GlassAlpha is an open-source toolkit for creating reproducible, regulator-ready audit reports for tabular machine-learning models. Licensed under Apache 2.0, it helps teams document model performance, fairness, explainability, calibration, and robustness through a command-line interface and Python API. Audits produce professional HTML or PDF reports, along with evidence packs and audit trails that record details such as configuration, data hashes, software versions, and random seeds.

The toolkit evaluates standard performance measures, group and intersectional fairness, individual consistency, dataset bias, and calibration. It can also provide feature importance, explanations for individual predictions, preprocessing verification, counterfactual recourse recommendations, and ECOA-related adverse-action reason codes. Users can start with built-in examples or audit models using their own data, and the documentation includes guides for machine-learning engineers, data scientists, compliance officers, and validators.

GlassAlpha is currently in beta, with functionality focused mainly on classification and tabular data. It runs offline and supports repeatable reporting, though PDF output may differ across operating systems. The project provides setup instructions, examples, technical references, and support through GitHub.