kitops-ml/kitops

What it solves

KitOps provides a standardized way to package, version, and share AI/ML projects. It addresses the difficulty of keeping model weights, datasets, prompts, and configurations synced and trackable across development, staging, and production environments, especially for organizations requiring high security and auditability.

How it works

KitOps uses OCI (Open Container Initiative) technology to bundle AI assets into "ModelKits"—immutable, layered artifacts stored in existing container registries. A Kitfile defines the contents of the package. Users manage these assets via a CLI or a Python SDK (PyKitOps), allowing them to pack, push, pull, and inspect models without needing to download the entire bundle.

Who it’s for

• DevOps and Platform Engineers who want to integrate AI models into existing CI/CD pipelines and Kubernetes deployments.
• Data Scientists who need to package and track experiments and models without managing infrastructure.
• Developers who want to treat AI models as versioned dependencies.

Highlights

• OCI-Based Storage: Uses standard container registries for AI asset management.
• Tamper-Proof: Every component is protected by SHA-256 digests for consistency and traceability.
• Security Integrations: Supports cryptographic signing with Cosign and transparency logging with Rekor.
• Flexible Unpacking: Allows users to pull only specific layers (e.g., just the model weights) instead of the full package.
• Standardized Formats: Supports both ModelKit and the CNCF ModelPack specification.