云玦·Not Wearable. It's Awarer
/

Human RSI is
our infinite game

Human RSI 是一场无限游戏

RSI trends are all about models. Could it be part of you?

当前自进化的主体和对象都是模型,能不能是人自己?

Build a no-setup, zero-app, personalized wearable agent system.

研发一种零配置,零预装,极致个人化的可穿戴智能体系统。

It fuses body, ambient, speech and preferred context into your own RSI model.

多模态融合身体、环境、语言和数字上下文偏好,每个人有自己能部署和迭代的RSI模型。

Not far off, everyone owns them under $200…

无需太久,200美金/用户就能开启……

01 / DEMO

Our Way In

以充满争议的形态验证原型和原理

Findings发现
7 天可穿戴评测视频封面 Bilibili · 7-day test · 18:49B站 · 7 天评测 · 18:49
2M+
saw one demo人看过一个 demo
5,000
danmu comments条弹幕评论
500+
into research进入典型研究
  1. Ugliest wearable of the year. 7-day test, trending on Bilibili & Douyin.

    一块今年最丑可穿戴的7天评测,在B站抖音自然进入热门精选。

  2. The debate wasn't health or focus, but AI seeing the tester's blind spots.

    观众的兴趣和争议,不是健康、提效,而是AI觉察了测试者自我遮蔽的模式。

  3. Nothing due through Q1 2027 has that agency, or multimodal and all-week wear.

    至27年1季度的我们所知待发售产品,在主动智能上落差很大,多模态 和 持久戴 也难以两全。

  4. US: top 30–40% of long-form content chases protocol and faster recursion.

    美国头部30-40%的创作者长内容消费,都是在探讨更好的想法、做法,加速自我递归。

Approach做法
  1. Early Fusion Memory — not an open-source stack.

    早融合记忆架构,有别于任何开源记忆工程。

  2. No Setup / No App / Zero Skill — create unique loops for you.

    零预装、自进化的智能体和硬件体验,一切能力都为一个人生成。

  3. Nearly done: 1st hardware — ultra-low power, proactive, instant memory.

    接近完成第一款超低能耗、完全主动、即时记忆的硬件定义。

02 / Wearable

Wearables have never been recursive

可穿戴产品从未改变你

Symptoms现象
  1. Open Oura or Whoop. It shows the metrics — does anyone know how to improve?

    打开 Oura /Whoop App,它展示指标之后,用户知道可以怎么改善吗?

  2. Add Plaud or Looki plugin into ChatGPT Health or Claude. Will you set up loops?

    +Plaud/Looki 接入 ChatGPT Health/Claude Loops,你会仔细设置循环任务吗?

  3. Coach, dietitian, doctor, mindful shows — can Apple Watch make it your protocol?

    教练、营养师、医生、心灵播客说的,Apple Watch能不能让你在日常实践反思?

  4. AI has never remembered or absorbed your taste or state. It only reads input.

    AI 从未觉察和习得你的品味和状态,都是读取你的输入。

Thesis想法
  1. Body, brain, ambient, context, wearable — a recursive loop has to be built.

    身体、大脑、环境、上下文、可穿戴,需要建立这样的递归循环。

  2. Data volume sets tunable parameters. Battery has to hold always-on.

    数据量决定了可调节参数量:续航一定要能Always-on。

  3. Personalization sets reward quality. So it has to be a Zero Skill start.

    个性化决定了Reward质量:一定是无预装的Zero Skill Start。

  4. Health, focus, creation — if AI's gain is first-order, you become the overseer.

    无论健康、效率、创作,如果AI只是做一阶加速,人就会沦为监工。

03 / Model

Why not Mem, PEFT, LoRA, or another SecondMe?

为什么不是Mem、PEFT、LoRA或各种SecondMe?

Tool library size across evolution steps
Tool library size across evolution steps工具库规模随进化步数的增长
Findings发现
  1. Opus 4.6 wrote wisely, 5.0 only codes. Mourning GPT-4o still trends in the US.

    Opus 4.6 写作绝佳,5.0只能代码。至今怀念GPT4o是美区热门话题。

  2. Dense models come close on domain skills, at a fraction of the cost.

    Dense model在特定能力上接近,人均后训练成本显著降低。

  3. MCP collections like OpenConnector rewire internet context.

    类似OpenConnector等MCP集合重新拉通互联网上下文。

  4. No other memory project carries multi-modal, real-time human alignment.

    Mem0等开源记忆项目,都无法承载多模态、实时的人类数据对齐。

  5. Hazy Cartridges, Transformer² — far from any consumer product feasibility.

    Hazy Cartridge、Transformer²,远没有消费产品工程可行性。

Approach做法
  1. SOTA model steers, follows expert recipes, patrols loops.

    最强模型发现和创造Skill,遵循专家意见和巡视问题。

  2. Dense models move from a cohort to a person.

    Dense Model 会从人群、领域,走向个人化。

  3. Reward comes from real context trajectories, and gamified preference.

    reward 来自真实的上下文轨迹,也来自游戏化互动中的偏好。

  4. A month of fused memory may hold 100x the tunable parameters of open work.

    1个月融合记忆形成的可调节参数,可能是现有学术和开源项目的100x。

  5. Privacy and a model of your own are one. More profitable, scalable revenue.

    隐私诉求和想要专属模型是一体两面,激活更高利润更持续的收入流。

04 / Team

Three threads of experience, now awakening

大家经历的三条线,现在撞到一起

2019 2020 2021 2022 2023 2025
Multimodal & self-evolving tools多模态与自进化工具
Personalization & trust个性化与人类信任
Consumer hardware消费硬件
2019

1bn USD Rev Management

10 亿美元营收盘子

Huawei India — the only full-stack unit outside China.

华为印度:消费者业务在中国以外,唯一产研销一体的国家公司。

2020

Multi-token prediction

Multi-token prediction

Now core to DeepSeek V4 and Qwen3 / 4-Next inference.

后来成了 DeepSeek V4、Qwen3/4-Next 的核心推理架构。

2021

M6 & PLUG, first adopter

M6 & PLUG 首个业务方

First Alibaba keynote as head of Intelligent Connectivity.

第一次在阿里发布会上,以智能互联总裁身份出场。

2023

Visual ChatGPT · 34k stars

Visual ChatGPT · 34k star

Core dev at Microsoft. First LLM tool-orchestration stack.

为微软做核心开发,当时第一个 LLM 多模态编排工具。

2023

Qwen's first personalized LLM

Qwen 第一个个性化大模型

For games and hardware. Its alignment set still ranks.

面向游戏与硬件;配套的对齐数据集至今在 ModelScope 前列。

2025

Ring and band — leading product experience

戒指和手环,都有领先产品经验

Software lead from RingConn,
hardware lead from Amazfit.

软件产品负责人来自 RingConn,
硬件负责人来自华米 Amazfit。

2026

Zero-Skill self-evolution

Zero-Skill 自进化

Our first tech report.

我们的第一篇技术报告。

2026 · Yunjue2026 · 云玦

The three threads converge

三条线合流

Scroll the chart sideways →图可以左右拖 →