Welcome to the open-source, Zimbra Machine Learning server for text and email classification. This server provides a solution for basic text classification requirements, including smartfolders by topic, sentiment, author, and social toxicity/obscenity, or an unlimited number of other classifications. All of this capability from defining a classifier to loading or creating a new vocabulary, training the classifier, and classifying content can be accessed through the GraphQL API or directly through the libraries. In the current version, classification is performed using a neural network designed by the server's original author and described below. The neural network is a multi-input convolutional graph without pooling that is the same for all types of tasks.