[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f2coyt8ks3kvus":3},{"code":4,"message":5,"data":6,"timestamp":69},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":40,"summary_md":41,"topic":42,"github_topics":45,"license":65,"homepage":66,"updated_at":67,"history":68},"Graphify-Labs\u002Fgraphify",4,"Turn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A \u002Fgraphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.","Python",134308,12829,6875,204,[16,22,27,31,35],{"id":17,"title":18,"points":19,"comments":20,"posted_at":21},49217306,"Graphify – Less token use for Claude Code",3,0,"2026-08-07",{"id":23,"title":24,"points":25,"comments":20,"posted_at":26},49195383,"Graphify – Turn any codebase into a queryable knowledge graph",2,"2026-08-06",{"id":28,"title":29,"points":25,"comments":20,"posted_at":30},49094378,"Graphify: Codebase Knowledge Graph","2026-07-29",{"id":32,"title":33,"points":25,"comments":20,"posted_at":34},48903692,"Show HN: Turn any folder of code, SQL schemas into a queryable knowledge graph","2026-07-14",{"id":36,"title":37,"points":38,"comments":20,"posted_at":39},48794162,"Graphify: Turn any codebase into a queryable knowledge graph",1,"2026-07-05","把代码库解析成可查询的知识图谱，让编程 Agent 沿着图找答案而不是满仓库 grep。","## 解决什么问题\n\n编程 Agent 在一知半解的情况下靠 grep 和逐个读文件来定位代码，大仓库里又慢又烧 token，还经常找不到正确的调用链。Graphify 把整个项目（代码、文档、PDF，甚至图片和视频）解析成一张图，Agent 用查询代替反复读文件。生成的 graph.json 会持久化，同一个仓库再提问时不用重扫一遍。\n\n## 有什么不同\n\n面对这个问题，主流做法是向量索引：切块、做 embedding、召回 top-k。Graphify 明确不走这条路，代码部分用 tree-sitter 做 AST 解析，结果是确定性的，不调用 LLM，也不建向量库，产出是一张可以真正遍历的图。由此带来两个实质差异：每条边都带 `EXTRACTED` 或 `INFERRED` 标签，能区分哪些关系是从源码直接读出来的、哪些是解析推导的；查询方式偏结构化（`explain`、`path A B`、按问题圈定子图），更适合调用链和依赖关系这类 embedding 不擅长的场景。文档、PDF、图片和 ADR 走的是另一条语义通道，那一步需要模型参与。\n\n## 适合谁\n\n在中等以上规模代码库上用 Claude Code、Cursor、Codex、Gemini CLI 等助手的开发者，尤其是跨文件耦合重、设计文档多的项目。对不愿意把源码送到第三方索引服务的团队，代码部分的本地解析是加分项。如果你要的是对散文式内容做语义搜索，或者仓库小到 grep 本来就够快，它帮不上什么忙；想要纯 RAG 技术栈的人应该去看向量库方案。\n\n## 上手成本\n\n需要 Python 3.10+，以及 uv 或 pipx。安装是 `uv tool install graphifyy`，再执行 `graphify install` 把 skill 注册到你的助手里。代码解析完全在本地，但文档和多媒体那一步要调用助手自带模型或你配置的 API key，所以「不出本机」的承诺只对代码成立。输出格式带有 schema 版本号，说明作者有兼容性意识，但也意味着跨版本集成可能出问题。许可证 Apache-2.0。README 中 Graphify Cloud 的商业推广占比很高，需要注意开源 CLI 和付费产品之间是否存在功能切分。\n\n## 点评\n\n项目 2026 年 4 月才创建，却已有 134k star、单周 +6,875，更像一次爆发式增长而不是沉淀下来的采用曲线。用确定性 AST 建图、给每条边标注来源，这个思路确实和基于 embedding 的检索不同，代码零 LLM 调用在成本和隐私上都是实打实的优势。风险有两方面：203 天的代码库配这个星数，成熟度需要自己验证；商业化路线也可能把能力往云端收。建议先拿一个你非常熟悉的大仓库试一次，用你自己的理解去检验图的质量，再决定要不要依赖它。",{"slug":43,"name":44},"developer-tools","开发工具",[46,47,48,49,50,51,52,53,43,54,55,56,57,58,59,60,61,62,63,64],"ai-agents","antigravity","ast","claude-code","code-analysis","code-search","codex","cursor","gemini","graphify","graphrag","knowledge-graph","leiden","llm","mcp","openclaw","rag","skills","tree-sitter","Apache-2.0","https:\u002F\u002Fwww.graphify.com","2026-10-24T15:35:37Z",[],1792868437]