Anthropomorphic AI: a toolkit for authoring and interacting with intelligent virtual agents for extended reality

In plain English

A University of Hamburg group released an open-source toolkit for building embodied virtual agents in XR, bundling speech, gaze, gesture, facial expression and vision with swappable foundation models and editable system prompts. Four developers in a pilot rated its usability as acceptable and all said they would use it again; of 43 people who talked to an agent built with it at IEEE VR 2025, the 34 who answered the survey rated likeability and ease of use highly, across roughly 120 minutes of dialogue and 429 interaction rounds. For practitioners this is a credible starting point for prototyping an embodied agent without assembling the speech-gaze-gesture pipeline yourself — though the evaluation is two small pilots, not a controlled study.