A model registry makes weights available beyond the computer that stores the original copy. Pushing an Ollama model to ollama.com publishes it under your account for other users to download and run on their own machines.
The source model must already be installed and licensed for redistribution. Public uploads can expose bundled prompts and metadata as well as weights, so a model containing secrets or private training data is unsuitable for public sharing.
Publication requires an Ollama account signed in on the server used for the push. The destination name begins with that account's username, while a second installation with an empty model store can confirm that the published model is available independently of the original local copy.
Steps to push an Ollama model:
- Inspect the locally installed source model.
$ ollama show smollm2:135m Model architecture llama parameters 134.52M context length 8192 embedding length 576 quantization F16 template source go_template Capabilities completion ##### snipped #####The source here is the small smollm2:135m text model. A missing-model error means it must be downloaded or created before publication.
Related: How to pull an Ollama model
Related: How to create a custom Ollama model - Review the source model's license before sharing it.
$ ollama show --license smollm2:135m
SmolLM2 carries the Apache 2.0 license. Redistribution must preserve its license and any applicable notices; copying the model to a new name does not change those obligations. A custom model may have additional restrictions from its base weights or training material.
- Test the source model with a short local prompt.
$ ollama run smollm2:135m "What is 2 plus 3? Answer in one short sentence."
The reply should give the sum 5. A load error or missing reply means the local model is not ready to share.
- Copy the source model to an unused name under your Ollama username.
A copy can replace an existing local destination, and a push can update a published tag. An unused destination avoids overwriting a model that someone already uses.
$ ollama cp smollm2:135m exampleuser/smollm2-demo success copied 'smollm2:135m' to 'exampleuser/smollm2-demo'
exampleuser represents your own account username, and smollm2-demo is the new model name. An omitted destination tag selects latest. This copy creates a local alias; it does not upload the model.
Related: How to copy an Ollama model - Push the namespaced model to ollama.com.
This uploads the model under your account for other people to download. Only models cleared for public distribution belong in this destination.
$ ollama push exampleuser/smollm2-demo retrieving manifest ##### snipped ##### pushing manifest success
A final success confirms the upload. An authorization error means the Ollama server's sign-in or destination namespace needs attention; browser sign-in alone does not authenticate a different server.
Related: How to sign in to Ollama - Check the model list on a second Ollama installation with an empty model store.
$ ollama list NAME ID SIZE MODIFIED
This installation needs Ollama running and enough disk space and memory for the model. A public model can be downloaded without signing in. An empty store prevents existing local layers from hiding a failed download.
- Pull the published name on the second installation.
$ ollama pull exampleuser/smollm2-demo pulling manifest ##### snipped ##### verifying sha256 digest writing manifest success
- Run the downloaded model on the second installation.
$ ollama run exampleuser/smollm2-demo "What is 2 plus 3? Answer in one short sentence." 2 + 3 = 5
The response should give the sum 5, although its wording can vary. A reply after the fresh pull confirms that another installation can download, load, and use the published model. The original and downloaded copies remain available.
Related: How to run a model in Ollama