Source code for the X Recommendation Algorithm (github.com)

X has open-sourced the comprehensive codebase behind its Recommendation Algorithm, the technology driving personalized content feeds across key product surfaces such as the For You Timeline, Search, Explore, and Notifications. This release offers the AI/ML community unprecedented insight into how X curates and ranks content by leveraging a complex interplay of data services, machine learning models, and software frameworks. Core components include real-time user action streaming, diverse embedding models like SimClusters and TwHIN for community and knowledge graph representations, and advanced ranking systems combining lightweight and heavy neural network rankers. Significantly, the repository showcases both candidate generation methods—such as graph-based user-post interaction traversal via GraphJet—and post-ranking strategies essential for relevance and user engagement optimization. Frameworks like Rust-based navi for model serving and product-mixer for feed construction highlight X’s architectural emphasis on performance and modularity. This transparency enables researchers and developers to study production-grade recommendation systems, experiment with state-of-the-art multi-task learning models used in notification ranking, and contribute to continuous improvement through community collaboration. By inviting open contributions and integrating feedback from global experts, X’s initiative fosters a collaborative ecosystem aimed at enhancing recommendation quality, fairness, and safety. It also signals a move towards greater openness in large-scale, real-time recommendation engineering—making this release a valuable resource for advancing research and innovation in personalized content delivery and user experience optimization.