MLflow

MLflow is an open source platform for managing the full lifecycle of a machine learning project: tracking experiments, packaging models in a standard format, model registry, and deployment. The software is distributed under Apache 2.0 license, entirely free, with no paid plans or features locked behind a subscription: it is a Linux Foundation project, not owned by any commercial vendor. It installs locally in a single command or deploys on your own tracking server. Databricks, which created and then contributed the project to the foundation, offers a managed version called Managed MLflow integrated into its platform, but with no dedicated published pricing: it is billed as part of overall Databricks consumption (DBU), either on a flat rate or on quote depending on the contract.

MLflow is not translated into French, neither on the project website nor in the technical documentation. With self-hosting, hosting and GDPR compliance depend entirely on the infrastructure chosen by the user, since MLflow itself stores nothing on third-party servers. The REST API and Python, R, and Java SDKs cover most use cases, with native integrations for scikit-learn, PyTorch, TensorFlow, XGBoost, Hugging Face Transformers, and Apache Spark. The honest limitation is the absence of an MLflow-specific managed service for small organizations that do not want to manage their own tracking server: without Databricks or self-hosting, you must turn to third-party offerings (such as AWS or Azure managed cloud services) to get a turnkey hosted version.