KitOps is an open source CNCF project for packaging, versioning, and sharing every part of an AI project as a single OCI artifact called a ModelKit. A ModelKit bundles models, datasets, code, prompts, agent skills, MCP (Model Context Protocol) servers, guardrail configurations, and policies into one versioned, immutable package, replacing the scattered repositories, zip files, and git submodules teams typically use to hold an AI project together. KitOps is the largest implementation of the CNCF ModelPack specification, the only openly governed standard for AI/ML packaging, and was built intentionally to meet enterprise needs for securing the AI supply chain for models and agents in production.
Security and provenance are core to the design. Every ModelKit component is protected by SHA-256 digests, and KitOps integrates with Cosign and other signers so teams can sign and verify models, agent skills, and MCP servers before anything runs. Each ModelKit carries the metadata needed for an AI Bill of Materials: dependencies, configurations, licenses, and lineage for everything inside the package. That evidence supports organizations working toward requirements like the EU AI Act and the NIST AI Risk Management Framework.
KitOps is registry-agnostic. ModelKits store in any OCI-compliant registry, including Docker Hub, Amazon ECR, Google Artifact Registry, Harbor, Artifactory, and Jozu Hub, so organizations keep their existing infrastructure and access controls with no lock-in to a proprietary ecosystem. ModelKits deploy anywhere, from public clouds to fully air-gapped environments, which matters for enterprises and government teams with strict security and compliance requirements.
The CLI mirrors familiar Docker commands (kit pack, push, pull, tag), so anyone comfortable with container workflows is productive quickly. For programmatic use, the PyKitOps Python SDK integrates with existing tools and pipelines like MLflow without rework. A GitHub Action and integrations with major CI/CD tools let teams build a ModelKit on every commit, push it to their registry, and promote versions based on policy.
The result is one secure, portable, versioned package for the entire AI project that works with the registries, pipelines, and tools teams already use.
Use KitOps to speed up and de-risk all types of AI/ML projects:
Predictive models
Large language models
Computer vision models
Multi-modal models
Audio models
etc...
Who Is the Company Behind KitOps?
