knowledge-rag

概述

Knowledge RAG 是一个完整的本地优先检索增强生成系统，能够在不依赖任何外部服务器或 API 密钥的情况下，即时搜索各种格式的文档。它实现了混合搜索，结合 BM25 关键词搜索、语义向量和交叉编码器重排序，以提供高精度结果。该系统使用 Python 和 FastMCP 框架构建，提供 12 个 MCP 工具，用于完整的 CRUD 操作、搜索、评估和文档管理。

试试问 AI

装完之后，这里有 5 个你可以让 AI 做的事：

• searchquery: str

  Search through indexed documents using hybrid search with semantic and keyword matching

  试试:Search for documents about machine learning试试:Find files related to cybersecurity threats
• getdoc_id: str

  Retrieve a specific document by its ID

  试试:Get document with ID abc123试试:Retrieve the content of doc xyz789
• addpath: str

  Add a new document to the index

  试试:Add the report.pdf to the knowledge base试试:Index the meeting_notes.md file
• updatepath: str

  Update an existing document in the index

  试试:Update the project_requirements.docx试试:Refresh the technical_specs.txt
• removepath: str

  Remove a document from the index

  试试:Remove the outdated_policy.pdf试试:Delete the deprecated_code.py file
• reindex

  Rebuild the entire index from scratch

  试试:Rebuild the entire knowledge base试试:Refresh all indexed documents
• list

  List all documents in the index

  试试:Show all indexed documents试试:List all files in the knowledge base
• stats

  Get statistics about the index

  试试:Show index statistics试试:How many documents are indexed?
• urlurl: str

  Index content from a web URL

  试试:Index this article https://example.com/security试试:Add documentation from https://api.example.com/docs
• similardoc_id: str

  Find documents similar to a reference document

  试试:Find documents similar to the security audit report试试:Show me files like the project plan
• evaluate

  Evaluate retrieval performance with MRR@5 and Recall@5 metrics

  试试:Evaluate search performance试试:Check retrieval quality metrics

可对比工具

安装

pip install knowledge-rag

{
  "mcpServers": {
    "knowledge-rag": {
      "command": "python",
      "args": ["-m", "knowledge_rag.server"],
      "env": {}
    }
  }
}

FAQ

支持哪些文件格式？

Knowledge RAG 支持 20+ 种格式，包括 PDF、Markdown、Python 代码、Word 文档、Excel 文件、Jupyter 笔记本以及各种编程语言（C++、JavaScript、TypeScript 等）。

它是否真的完全本地运行，没有云依赖？

是的，所有处理都通过 ONNX 在本地运行。无需服务器、API 密钥，数据也不会离开您的机器。

Hacker News 讨论

开发者社区最近的相关讨论。

• Knowledge-RAG – Local RAG for Claude Code with hybrid search and cross-encoder▲ 2
  帖子 by lyonzin · 2026-03-19

knowledge-rag 对比