Netflix open‑sources Metaflow

Machine learning operations software is the plumbing that moves an artificial intelligence project from a laptop to production servers. Netflix has released its Metaflow framework as open-source code on GitHub under the Apache 2.0 license. (github.com) Netflix’s public repository says Metaflow now supports more than 3,000 artificial intelligence and machine learning projects inside the company. The same README says those projects run hundreds of millions of compute jobs, process petabytes of data, and manage tens of petabytes of models and other artifacts. (github.com) Metaflow’s core pitch is that a team can write a workflow once in Python, test it locally, and then send heavier steps to cloud infrastructure. Its documentation says the system handles versioning, dependency management, orchestration, and storage so users can track what code, data, and compute produced a result. (docs.metaflow.org 1) (docs.metaflow.org 2) The release lands as companies try to turn more prototype artificial intelligence work into repeatable products. Netflix said in a November 2025 engineering post that Metaflow works with Maestro, its workflow orchestrator, which powers nearly every machine learning and artificial intelligence system at the company. (netflixtechblog.com) That makes Metaflow part of a larger shift in artificial intelligence tooling: companies want the flexibility of notebooks and scripts, but they also need audit trails, scheduling, and access to large pools of central processing units and graphics processing units. Metaflow’s documentation says users can start with local Python code and then mix in cloud-scale compute patterns as workloads grow. (docs.metaflow.org) Netflix has described Metaflow as an open-source project before, but the current GitHub materials present a broader picture of outside use. The repository says the software powers projects at Amazon, DoorDash, Dyson, Goldman Sachs, Ramp, and other companies. (github.com) Outside case studies back up at least part of that adoption claim. Outerbounds, a company built around commercial services for Metaflow, says Amazon Prime Video used the framework to speed experimentation and scale machine learning workflows. (outerbounds.com) Netflix has been building around Metaflow for years. In a 2024 post, the company said its machine learning platform team uses Metaflow as the foundation for tools that help data scientists and machine learning practitioners build and manage different kinds of production systems. (netflixtechblog.com) The practical bet is simple: if more companies adopt the same workflow layer Netflix uses internally, Metaflow can move from a Netflix tool to a wider standard for running data-heavy artificial intelligence systems. The code is now public, the documentation is live, and the next test is whether outside teams contribute to it. (github.com) (docs.metaflow.org)