KitOps is an open-source DevOps tool, now a CNCF project, for packaging and versioning everything an AI/ML system needs — models, datasets, code, and configuration — into a single OCI artifact. It has around 1,400 stars on GitHub and is under active development in 2026.
ML projects sprawl across model files in one system, datasets in another, and parameters in someone’s notebook, which makes releases hard to reproduce and audit. KitOps defines a packaging format called a ModelKit that bundles model weights (including GGUF files), datasets, code, and configuration together, stored and distributed through the same OCI registries and tooling — Docker-style registries, Kubernetes deployment flows — that teams already use for containers. Artifacts can be signed, verified, and pulled selectively, so a deployment can fetch just the model without the training data. Because it rides on OCI standards, KitOps slots into existing CI/CD pipelines and platform engineering practices instead of demanding a separate ML-only stack, and it supports compliance workflows that require artifact integrity, provenance, and audit trails. The project is Apache-2.0 licensed and works with models from Hugging Face and frameworks such as PyTorch, scikit-learn, and TensorFlow.
Free. KitOps is fully open source under the Apache-2.0 license; commercial platforms such as Jozu Hub build on it, but the tool itself is free.
MLOps and platform engineers who want container-grade versioning, distribution, and governance for AI models and their accompanying assets.