A preservation network for open-weight AI

Keep open models open.

Open-weight AI models can be deleted, paywalled, or regulated away. ossllms keeps them alive on a network no single company or government controls.

And the models you already have help keep them alive for everyone.

Install

V0 is live through the Python package with torrent support. Install it, pull verified bytes into the normal HF cache, then contribute supply from the same cache.

current proof — local checkout

python -m pip install 'ossllms[torrent]'

local checkout

python -m pip install -e 'sdk[torrent,dev]'

published package

python -m pip install 'ossllms[torrent]'

Then run ossllms search minilm, pull the MiniLM ref it returns, and use ossllms contribute --dry-run before seeding. PyPI install and public MiniLM seed proof are already green.

The mission

“Our goal is to advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return.

Since our research is free from financial obligations, we can better focus on a positive human impact. We believe AI should be an extension of individual human wills and, in the spirit of liberty, as
broadly and evenly distributed as possible.” — OpenAI, “Introducing OpenAI,” December 11, 2015.

That promise — AI “as broadly and evenly distributed as possible” — is what ossllms makes literal. Open models, held in pieces by everyone. The decentralization is the mission.

What you do

Current V0 commands. Each one matches the package surface that ships now.

pull

ossllms pull hf/sentence-transformers/[email protected]+1110a243fdf4 # get verified bytes into your HF cache

Download a model, integrity-checked end to end, straight into the Hugging Face cache your tools already read from.

contribute

ossllms contribute # seed matches already in your HF cache

Point ossllms at the weights already on your disk and attach to or start a managed worker. Use --dry-run to inspect the plan first.

verify

ossllms verify hf/sentence-transformers/[email protected]+1110a243fdf4 # re-check cached bytes

Audit local model bytes against the signed manifest without running or loading the model.

pull by hash

ossllms pull sha256:… # pull by hash, provably the real bytes

Ask for an exact content hash and get exactly those bytes — provably the model you meant, from whoever has it.

Why you can trust it

Every pull passes three independent checks before it lands in your cache.

Integrity

Every file is hashed and verified, so the bytes you receive are the bytes that were published — no corruption, no tampering in transit.

Authenticity

A pinned key signed the release. You know who vouched for this model, not just that a file with the right name exists.

Matches HF

The content is identical, byte for byte, to Hugging Face’s own official files for the same model.

A note on words: “Verified” here means verified origin and integrity — that the bytes are authentic and unaltered. It does not mean “safe to run.” Treat model weights like any other code you didn’t write.

In the terminal

V0 is a CLI-first path: search the default catalog, pull verified bytes into the HF cache, then contribute from the same cache. V1 adds the signed-catalog website directory and TUI selector.

ossllms — search minilm

  REF                                                    VERSION
  ──────────────────────────────────────────────────────────────────────
  hf/sentence-transformers/all-MiniLM-L6-v2              0.0.1+1110a243fdf4

  ✓ matches HF official
  ✓ Apache-2.0
  Use: ossllms pull hf/sentence-transformers/[email protected]+1110a243fdf4

Pull & contribute

ossllms — pull + contribute MiniLM

  Pulling all-MiniLM-L6-v2 from the default catalog …

  ✓ model.safetensors                  verified
  ✓ tokenizer.json                     verified
  ✓ config.json                        verified

  ✓ files verified
  ✓ matches HF official

  placed in the Hugging Face cache

  $ ossllms contribute --status
  reachability: serving
  download_bytes: 0

Catalog directory

The current public catalog lists 2 pullable model releases. This is static HTML generated from signed catalog manifests, so it works without JavaScript.

sentence-transformers/all-MiniLM-L6-v2

hf/sentence-transformers/[email protected]+1110a243fdf4

Matches HF @1110a243fdf4

Version: 0.0.1+1110a243fdf4
Store: hf
License: Apache-2.0
Size: 87.3MiB
Files: 12
Signed: ossllms
Availability: torrent metadata + 1 web seed

ossllms pull hf/sentence-transformers/[email protected]+1110a243fdf4

ossllms pull sha256:98ccb431c012ebfe976280fbd45aea4cec7409935868ccecf3954370f96732a1

ossllms/smoke-tiny

oslm/ossllms/[email protected]

HF provenance missing

Version: 0.0.1
Store: oslm
License: Apache-2.0
Size: 217B
Files: 4
Signed: ossllms
Availability: bundled local bytes

ossllms pull oslm/ossllms/[email protected]

ossllms pull sha256:956ac1c3125b1ee2e78d127ed96f6366912cd6e9322db4aeae2b87f3ee0b81ec

Get started

Install with torrent support, find MiniLM, pull it into the HF cache, then contribute from that cache. V0 is proven through PyPI and the public MiniLM seed path.

get started

python -m pip install -e 'sdk[torrent,dev]'

ossllms search minilm

ossllms pull hf/sentence-transformers/[email protected]+1110a243fdf4

ossllms contribute --dry-run

ossllms contribute

Run ossllms contribute --status or ossllms contribute --self-test to see whether you are actually serving. V1 brings the interactive TUI selector over the same verified catalog.