HackerNews中文版

最近一直在研究 RAG 系统，反复遇到同一个问题：它们能检索到相关的文本块，但会丢失文本块之间的关系。在处理实际系统（文档、API、基础设施）时，这个问题很快就会显现出来，因为理解事物之间的联系比仅仅找到相似的文本更重要。所以我构建了 Mindex： https://usemindex.dev/ 它结合了语义搜索和知识图谱层，因此它不仅仅返回孤立的文本块，还能连接文档并展示它们之间的关系。它通过 CLI 和 MCP 工作，因此您可以直接将其插入到 Claude Code、Cursor 或您自己的 Agent 等工具中。我还添加了一个可视化对比，展示了朴素的 RAG 与基于图的检索，这使得差异更加清晰。非常希望得到反馈——特别是来自使用 RAG 构建或从事开发者工具开发的人。

查看原文

I’ve been working a lot with RAG systems recently, and kept running into the same issue: they retrieve relevant chunks, but lose the relationships between them.This becomes a problem pretty quickly when dealing with real systems (docs, APIs, infra), where understanding how things connect matters more than just finding similar text.So I built Mindex: https://usemindex.dev/It combines semantic search with a knowledge graph layer, so instead of returning isolated chunks, it can connect documents and surface how they relate.It works via CLI and MCP, so you can plug it directly into tools like Claude Code, Cursor, or your own agents.I also added a visual comparison showing naive RAG vs graph-based retrieval, which makes the difference clearer.Would love feedback — especially from people building with RAG or working on developer tooling.

别再用幼稚的 RAG 了——给 AI 上下文添加关系