GlassAlpha is an open-source toolkit for producing reproducible, regulator-ready audits of tabular machine-learning models. Licensed under Apache 2.0, it helps teams create transparent documentation and evidence for model review, with reports available in HTML and PDF. The software is currently in beta, and its core functionality is supported by extensive documentation and testing.

Through a command-line interface or Python API, GlassAlpha analyzes model performance, fairness, explainability, calibration, and robustness, then combines the results into a professional audit report. Its capabilities include group and intersectional fairness testing, dataset-bias detection, feature importance, individual prediction explanations, preprocessing verification, and a reproducibility trail recording relevant data, configuration, and software details. Additional commands generate adverse-action reason codes and counterfactual recourse recommendations.

GlassAlpha provides quick-start materials, interactive notebooks, configuration guidance, technical references, and examples covering credit, healthcare, and fraud detection. It is designed to run offline and produce deterministic reports for consistent review. Current limitations include a focus on tabular classification, limited regression support, and no distributed-computing capability.