Metaflow is an open-source Python framework for building, tracking and deploying machine learning and data science workflows, originally developed at Netflix and open-sourced in 2019. It supports local development with automatic result tracking, notebook-friendly workflows, deployment across AWS, Azure, GCP and Kubernetes with GPU and multi-core compute support, automatic variable versioning, and data warehouse integration, with recent additions including recursive and conditional workflow steps and checkpointing.
Automatic versioning and tracking of every variable and data flow across a workflow’s steps, without requiring a data scientist to build that instrumentation themselves, is what separates production-grade ML workflow tooling from a script that happens to work on one person’s laptop – reproducibility and experiment tracking are exactly where ad-hoc data science pipelines break down at scale. Real, named production adoption (23andMe, CNN, Realtor.com) alongside Netflix’s own internal use is a genuine, checkable signal of production-readiness for a workflow framework specifically, a category where a lot of tools look good in a demo but haven’t proven themselves at scale.
This is a free, open-source Python framework, not a hosted product with a signup flow or a support team on call – using it well requires real Python and ML engineering competence, and deployment/infrastructure setup is the user’s own responsibility unless paired with a managed platform. No pricing exists because there’s nothing to buy; the cost is entirely in the engineering time to adopt it well.
