Manage and serve LLM models as standard OCI artifacts. Ollama-, OpenAI-, and Anthropic-compatible APIs from one small binary. Zola 2026-09-04T00:00:00+00:00 https://llmmanorg.github.io/blog/atom.xml 2026-09-04T00:00:00+00:00 2026-09-04T00:00:00+00:00 Unknown https://llmmanorg.github.io/blog/an-aggregation-of-llmman/

<p>A group of manatees is called an aggregation. As of this week, so is a group of <code>llmman serve</code> daemons.</p>

2026-09-03T00:00:00+00:00 2026-09-03T00:00:00+00:00 Unknown https://llmmanorg.github.io/blog/verify-the-model-not-the-hub/

<p>llmman has always content-addressed everything it pulls. Every blob is re-hashed on the way in, so the bytes always match what the registry said they would be. That proves nothing about who put them there.</p>

2026-09-01T00:00:00+00:00 2026-09-01T00:00:00+00:00 Unknown https://llmmanorg.github.io/blog/mirror-hugging-face-models-into-any-oci-registry/

<p>I’ve been building llmman, a CLI that packages LLM models as OCI artifacts and serves them over Ollama-, OpenAI- and Anthropic-compatible APIs.</p> <p>The reported Nvidia/Hugging Face deal is why I’m writing this now. In last week’s discussion, the alternatives people reached for were torrents and ModelScope, and one question went unanswered: is there a third option that isn’t a US GPU monopoly or a Chinese hyperscaler? I think the boring answer is the registry infrastructure a lot of us already run.</p>

2026-08-31T00:00:00+00:00 2026-08-31T00:00:00+00:00 Unknown https://llmmanorg.github.io/blog/introducing-llmman/

<p>Model weights are large binary blobs with a name, a version, and a dependency on nothing. That is exactly the shape of an OCI artifact, and the infrastructure for moving OCI artifacts around is already deployed almost everywhere: OCI registries, mirrors, caches, auth, signing, retention policies, quotas.</p> <p><code>llmman</code> is a command-line tool built on that premise. Models are packaged as standard OCI artifacts and stored in any compatible OCI registry (Docker Hub, GHCR, quay, or something self-hosted), and <code>llmman serve</code> exposes Ollama-, OpenAI-, and Anthropic-compatible HTTP APIs on top of the local store.</p>