Abstract: Systems and methods for training an agent-based assistant model are provided. In embodiments, a method includes: obtaining biometric data of a user from a software application utilizing an assistant model that determines functions of the software application; filtering the biometric data based on predetermined categories, thereby extracting select biometric data; training a first version of the assistant model based on the select biometric data, thereby generating an updated assistant model; generating a summary of changes including changes to the first version of the assistant model that occurred during the training; and sending the summary of changes to a remote federated learning server, wherein the federated learning server trains a general version of the assistant model based on the summary of changes and other summary of changes received from computing devices of other users, thereby generating an updated general version of the assistant model.