The Algorithm ML

The Algorithm ML is the open-source machine learning codebase behind Twitter’s recommendation system, published by Twitter with more than 10,600 stars on GitHub as part of its algorithm transparency release.

Overview

In 2023 Twitter did something no major social platform had done: it published the actual source code of the system that decides what hundreds of millions of people see. This repository is the machine-learning half of that release — the model code used to rank and recommend content for the Home timeline and related surfaces. It contains production ML projects rather than toy examples: representation learning that embeds users and content into shared spaces, community and similarity models that group accounts by interest, and the training and scoring code used to predict engagement. For engineers, it is a rare look at recommendation at true industrial scale, with the compromises and engineering patterns that scale forces on you. For researchers and the public, it is primary-source material on how a major platform’s feed actually worked. The repository is released under the AGPL-3.0 license; the codebase has been quiet since 2024, so treat it as a reference implementation and a historical artifact of a production system rather than a living framework.

Key Features

• Production recommendation models – the real ML code behind Twitter’s Home timeline ranking.
• Representation learning – embedding models that place users and content in shared spaces.
• Community detection – similarity models that group accounts into interest communities.
• Industrial-scale patterns – training and serving code shaped by hundreds of millions of users.
• Transparency primary source – study exactly how a major feed algorithm was implemented.
• Full training pipelines – not just model definitions, but the surrounding ML infrastructure.
• AGPL-3.0 open source – the complete codebase, free to read, run, and learn from.

💳 Pricing

The Algorithm ML is completely free and open source under the AGPL-3.0 license. Anyone can download, study, and run the code; there is no commercial product or paid tier attached to it.

Who Is It For

Machine learning engineers and researchers who want to study recommender systems at real production scale, and technically curious readers who want to see — in code, not in press releases — how a major social feed was ranked.