TFX (TensorFlow Extended) is Google’s open-source, end-to-end platform for deploying production machine learning pipelines. With around 2,200 stars on GitHub and continued maintenance in 2026, it packages the practices behind Google’s own ML systems into reusable components that carry a model from data ingestion to serving.
A TFX pipeline is a sequence of components: ExampleGen ingests data, StatisticsGen and SchemaGen profile it, ExampleValidator catches anomalies, Transform engineers features consistently between training and serving, Trainer trains the model, Tuner optimizes hyperparameters, Evaluator gates quality, and Pusher deploys approved models. Pipelines can run on Apache Beam and be orchestrated with Airflow or Kubeflow.
TFX also includes ML Metadata, a library that records every artifact a pipeline produces — datasets, schemas, trained models, evaluation results — giving teams lineage and reproducibility for audits. Because TFX grew out of the platform Google uses for its own production models, its defaults reflect hard-won lessons about data drift, training-serving skew, and safe model rollouts, which is why it remains a reference architecture for production ML even outside the TensorFlow ecosystem.
TFX pipelines are defined in Python with the TFX SDK, and individual components can also be used standalone — many teams adopt only Data Validation or Transform before committing to the full platform. Interactive notebook execution is supported for developing pipeline components step by step, and the same definitions can then be exported to an orchestrator for scheduled production runs.
TFX is free and open source under the Apache-2.0 license. You only pay for your own compute and storage infrastructure.
ML engineers and platform teams productionizing TensorFlow models who need repeatable, validated, end-to-end pipelines rather than notebook workflows.