Show HN: Llmcc – 用于代码理解的多深度架构图
1 分•作者: allenanswerzq•6 个月前
嗨，HN 社区，祝大家新年快乐！
我一直在开发 llmcc，这是一个为代码库生成多“缩放级别”架构视图的工具。 想法是：编码代理（和人类）需要从不同的距离理解系统——有时你需要高层次的 crate/模块边界，有时需要低层次的符号关系。
它使用 PageRank 来呈现最重要的节点，因此你可以在几秒钟内获得 60 万行代码库的聚焦视图。
目前仅支持 Rust，但其架构设计支持多种语言。
我开发这个工具是因为我经常看到编码代理在大型代码库中迷失方向——它们要么过度聚焦（丢失上下文），要么试图处理太多内容（达到 token 限制）。拥有一个树状视图，可以在其中上下移动，似乎是正确的原始方法。
仓库：<a href="https://github.com/allenanswerzq/llmcc" rel="nofollow">https://github.com/allenanswerzq/llmcc</a>，可以查看 SVG 或基准测试。
期待反馈：
- 你们觉得它有用吗，会继续投入时间开发吗？
- 你们希望接下来支持哪些语言？
- 关于这个项目，对于编码代理有什么想法吗？
乐于回答问题！
查看原文
Hi, HN community, happy new year to all!<p>I've been working on llmcc, a tool that generates architecture view at multiple "zoom levels" for codebases. The idea: coding agents (and humans) need to understand systems from different distances—sometimes you need the high-level crate/module boundaries, sometimes the low-level symbol relationships.<p>It uses PageRank to surface the most "important" nodes, so you can get a focused view of a 600K LoC codebase in seconds.<p>Currently Rust-only, but the architecture is designed to support multiple languages.<p>Built this because I kept seeing coding agents get lost in large codebases—they'd either get too zoomed in (missing context) or try to process too much (hitting token limits). Having a tree-like view where you can walk up/down seemed like the right primitive.<p>Repo: <a href="https://github.com/allenanswerzq/llmcc" rel="nofollow">https://github.com/allenanswerzq/llmcc</a>, checked out for svgs or benchmark.<p>Looking for feedbacks:
- Do you guys feel could be useful and continue spend time to for it?
- What languages would you want next?
- Any kinds of ideas regarding this project for coding agents?<p>Happy to answer questions!