Alibi is an open-source Python library from Seldon that implements a broad collection of algorithms for explaining machine learning models. The GitHub repository has gathered around 2,650 stars and was last actively updated in 2025. Where many explainability tools cover a single method, Alibi brings black-box, white-box, tabular, text, and image explanations together under one consistent API, with a focus on production-quality implementations.
The library includes well-known techniques such as Anchors, integrated gradients, Kernel SHAP, counterfactual explanations, and contrastive methods, each returning structured explanation objects that are easy to visualize and serialize. Methods work with models from scikit-learn, TensorFlow, PyTorch, and anything exposing a predict function, and many scale to realistic dataset sizes. Comprehensive documentation and examples make it a reference implementation frequently cited in explainable-AI research and industry practice alike.
Alibi’s maintainers place unusual weight on correctness and documentation for a research-adjacent library: each algorithm ships with theory notes, references, and guidance on when it should and should not be used. The project also spawned the separate Alibi Detect library for outlier and drift detection, reflecting Seldon’s broader focus on trustworthy ML operations. For practitioners, Alibi is often the first serious stop when an explanation feature needs to survive contact with real users and auditors.
Alibi is free and open source. It is maintained by Seldon as a community library with no paid tier; enterprise MLOps offerings from Seldon are separate products.
Data scientists and ML engineers who need to explain model predictions for debugging, stakeholder trust, fairness reviews, or regulatory documentation.