How to pull models from custom model libraries beyond the default

What this does

Downloads a model whose manifest is hosted at a third-party location rather than the default community library. The namespace-prefixed name routes the request to an alternate registry, and Ollama stores the result locally just like a standard pull.

Steps

1. Confirm the custom reference format. Custom registries use a slash-delimited path: namespace/model-name with optional :version suffix. Retrieve this from the third-party documentation or manifest file.

2. Pull with the namespaced reference. Ollama resolves the manifest from the custom registry and begins downloading blobs.

   ollama pull exampleuser/custom-model:1.0
   

   Expected output: Progress bars for each blob layer, concluding with success.

3. List to confirm the model is registered locally. The model now appears by its full reference in the local inventory.

   ollama list
   

   Expected output: A new row in the table whose NAME column reads exactly exampleuser/custom-model:1.0.

4. Launch using the full reference. Runs the model in the same way as any library model.

   ollama run exampleuser/custom-model:1.0
   

   Expected output: >>> Send a message (type /bye to exit).

Verification

ollama show exampleuser/custom-model:1.0 2>&1 | head -4
# Expected: manifest block displaying architecture and digest fields confirmed against the remote reference

Common failures

• Error: not found: manifest - Registry is unreachable or the path is incorrect; verify the URL or manifest location.
• authentication required - The remote registry demands credentials; use environment variables or a credentials helper if supported.
• invalid reference format - Missing namespace or using an unsupported separator; check documentation for the required naming convention.
• network unreachable - Firewall or proxy blocking the registry domain; configure a proxy or whitelist the domain.
• incompatible quantisation - The custom library used a format Ollama does not support; fall back to a compatible variant or re-export the weights.

Related guides

• How to download and run a model with Ollama
• How to verify model integrity after downloading