LUMENAA — Edge-Native Assistive Vision Agent

LUMENAA is an assistive AI agent for blind and low-vision users. It describes what's ahead, reads labels and text, flags hazards, and remembers where objects were last seen — the kind of help that bears on independence and safety, not just convenience.

Vision-language models make this possible, but two things block dependable real-world use. Running everything in the cloud is expensive, slow, and useless offline while a person is on the move. And these models sometimes state things confidently that aren't true — a wrong distance, a misnamed person — which a blind user has no easy way to catch.

LUMENAA is built around one idea: do the least the situation actually needs, and only claim what it can back up. It answers each request at the cheapest capable level on the device, and only reaches out to the cloud when the on-device answer genuinely isn't enough. On the trust side it never invents a distance — it states only what it can verify, hedges when it's unsure who or what it's looking at, and always voices hazards first.

On 150 real questions written by blind users, LUMENAA handled about half of them fully on-device with no meaningful loss in accuracy. A lightweight visual memory means it rarely re-describes the same object, cutting repeat cloud calls by roughly 90%. An audit of every answer found zero invented distances — and these cost and safety guarantees held across five hardware tiers, from a desktop down to a 4-watt iPhone.