the places where people share and discuss them, Reddit, Discord, are the obvious next point of control: one policy change and a model can no longer even be linked or talked about. Torrents alone don't fix this, because the index and the discussion are still the choke point. On HuggingSocial the listings, discussion and moderation are P2P too, so as long as a model is seeded, nobody can take it down, delist it, or stop people from talking about it.

What is the social layer?

Bitsocial: a peer-to-peer social protocol on IPFS/libp2p where communities are owned by keypairs. A hub is a Bitsocial community; a model listing is a post in it.

How do you prevent spam if anyone can publish?

Each hub sets its own challenge. Publishing is a pubsub exchange: the publisher gossips the signed post to the network, the hub's operator node hears it and replies with its challenge (a captcha, an allowlist, a proof of payment, anything that can be coded basically), and only answers that pass get written into the hub's record. A post that never passes never enters the record, so readers never see it. Spam control is per hub, not platform-wide.

Is this a crypto project? Is there a token?

No. HuggingSocial has no token and will not have one.

Who hosts the weights?

Peers in BitTorrent swarms, HTTP mirrors (including Hugging Face), IPFS. Downloaders seed by default, which is what keeps the long tail of models alive.

What if the torrent has no seeders?

A listing isn't just one torrent, it's every known way to get the files. The client tries them all: the BitTorrent swarm, Hugging Face and its mirrors over HTTP, IPFS. If the swarm is dead but any mirror still has the bytes, the download works, and once you have them you seed, reviving the swarm for the next person.

How do I know a download is genuine?

Listings are signed by the publisher and list the sha256 of every file. The client hashes what it downloads and verifies checksum.

Is this legal?

HuggingSocial is a p2p client. There is no default hub, and each hub is responsible for what it accepts. The same way a browser isn't responsible for the sites you visit.

Can model authors get paid?

Listings carry the author's own sponsor links (GitHub Sponsors, Ko-fi, crypto addresses…). Direct tipping from the client is on the roadmap.

Interested?

Get an email when there's something to run, or tell us how you'd like to help.