DVC (Data Version Control) is an open-source tool that brings Git-style versioning to data and machine learning models. Large datasets and model files do not belong in a Git repository, so DVC stores them in your normal cloud or on-premise storage while keeping small metafiles in Git that record exactly which version every experiment used. On top of versioning it adds lightweight pipelines and experiment tracking, so ML projects become reproducible end to end. DVC has nearly 16,000 stars on GitHub, is maintained by Treeverse, and was actively developed in 2026.
DVC works through a command-line tool and a VS Code extension. You version data and models with simple commands, define processing and training steps as pipelines, and DVC reruns only the steps affected by a change, like a build system for ML. Experiments are tracked and compared locally in your Git repo, parameters, metrics and plots included, with no server to set up. Because the heavy files live in S3, Azure, Google Cloud or SSH storage of your choice while Git holds the pointers, teams can share datasets and reproduce anyone’s experiment exactly. It fits any language or framework, since it versions files and commands rather than code.
DVC is completely free and open source under the Apache-2.0 license. There is no paid tier for the tool itself; you pay only for your own storage.
DVC is for data scientists and ML teams who need reproducibility: anyone tired of final_v2 datasets, untracked experiments and models nobody can rebuild will benefit from versioning data and pipelines alongside code.