MemoRing
中文中 ENEN 日本語日

Science & research

AI has a model
of the world. People don't.

Understanding a person means reading three things at once: memory, emotion, and cognition. This page is about how we study people — the research itself, not what a product can do.

Research illustration: measurement axes with a network of memory nodes

Three lines have to hold at once

AGI is already learning how the world works, but there is still no independent model of a person. We call that goal The First Human World Model — not a system that knows the world better, but one that understands one particular human being.

Memory

Memory is continuous, not a pile of isolated records. Only once fragments are connected across long stretches of time does understanding a person become possible — which is why long-term memory is our primary research direction.

Emotion

The same sentence means different things in different tones. Emotion isn't metadata attached to a memory; it's a necessary dimension of meaning, and one of the markers of how strongly something is held.

Cognition

How people group concepts, form associations, and forget — cognitive structure decides what form a memory should be stored in, and what form a machine should hand it back in.

Disclosure

Shared-origin research depth — not a third-party endorsement.

The research on this page comes from a body of work that shares our origins: the Tianqiao and Chrissy Chen Institute (TCCI) was founded by Tianqiao Chen and Chrissy Luo, and has publicly documented funding and personnel ties to Shanda Group — the institute states as much on its own site.

So please read this page as our own research depth, not as an independent third party vouching for MemoRing. We state it plainly on purpose: on a question like this, saying nothing is the less credible option.

We're not a startup that bought a model and wrapped a device around it. Behind this ring sits a decade of sustained investment in brain science and AI research.

Three bodies of work

Each is written to the institute's own published wording. We publish only what can be checked: where the institute has not disclosed something — start year, principal investigator, journal, identifier — we leave it blank rather than fill it in.

  1. Asset 01 · Long-term memory models

    Making long-term memory the foundation of AI

    1st placeGAIA leaderboard
    40.53%success rate
    2024-10-23date achieved
    arXiv:2410.15665paper

    The OMNE multi-agent framework took the top spot on the Hugging Face GAIA leaderboard on 2024-10-23 with an overall success rate of 40.53%; the accompanying paper is "Long Term Memory: The Foundation of AI Self-Evolution" (arXiv:2410.15665). The same line of work also includes the open-source frameworks MiroMind and Miroflow. The institute has additionally donated RMB 50 million to Fudan University to establish a research institute whose primary focus is long-term memory.

    Anyone can verify this one independently: arXiv:2410.15665 →

  2. Asset 02 · Ageing and cognition cohort

    The Shanghai Aging Study

    nearly 4,000Shanghai residents
    over a decadefollow-up
    memory · thinking · brain scansdata collected

    The Shanghai Aging Study: nearly 4,000 Shanghai residents followed for over a decade, with data covering memory, thinking, and brain scans.

    This is a research project concerned with memory and cognition in older adults. It has no relationship to what MemoRing does, and makes no claim of any kind about the product.

  3. Asset 03 · Sleep EEG

    Brain-electrical data recorded during sleep

    nearly 6,000recordings

    Nearly 6,000 sleep EEG recordings.

    On the product side this is inspiration only: part of how we think about the way memories get organised comes from the perspective of sleep research. No product capability is implied here.

From multimodal memory to a Human World Model

There's only one throughline: turn a person's memory into a layer that can be computed on, that keeps growing, and that the person themselves controls.

  1. Track 01

    Multimodal memory input

    Speech, text, tone, emotional signal — the raw material comes from your actual day, not from samples collected in a lab.

  2. Track 02

    The Memory Layer

    The real asset was never the hardware. Ring, card, glasses — those are doorways. The memory layer is the thing being built for the long run, and it belongs to you.

  3. Track 03

    The memory graph

    Memories linked by person, event, and topic, so they can be searched, reasoned over, and grown. Isolated records sink; connected memories compound.

  4. Track 04

    Human World Model

    A model of a person: the prerequisite for moving from embodied intelligence to embodied AGI. The Digital Self will become infrastructure in the age of AI — and it has to be built on memory you control.

Four kinds of people in one room

Brain science asks the right question, AI turns the question into a model, hardware makes sure the data comes from real life, and product makes the result useful to an ordinary person.

Brain scienceresearch backgrounds in memory, emotion, and cognition
Artificial intelligencelong-term memory and multi-agent systems
Hardwarelow-power devices you can actually wear all day
Productturning research into something people want to use daily

This page describes research work. MemoRing is a consumer product and is not intended to diagnose, treat, or prevent any disease.

Be first to MemoRing

Pricing and launch details are coming soon. Leave your email and you'll hear it here first — waitlist members get priority access.

We'll only use your email for MemoRing updates. No spam, no sharing with third parties. Ever.