About SightNiq

The problem

People who are blind or have low vision get great help from general-purpose chatbots, but those tools forget every conversation the moment it ends. The third time you ask “which one is the blood pressure med?” the assistant has no idea — it doesn't remember the first two times. Every question starts from zero.

That's fine for a curious experiment. It's painful when it's daily life.

How SightNiq is different

SightNiq is built around a persistent memory layer. When you save a useful answer — “the rectangular bottle is blood pressure, the round one is allergy” — it sticks. The next time you ask about a bottle in the kitchen, Claude already knows.

Behind the scenes, every image goes through an optional object pre-pass before Claude sees it. That means hazards (a car, a step, a hot pan) get prioritised in the answer, not buried in a paragraph describing the scene.

The caregiver view

Family members, carers, or anyone you trust can see a read-only feed of what you've asked the app to help with. They don't need to install anything — it's a private URL that only people you invite can open. They can't change anything, just see.

Privacy posture

Raw images are not stored when you ask a one-off question — only the text question and Claude's text answer. Memories you explicitly save are kept, scoped to your account, behind Supabase Row Level Security. Caregiver links are one-token-per-invite and you can revoke them. See privacy for the long version.

What's next

Walking mode — continuous on-device hazard detection that interrupts audio to say “car ahead” — is wired in code and lights up once we ship the native build. Ray-Ban Meta glasses integration is slated for the sprint after.