@@ -36,7 +36,7 @@ For example:
 
 If you only look vertically, these sessions appear independent. But if you scan horizontally, you'll find that A, C, and E all belong to the "registration/login" theme.
 
-**Theme tracing** is: treat each turn on the timeline as a cylinder, where the height represents the turn's computation/thinking depth, and the color/tag represents the theme. We can even mine from Kimi Code's already compressed and archived context, perform deep horizontal scanning, find cylinders (turns) of the same color, re-establish their associations, form a theme, and store it in memory.
+**Theme tracing** is: treat each turn on the timeline as a cylinder, where the height represents the turn's computation/thinking depth, and the color/tag represents the theme. We can even mine from Kimi Code's already compressed and archived context, perform deep horizontal scanning, find cylinders (turns) of the same color, re-establish their associations, form a theme, and store it in memory. Three weeks later, when you're about to upgrade your "registration/login" module or fix a bug, you don't have to re-explain the original design rationale to an agent whose context has been compressed countless times. Instead, your agent can say: "I just traced the 'registration/login' theme. Based on the design decision from three weeks ago and the two adjustments made since then, we should use solution B for the current problem."
 
 > The diagram below shows how `kimi-memory` views conversation history: vertical bars are turns on the timeline, thick horizontal lines are clusters, gray boxes are sessions, and colored brackets connect related turns/clusters across sessions into theme lines.
 
 
@@ -37,7 +37,8 @@
 
 如果只看纵向，这些 session 彼此独立。但如果横向扫描，会发现 A、C、E 都属于"注册登录"这一主题。
 
-**主题追溯**就是：把时间线上的每个 turn 看作一个圆柱，圆柱的高度代表该 turn 的计算/思考深度，颜色/标签代表主题。我们甚至可以从kimi code已经压缩归档的上下文中再次挖掘，进行深度横向扫描，把同色圆柱（对话轮turns）找出来，重新建立关联，组成theme并存入记忆。
+**主题追溯**就是：把时间线上的每个 turn 看作一个圆柱，圆柱的高度代表该 turn 的计算/思考深度，颜色/标签代表主题。我们甚至可以从kimi code已经压缩归档的上下文中再次挖掘，进行深度横向扫描，把同色圆柱（对话轮turns）找出来，重新建立关联，组成theme并存入记忆。当我们三周以后准备升级的我们的“注册登录”模块，或者修复某个bug的时候，不是对着已经压缩过无数次上下文的agent，重新描述对其当时的设计思路和方案，而是你的agent说：“刚刚通过trace_theme("注册登录")这个主题，发现根据3周前的设计决策和中间经历的2次调整改进，现在的问题我们应该用方案B。”
+
 >下图展示了 `kimi-memory` 如何看待对话历史：竖条是时间轴上的 turn，粗横线是簇，灰框是 session，彩色括号把跨 session 的相关 turns / clusters 串联成主题线。
 
 ![时间轴上的 turns、簇、session 与主题追溯](./assets/contextFlow.svg)