[FEATURE]: Collapsible thinking steps with AI-generated summary titles to reduce Desktop lag (inspired by Trae) #29212

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

English：
Is your feature request related to a problem? Please describe.
The OpenCode Desktop app experiences noticeable lag when the "thinking process" output grows long. Rendering a massive, permanently expanded text block consumes resources and makes the UI sluggish. Additionally, even when expanded, it is hard to quickly understand what the model is currently doing without reading through the full output.

Describe the solution you'd like
I would like the thinking process section to be:

Collapsible – so that by default only a compact representation is shown, improving rendering performance and reducing UI clutter, similar to how Trae handles it.

Summarized with a short AI-generated title – the collapsed state should display a brief, dynamic summary (e.g., "Searching codebase for API definitions", "Analyzing error logs", "Structuring response") generated by the model itself, indicating what it is currently working on. This way users can monitor progress at a glance without needing to expand the full text.

These changes would drastically improve the Desktop experience, especially during long reasoning chains.

Additional context
Screenshots comparing the two interfaces:

OpenCode Desktop (current) – Thinking process fully expanded, causing lag and lacking summaries.

Trae (desired behavior) – Thinking steps are collapsed by default, each showing a concise summary of the current task.

中文

您的功能请求是否与某个问题相关？请描述。
OpenCode 桌面版在思考过程输出变长时会出现明显卡顿。渲染一个巨大且始终展开的文本块非常消耗资源，导致界面迟钝。此外，即便展开后，如果不逐行阅读，也很难迅速了解模型当前到底在做什么。

描述您想要的解决方案
我希望思考过程区域能够：

可折叠 – 默认仅显示紧凑的概括信息，从而提升渲染性能、减少界面混乱，类似 Trae 的做法。

附带AI生成的简短小标题 – 折叠状态下应显示一个动态的简短摘要（例如“正在搜索代码库中的 API 定义”、“正在分析错误日志”、“正在组织回复结构”），由模型自己生成，用于指示当前正在处理什么。这样用户无需展开就能一目了然地掌握进度。

这些改进将极大提升桌面版在长推理链过程中的使用体验。

附加上下文
两款界面的对比截图：

OpenCode（当前） – 思考过程完全展开，造成卡顿且缺少概括信息。

Trae（期望行为） – 思考步骤默认折叠，每步显示当前任务的简洁摘要。