KitOps

About KitOps

KitOps is a packaging and versioning tool for AI projects, distributed as a CLI (Kit) and a Python SDK (PyKitOps). It bundles models, datasets, code, agent skills, MCP servers, guardrail configs, and policies into a single versioned, immutable OCI artifact called a ModelKit, which can be stored in any OCI-compliant registry such as Docker Hub, ECR, GCR, Harbor, Artifactory, or Jozu Hub. Individual components within a ModelKit can be pulled separately rather than requiring the whole package.

The tool supports signing and verification with Cosign, SHA-256 hashing of every component for tamper detection, and generation of AI Bills of Materials mapped to EU AI Act and NIST AI RMF requirements. It integrates with CI/CD through a GitHub Action or other major CI/CD tools, and can automatically generate a container from a ModelKit. It is open source under the Apache 2.0 license.

KitOps is aimed at organizations running AI projects across data science, application development, and deployment teams, where models, datasets, and configurations need to be handed off between teams with version tracking and integrity verification. It positions itself as an alternative to using Git (which handles large model/dataset files inefficiently via LFS) or ad-hoc combinations of zip files and git submodules for coordinating AI project artifacts.

Key features

• Packages models, datasets, agent skills, MCP servers, and configs into a single versioned OCI artifact
• Signs and verifies artifacts with Cosign and generates SBOM-ready evidence with SHA-256 hashing
• Works with existing OCI-compliant registries like Docker Hub, ECR, GCR, Harbor, and Artifactory
• Docker-like CLI (kit pack, push, pull, tag) with a Python SDK (PyKitOps) for MLflow and other pipelines
• CI/CD integration via a GitHub Action or other major CI/CD tools
• Generates AI Bills of Materials mapped to EU AI Act and NIST AI RMF requirements

Frequently asked questions