Permissioned voice + vision agent loop: point and ask, explicit approval, no raw audio retained.
Jarvis is the multimodal intelligence loop inside the Portals platform — a permissioned, voice-first assistant that can see what you are looking at and act on it.
Capabilities span voice and vision understanding (ask about what is in front of you, hands-free), open-vocabulary “point & ask”, and raycast pointing for near and far selection — across mobile, headset and the web.
The design is built around explicit approval and minimal retention: it reads only the context it has been granted, asks before it sends, edits, schedules or publishes, and does not retain raw audio or screen frames. Every capability is held to evidence — a requirement counts as verified only when a named test cites it.
Jarvis is one third of the platform trio described in the CVPR 2026 work: Portals · XRAI · Jarvis.
