MLflow
| Initial release | June 5, 2018 |
|---|---|
| Repository | github |
| Written in | Python, JavaScript, TypeScript, Java, R |
| Engine | |
| Type | AI engineering platform, LLMOps, MLOps |
| License | Apache License 2.0 |
| Website | mlflow |
Search MLflow on Amazon.
MLflow is an open source platform for machine learning, LLM applications, and AI agents. Originally developed for machine learning lifecycle management, it provides experiment tracking, a model packaging format, a model registry, and model deployment functionality. Later releases expanded the platform to support LLM applications and AI agents, adding tools for observability, evaluation, prompt management, and LLM access control. Originally created by Databricks and first released in June 2018,[1] MLflow is licensed under the Apache License 2.0 and is a Linux Foundation project.[2][3]
Capabilities
Machine learning
For machine learning and deep learning workflows, MLflow covers the development lifecycle from experimentation to deployment. It provides experiment tracking for logging and comparing model parameters, metrics, and files across training runs. Models can be packaged in a standardized format compatible with frameworks including scikit-learn, PyTorch, TensorFlow, and Spark ML, and managed through a model registry.[4][5] MLflow also supports hyperparameter optimization and model serving via REST APIs.
AI agents and LLMs
MLflow 3.0, released in June 2025, added support for building and deploying AI agents and LLM applications.[6][7] According to project documentation, the release introduced tools for capturing execution traces compatible with the OpenTelemetry standard, an evaluation framework measuring response quality using LLM-as-a-judge scoring and human feedback, versioned storage for prompt templates, and an AI gateway for routing requests and controlling access to LLM providers such as OpenAI, Anthropic, and Google Gemini.[8][9]
History
| Version | Date | Notes |
|---|---|---|
| 0.1.0 | June 2018 | Initial release, announced at the Databricks Spark+AI Summit.[10][11][12] |
| 1.0 | June 2019 | Added loss curve tracking, performance improvements, and Windows support.[13][14][15] The Model Registry was introduced shortly thereafter.[16] |
| 2.0 | November 2022 | Introduced a model evaluation SDK, overhauled UI, and pre-built ML training pipelines.[17][18] |
| 3.0 | June 2025 | Added tracing, evaluation, prompt management, and gateway features for AI agents and LLM applications.[19][6] |
Adoption
MLflow is offered or supported by several vendors and cloud platforms. Databricks, the company that originally created MLflow, offers it as a managed service within its data and AI platform.[20] Amazon Web Services integrated MLflow within Amazon SageMaker,[21] and Azure Machine Learning supports MLflow tracking and model deployment natively.[22] Canonical released Charmed MLflow, an enterprise distribution for Ubuntu.[23] InfoWorld included MLflow in its annual Best Open Source Software awards in 2019[24] and 2021.[25]
References
- ↑ "MLflow 0.1.0". PyPI. 2018-06-05. Retrieved 2026-03-14.
- ↑ "LF AI & Data Landscape". Retrieved 2026-03-14.
- ↑ "MLflow is now a Linux Foundation project". InfoWorld. 2020-06-25. Retrieved 2026-03-14.
- ↑ Schlegel, Marius; Sattler, Kai-Uwe (2022). "Management of Machine Learning Lifecycle Artifacts: A Survey". ACM SIGMOD Record. Retrieved 2026-03-14.
- ↑ "Machine learning operations landscape: platforms and tools". Artificial Intelligence Review. Springer Nature. 2025. Retrieved 2026-03-14.
- ↑ 6.0 6.1 "Announcing MLflow 3.0". MLflow. 2025-06-09. Retrieved 2026-03-14.
- ↑ "Databricks Data + AI Summit 2025: Five takeaways for data professionals, developers". InfoWorld. 2025-06-01. Retrieved 2026-03-14.
- ↑ "MLflow Tracing". Retrieved 2026-03-14.
- ↑ "MLflow AI Gateway". Retrieved 2026-03-14.
- ↑ "Introducing MLflow: An Open Source Machine Learning Platform". Databricks. 2018-06-05. Retrieved 2026-03-14.
- ↑ "MLflow: A System for Machine Learning Lifecycle Management" (PDF). IEEE Data Engineering Bulletin. 2018. Retrieved 2026-03-14.
- ↑ "Databricks releases MLflow, runtime for ML and Databricks Delta at Spark + AI Summit". SD Times. 2018-06-06. Retrieved 2026-03-14.
- ↑ "MLflow 1.0.0". GitHub. Retrieved 2026-03-14.
- ↑ "Announcing the MLflow 1.0 Release". Databricks. 2019-06-06. Retrieved 2026-03-14.
- ↑ "Databricks wants one tool to rule all AI systems – coincidentally, its own MLflow tool". The Register. 2019-06-07. Retrieved 2026-03-14.
- ↑ "Introducing the MLflow Model Registry". Databricks. 2019-10-17. Retrieved 2026-03-14.
- ↑ "MLflow 2.0.1". GitHub. Retrieved 2026-03-14.
- ↑ "Announcing Availability of MLflow 2.0". Linux Foundation. Retrieved 2026-03-14.
- ↑ "MLflow 3.0". GitHub. Retrieved 2026-03-14.
- ↑ "MLflow guide". Databricks. Retrieved 2026-03-14.
- ↑ "AWS brings managed open source MLflow to Amazon SageMaker". VentureBeat. Retrieved 2026-03-14.
- ↑ "MLflow and Azure Machine Learning". Microsoft. Retrieved 2026-03-14.
- ↑ "Canonical Launches Charmed MLflow to Simplify Management and Maintenance of ML Workflows". InfoQ. 2023-10-01. Retrieved 2026-03-14.
- ↑ "The best open source software of 2019". InfoWorld. 2019-10-01. Retrieved 2026-03-14.
- ↑ "The best open source software of 2021". InfoWorld. 2021-10-01. Retrieved 2026-03-14.
External links
Category:AI software Category:Open-source artificial intelligence Category:Software using the Apache license Category:Artificial intelligence Category:Free and open-source software Category:Linux Foundation projects
This article "MLflow" is from Wikipedia. The list of its authors can be seen in its historical and/or the page Edithistory:MLflow. Articles copied from Draft Namespace on Wikipedia could be seen on the Draft Namespace of Wikipedia and not main one.
