https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/技术 × 产品 × 机器人Hugo 0.160.1 & FixIt v0.4.6-20260517062719-ed8ec5c2zh-CNruohuyang@163.com (Cygnus Yang)ruohuyang@163.com (Cygnus Yang)Sat, 23 May 2026 00:00:00 +0800https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E7%AC%AC%E4%B8%80%E7%AB%A0-gbrain%E6%98%AF%E4%BB%80%E4%B9%88/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E7%AC%AC%E4%B8%80%E7%AB%A0-gbrain%E6%98%AF%E4%BB%80%E4%B9%88/<h2 class="heading-element" id="从问题说起"><span>从问题说起</span> <a href="#%e4%bb%8e%e9%97%ae%e9%a2%98%e8%af%b4%e8%b5%b7" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><h3 class="heading-element" id="传统知识管理为什么失败"><span>传统知识管理为什么失败</span> <a href="#%e4%bc%a0%e7%bb%9f%e7%9f%a5%e8%af%86%e7%ae%a1%e7%90%86%e4%b8%ba%e4%bb%80%e4%b9%88%e5%a4%b1%e8%b4%a5" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h3><p>知识管理软件已经存在 30 年，但绝大多数人仍然无法有效维护自己的知识库。为什么？</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E7%AC%AC%E4%BA%8C%E7%AB%A0-gbrain%E8%83%BD%E5%81%9A%E4%BB%80%E4%B9%88/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E7%AC%AC%E4%BA%8C%E7%AB%A0-gbrain%E8%83%BD%E5%81%9A%E4%BB%80%E4%B9%88/<h2 class="heading-element" id="核心能力全景"><span>核心能力全景</span> <a href="#%e6%a0%b8%e5%bf%83%e8%83%bd%e5%8a%9b%e5%85%a8%e6%99%af" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><p>GBrain 不是单一功能工具，而是一个完整的能力矩阵，覆盖知识管理的全生命周期。</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E7%AC%AC%E4%B8%89%E7%AB%A0-gbrain%E5%A6%82%E4%BD%95%E5%B7%A5%E4%BD%9C/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E7%AC%AC%E4%B8%89%E7%AB%A0-gbrain%E5%A6%82%E4%BD%95%E5%B7%A5%E4%BD%9C/<h2 class="heading-element" id="整体架构"><span>整体架构</span> <a href="#%e6%95%b4%e4%bd%93%e6%9e%b6%e6%9e%84" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><p>GBrain 采用四层架构设计，每层有明确的职责边界。</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E7%AC%AC%E5%9B%9B%E7%AB%A0-%E6%A0%B8%E5%BF%83%E6%A6%82%E5%BF%B5%E6%B7%B1%E5%BA%A6%E8%A7%A3%E6%9E%90/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E7%AC%AC%E5%9B%9B%E7%AB%A0-%E6%A0%B8%E5%BF%83%E6%A6%82%E5%BF%B5%E6%B7%B1%E5%BA%A6%E8%A7%A3%E6%9E%90/<h2 class="heading-element" id="知识模型两层页面结构"><span>知识模型：两层页面结构</span> <a href="#%e7%9f%a5%e8%af%86%e6%a8%a1%e5%9e%8b%e4%b8%a4%e5%b1%82%e9%a1%b5%e9%9d%a2%e7%bb%93%e6%9e%84" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><p>每个知识页面由两部分组成，用 <code>---</code> 分隔。</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E7%AC%AC%E4%BA%94%E7%AB%A0-%E5%AF%B9%E6%99%BA%E8%83%BD%E4%BD%93%E7%9A%84%E5%90%AF%E5%8F%91/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E7%AC%AC%E4%BA%94%E7%AB%A0-%E5%AF%B9%E6%99%BA%E8%83%BD%E4%BD%93%E7%9A%84%E5%90%AF%E5%8F%91/<p>前四章介绍了 GBrain 是什么、能做什么、如何工作，以及它背后的核心概念。真正重要的并不只是 GBrain 这套实现本身，而是它揭示了一个更普遍的判断：</p> <p><strong>当前智能体的核心瓶颈，已经越来越不是“不会推理”，而是“不会稳定地积累、组织、调用过去的经验”。</strong></p> <p>传统知识管理失败，是因为维护成本太高；传统智能体不稳定，是因为每次都像重新开始。GBrain 的价值，在于把这两件事连接起来：让知识维护不再依赖人的持续勤奋，而成为系统正常运行的副作用。</p> <h2 class="heading-element" id="从-gbrain-提炼出的五个判断"><span>从 GBrain 提炼出的五个判断</span> <a href="#%e4%bb%8e-gbrain-%e6%8f%90%e7%82%bc%e5%87%ba%e7%9a%84%e4%ba%94%e4%b8%aa%e5%88%a4%e6%96%ad" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><h3 class="heading-element" id="1-记忆系统比单轮聪明更重要"><span>1. 记忆系统比单轮聪明更重要</span> <a href="#1-%e8%ae%b0%e5%bf%86%e7%b3%bb%e7%bb%9f%e6%af%94%e5%8d%95%e8%bd%ae%e8%81%aa%e6%98%8e%e6%9b%b4%e9%87%8d%e8%a6%81" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h3><p>当模型能力持续提升后，很多智能体的真正短板不再是“答不出来”，而是：</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E6%80%BB%E7%BB%93/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/03-gbrain/%E6%80%BB%E7%BB%93/<p>GBrain 通过以下核心设计实现将知识管理维护成本降至近乎零的个人智能系统：</p> <h2 class="heading-element" id="五大核心设计"><span>五大核心设计</span> <a href="#%e4%ba%94%e5%a4%a7%e6%a0%b8%e5%bf%83%e8%ae%be%e8%ae%a1" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><pre><code>graph TB A[&#34;MECE 目录&lt;br/&gt;单页单实体&#34;] B[&#34;两层页面模型&lt;br/&gt;编译真理&#43;时间线&#34;] C[&#34;自动丰富&lt;br/&gt;信号触发&#34;] D[&#34;混合检索&lt;br/&gt;向量&#43;关键词&#43;RRF&#34;] E[&#34;技能系统&lt;br/&gt;25 个可组合技能&#34;] F[知识管理维护成本降至近乎零] --&gt; A F --&gt; B F --&gt; C F --&gt; D F --&gt; E style A fill:#f9f,stroke:#333,stroke-width:2px style B fill:#f9f,stroke:#333,stroke-width:2px style C fill:#f9f,stroke:#333,stroke-width:2px style D fill:#f9f,stroke:#333,stroke-width:2px style E fill:#f9f,stroke:#333,stroke-width:2px style F fill:#fff,stroke:#333,stroke-width:4px</code></pre><h3 class="heading-element" id="1-mece-目录"><span>1. MECE 目录</span> <a href="#1-mece-%e7%9b%ae%e5%bd%95" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h3><ul> <li>单页单实体，防止知识腐烂</li> <li>每个知识片段通过决策树落入精确一个目录</li> <li>无重复页面，无歧义</li> <li>inbox/ 作为模式演进标志</li> </ul> <h3 class="heading-element" id="2-两层页面模型"><span>2. 两层页面模型</span> <a href="#2-%e4%b8%a4%e5%b1%82%e9%a1%b5%e9%9d%a2%e6%a8%a1%e5%9e%8b" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h3><ul> <li>编译真理：可重写，始终当前，结构化状态</li> <li>时间线：仅追加，不可重写，反向时间序证据</li> <li>合成预计算，不像 RAG 每次查询都从头推导</li> </ul> <h3 class="heading-element" id="3-自动丰富"><span>3. 自动丰富</span> <a href="#3-%e8%87%aa%e5%8a%a8%e4%b8%b0%e5%af%8c" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h3><ul> <li>每个信号触发丰富流水线，无需人工记忆</li> <li>分层策略：Tier 3（存根）→ Tier 2（Web+社交）→ Tier 1（完整）</li> <li>重要度自动提升，无需显式告知</li> </ul> <h3 class="heading-element" id="4-混合检索"><span>4. 混合检索</span> <a href="#4-%e6%b7%b7%e5%90%88%e6%a3%80%e7%b4%a2" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h3><ul> <li>向量搜索：HNSW 余弦相似度，语义理解</li> <li>关键词搜索：tsvector 全文搜索，精确匹配</li> <li>RRF 融合：兼顾两者优势</li> <li>4 层去重：源感知 + 编译真理保证</li> </ul> <h3 class="heading-element" id="5-技能系统"><span>5. 技能系统</span> <a href="#5-%e6%8a%80%e8%83%bd%e7%b3%bb%e7%bb%9f" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h3><ul> <li>25 个可组合技能</li> <li>技能路由系统（RESOLVER.md）</li> <li>链式调用机制</li> <li>Always-on 技能（signal-detector, brain-ops）</li> </ul> <h2 class="heading-element" id="系统自成长机制"><span>系统自成长机制</span> <a href="#%e7%b3%bb%e7%bb%9f%e8%87%aa%e6%88%90%e9%95%bf%e6%9c%ba%e5%88%b6" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><pre><code>flowchart TD A[用户操作] --&gt; B[技能处理] B --&gt; C[信号捕获] B --&gt; D[实体提取] C --&gt;|并行| E[丰富触发] D --&gt; E E --&gt; F[知识更新] F --&gt; G[索引同步] G --&gt; H[知识库被动增长] I[时间流逝] --&gt; J[重要度提升] J --&gt; K[自动丰富升级] K --&gt; H L[下次查询] --&gt; M[上下文更丰富] M --&gt; H style H fill:#e6f3ff,stroke:#fff,stroke-width:3px style I fill:#95d5b2,stroke:#fff,stroke-width:2px style L fill:#95d5b2,stroke:#fff,stroke-width:2px</code></pre><p>系统在睡眠中自我成长——每个周期都增加知识。智能体在会议后丰富人物页面，下次该人物出现时，智能体已具备上下文。差异每日累积。</p>