Repository View

The top section of the repository view starts with the same information as the repository tile in the catalog (see the Private Registry page for more on the Catalog and Status pages). However, to the right of the repository name is a green tag name and drop-down button. Clicking the tag name opens a modal where you can select from all the tags in the repository. The information on the screen applies to whichever tag is shown beside the repository name.

At the top right there are options for:

• Viewing the repository timeline: lists major actions that have been taken with the repository
• Changing visibility: choose from public (default), or private
• Adding the repository to favorites

Below the pull tag command and copy button are a series of tabs to work with the ModelKit tag.

Deploy

The Deploy tab generates a ready-to-run inference container directly from this ModelKit. Select a runtime from the dropdown list - vLLM, Llama.cpp, or others available in your deployment - and Hub composes the container on demand, merging the runtime with the model weights from this ModelKit. There is no image to build or maintain. The same ModelKit can be deployed to multiple runtimes by selecting a different type; the model weights are identical in each. Copy the generated command and run it in Docker or drop the image reference into a Kubernetes manifest.

For more information see Jozu Rapid Inference Containers.

ModelKit Contents

The ModelKit Contents tab shows what's inside the ModelKit. It lists package metadata, model, prompts, datasets, codebases, and docs. In the top right of each section is a button that copies the kit unpack command to extract only that section of the ModelKit - this is valuable when you only need to pull a small part of an otherwise large ModelKit.

The parameters portion of the Model section can contain any JSON-formatted content. It's often used to capture:

• Model training run results
• Hyperparameters or configuration guidance for specific environments
• Performance information
• Lineage information

The prompts section shows you each of the packaged prompts or skills. You can click the "Show" link on the right to view the full prompt or skill in a new window.

Datasets, code, and docs include a preview function - clicking "Show Preview" on the right will open the first several rows of data or lines of code/text. Not all data can be previewed.

Security Report

The Security Report tab shows Jozu Hub's vulnerability scan results for this ModelKit. The Hub scans across ten threat categories that focused on AI artifacts only (there are more details in the Security Scanning page). Scans run automatically on push; on self-hosted deployments they can also run on a schedule with a customizable scanner configuration.

In addition to showing results in the UI, scan results are signed and attached to the ModelKit as attestations so they travel with the artifact wherever it is pulled allowing local evaluation of security from any location without requiring a connection back to the Hub.

ModelKit Diff

The ModelKit Diff tab shows the content layer diff between two ModelKit tags. The right-hand pane always shows the currently selected tag and in the right-most column you can select the file size units to use. To change the ModelKit you're diffing against in the left-hand pane click the name of the tag at the top of that pane. In the resulting modal you can select other tags or repositories.

Tag Timeline

The Tag Timeline tab shows the changes made to the ModelKit over the selected period. For a more granular view of changes administrators can choose to download the audit log from the repository timeline view (at the top-right of the Repository view).

Model Card

If a ModelKit includes a README.md in the docs layer it will be rendered in the Model Card tab.

Add Repository

You can add a new empty repository manually, import a repository from Hugging Face, or create a repository automatically by pushing a ModelKit to Jozu Hub. For more information see the Creating Repositories doc page.