CitrusX

Citrusx is a validation, explainability and monitoring platform for AI and large language model risk. It is aimed at teams that deploy models in regulated settings such as credit risk and need evidence that those models are accurate, robust, fair and compliant. The platform validates model performance, explains decisions globally for the whole model and locally for each prediction or group of predictions, and flags anomalies, vulnerabilities and heavy reliance on a small number of features that could make a model easy to manipulate. In production, it continuously monitors models for drift and degradation, sends live alerts when a model makes mistakes, and explains each new prediction in real time with its error and certainty. It suggests actions to reduce risk and generates customized reports for different audiences, including data scientists, data science managers, model risk managers, executives and regulators. Data scientists work through a web interface and an SDK. The vendor says its tools help with obligations under regulations such as the EU AI Act and AIDA, and it compares its explainability methods with SHAP and LIME. The vendor does not publish plan prices. Prospective customers are directed to contact the company. The website is behind a bot challenge that blocks some automated requests but loads normally in a browser.