Eye gaze in the AAC reference app uses MediaPipe, served from the app's own site under vendor/mediapipe--/ by third_party/mediapipe/vendor.sh. MediaPipe Tasks Vision (@mediapipe/tasks-vision, the version in the directory name), by The MediaPipe Authors (Google). Licensed under the Apache License, Version 2.0 (LICENSE.txt beside this file). Source: https://www.npmjs.com/package/@mediapipe/tasks-vision (package.json "license": "Apache-2.0"), https://github.com/google-ai-edge/mediapipe Changed by the AAC build: in vision_bundle.mjs, the usage-metrics logger's flush() drops its queue instead of sending it to https://odml.pa.googleapis.com/v1/log (the library otherwise sends usage metrics to Google every minute and on close, with no option to turn this off; https://goo.gle/mediapipe-privacy, read 2026-09-29). No other file is changed. face_landmarker.task (face_landmarker/float16/1), unchanged. From https://storage.googleapis.com/mediapipe-models/face_landmarker/face_landmarker/float16/1/face_landmarker.task It bundles three models, each "Licensed under Apache License, Version 2.0" according to its model card (read 2026-09-29, linked from https://developers.google.com/edge/mediapipe/solutions/vision/face_landmarker): BlazeFace (short range) https://storage.googleapis.com/mediapipe-assets/MediaPipe%20BlazeFace%20Model%20Card%20(Short%20Range).pdf Face Mesh V2 https://storage.googleapis.com/mediapipe-assets/Model%20Card%20MediaPipe%20Face%20Mesh%20V2.pdf Blendshape V2 https://storage.googleapis.com/mediapipe-assets/Model%20Card%20Blendshape%20V2.pdf