b stack radar
Metaflow logo

Metaflow

6.9/10

Open-source Python framework from Netflix for building, scaling, and deploying ML and AI workflows on your own cloud or Kubernetes infrastructure.

Data last updated November 2026 · Our methodology

Overview

Metaflow is an Apache 2.0 licensed Python framework that Netflix built for its own data science work and open-sourced in 2019. Data scientists write workflows as plain Python steps and debug them on a laptop. They can then send individual steps to cloud compute with GPUs, many cores, or large memory, and deploy the whole flow to a production scheduler without rewriting it. Metaflow versions variables and data between steps automatically, which gives teams experiment tracking and reproducibility. It also handles dependencies, including uv support. Recent releases added a checkpoint decorator for long training jobs, recursive and conditional steps for agentic systems, and AWS Trainium support. On AWS it can run on AWS Batch with Step Functions and S3, or on EKS. On Azure and Google Cloud it runs on AKS or GKE, with Blob Storage or Cloud Storage as its datastore. It also runs on any Kubernetes cluster, including on-premise clusters with MinIO, and can deploy to Argo Workflows or Airflow. The framework is free, and you pay only for the infrastructure it runs on. Outerbounds, the company behind the commercial managed platform, was acquired by Anaconda in April 2026 and is being folded into the Anaconda Platform. Commercial pricing is not published. Metaflow suits Python-first ML teams at startups and enterprises that already have a cloud account. Its main limitation is that it does not host anything itself. Your team has to set up and operate the compute, storage, Kubernetes, and scheduler underneath it, and its Airflow integration has fewer capabilities than the Argo and Step Functions integrations.

Our Rating

6.9
out of 10

AI Software7/10

Integration
7
Analytics
5
Customization
8
Ease of Use
6
Support
5
Cost
9
Scalability
8
Security
6

Hosting & Infrastructure6.9/10

Integration
7
Support
5
Cost
9
Customization
8
Security
6
Reliability
7
Ease of Use
6
Analytics
5
Scalability
8

Who Metaflow is best for

Best software for Developers

Good fit

  • ML engineers and data scientists write workflows in plain Python and deploy them from the command line.

  • Netflix runs it in production, and the commercial version is now part of Anaconda's enterprise platform.

  • It is free and installs with pip, though a team still has to set up its own cloud infrastructure.

Less suited

  • Product teams with ML features can run it on the cloud accounts they already have.

Pros & Cons

Pros

  • The same Python flow runs on a laptop and in production schedulers such as Argo Workflows or AWS Step Functions without code changes.
  • Variables and data are versioned automatically between steps, which gives teams experiment tracking without a separate tool.
  • It runs on AWS, Azure, Google Cloud, and on-premise Kubernetes, so teams are not tied to one cloud.
  • The framework itself is free under Apache 2.0, so the only costs are the cloud infrastructure it runs on.

Cons

  • It is not a hosting service; your team has to set up and run the compute, storage, and scheduler underneath it.
  • Outside AWS, every deployment runs on Kubernetes, so you need to operate a cluster.
  • Metaflow's own documentation says its Airflow integration has more limitations than the Argo and Step Functions integrations.
  • Commercial support and managed-platform pricing are not public, and the managed offering is being folded into Anaconda after the 2026 acquisition.

Key Features

API and CLI

Not included

  • Managed WordPress
  • Managed Kubernetes
  • Shared Hosting

Metaflow Pricing

Open Source

Free
  • Community support
  • Apache 2.0 license
  • You pay your cloud provider for compute and storage
  • Runs on your own AWS, Azure, Google Cloud, or Kubernetes infrastructure
Start Free

Managed platform (Outerbounds, now part of Anaconda)

Contact for pricing
  • Commercial managed Metaflow platform
  • No public pricing; the former Outerbounds pricing page now redirects to Anaconda
Get Started
Open interactive pricing calculatorGet a tailored cost breakdown for your team size, billing cycle, and add‑ons.

Frequently Asked Questions

Yes. The framework is open source under the Apache 2.0 license. You pay only for the cloud or Kubernetes resources it runs on.

It supports AWS through Batch and Step Functions or through EKS, Azure through AKS with Blob Storage, Google Cloud through GKE with Cloud Storage, and any Kubernetes cluster, including on-premise clusters with MinIO.

Netflix created it and hosts the GitHub repository. Outerbounds, the commercial company behind it, was acquired by Anaconda in April 2026, and Anaconda has said it will keep developing Metaflow as open source.

It can deploy to Argo Workflows, AWS Step Functions, and Apache Airflow. A local orchestrator is available for prototyping.

There was a managed platform from Outerbounds, which is now part of Anaconda. The former Outerbounds pricing page redirects to Anaconda, and no public price is listed.

Ranked near Metaflow in AI Software

See Metaflow in the list

Metaflow is #72 in AI Software by our score.

  1. #69
    Dograh logo
    Dograh
  2. #70
    fal logo
    fal
  3. #71
    Mural logo
    Mural
  4. #73
    tl;dv logo
    tl;dv
  5. #74
    MindStudio logo
    MindStudio
  6. #75
    Open WebUI logo
    Open WebUI