A visual memory assistant that remembers where things are, answers everyday questions, and brings a trusted person to help when it matters.
Local-first and private by default. Built for our parents, on a single Acer GN100.
Not a cloud model answer. Reasoning, perception, speech, and memory run on the device.
· Runs on 
“Where did I leave my keys?”
“On the living-room coffee table at 10:42, based on the last confirmed placement. I have not confirmed a newer location.”
The problem
We build MEMO for our parents.
It helps them keep track of important belongings, such as keys or a wallet, so they can ask where something is and get its last confirmed location. It gives them a simple, hands-free way to ask everyday questions.
But not every situation should be left to AI. When judgment, reassurance, or human reasoning is especially valuable, MEMO connects our parents with family members they know and trust. It does not try to replace family support. It keeps that support within reach.
Our goal is to give our parents greater confidence and independence while giving their families peace of mind, and a trusted way to help when it matters.
What it does
Each experience is a real actor journey. The system does not just answer, it knows what it does not know.
Visual memory
A parent asks where an object is. MEMO returns the last confirmed placement, honest about whether that location is current, historical, ambiguous, or unknown.
Local assistant
A parent asks out loud. MEMO listens, reasons on device, and replies privately. No account, no cloud, no screen required.
Remote human help
A parent says “call my remote assistant.” A trusted family member’s phone rings. They join to see what the parent sees and speak with them.
See it in action
A real run on the glasses: register an object, ask where it is, then bring a remote helper in. It plays right here.
The stack
Every capability runs on one device. Private by default, no hosted model required. Each model links to its page.
Where it stands
Open an item to see the technical detail behind it.
The projected Console is the proof surface. MEMO is a memory aid, not a guarantee of an object’s current location, and not a safety-critical system.
How it runs
When help is needed, the wearer’s request opens a LiveKit room to a paired helper phone. Human judgment stays within reach.
Privacy by design
MEMO never stores full-length video in the backend. The media gateway retains no raw media, and perception keeps only a short, bounded window of frames it has already sampled and relayed, evicting anything older than that window. What survives is evidence.
The media gateway is built with a hard zero for raw media. The live stream is never recorded to the backend.
When the camera confirms an object such as a keychain, the system keeps a still frame and a short clip of that event window, a few seconds long, as the evidence of what it saw.
That evidence is stored with the placement in the local memory service. Ask later, “where is my keychain?” and MEMO shows the last confirmed location with the evidence to back it up when a text answer alone is not enough.
Everything else stays local: the raw stream, transcripts, and inference never leave the device. Keeping a minimal evidence clip is the deliberate exception that lets the memory answer honestly.