KitOps Getting Started Tutorial

KitOps is the open-source CLI tool for packaging and sharing complete AI/ML projects using the ModelKit format.

In this guide, you'll:

1. Unpack and inspect a sample ModelKit
2. Package your own model and data
3. Push it to a remote registry for collaboration or deployment

Prefer to use KitOps with your favorite MLOps or other tool? Check out our integrations

Prerequisites

• Install the Kit CLI
• Verify it by typing kit version in a new terminal
  • If you get an error check your PATH
• Create and navigate to a new folder (e.g., KitStart)

Step 1: Log In to a Registry

You can use any OCI-compatible registry. We’ll use Jozu Hub for this example:

kit login jozu.ml

Use the email and password you signed up to your registry with.

Trouble? See the kit login docs.

Step 2: Unpack a Sample ModelKit

TIP

If you already have a model or dataset on your machine navigate to the directory where the files are and run kit init . in your terminal to build a Kitfile automatically.

We’ll pull and unpack a fine-tuned Llama 3 model:

kit unpack jozu.ml/jozu-quickstarts/fine-tuning:latest

This unpacks all files to the current directory:

.
├── Kitfile
├── README.md
├── llama3-8b-8B-instruct-q4_0.gguf
├── lora-adapter.gguf
└── training-data.txt

Step 3: Pack your ModelKit

Use the kit pack command:

// Replace <your-name> with your Jozu.ml user
// [!code word:/your-username]
kit pack . -t jozu.ml/your-username/finetune:latest

This saves the ModelKit locally under the latest tag. Verify:

kit list

Step 4: Push to a Remote Registry

Now push your ModelKit to share it:

// Replace <your-name> with your Jozu.ml user
// [!code word:/your-username]
kit push jozu.ml/your-username/finetune:latest

💡 If you see an error, check that your target repository exists and that you have permission to push.

Next Steps

If you'd like to learn more about using Kit, try our Next Steps with Kit document that covers:

• Creating a container or Kubernetes deployment from a ModelKit
• Signing your ModeKit
• Making your own Kitfile
• The power of unpack
• Tagging ModelKits
• Keeping your registry tidy