Skip to content
BranchesTags

Repository files navigation

LLM Wiki Plugin

AI-powered personal knowledge operating system for Obsidian

Python 3.10+ Claude Code Obsidian License MIT

Knowledge Graph

Architecture

graph LR
    subgraph "Raw Sources"
        A1[articles/]
        A2[books/]
        A3[qa/]
    end

    subgraph "Ingest Engine"
        B1[wiki:ingest]
        B2[ingest-loop --engine=qwen]
    end

    subgraph "Wiki Layer"
        C1[entities/]
        C2[concepts/]
        C3[syntheses/]
        C4[qa-insights/]
    end

    subgraph "Index & Graph"
        D1[index.md]
        D2[BM25 Index]
        D3[graph.json]
        D4[graph.html]
        D5[statistics.html]
        D6[wiki/*.html]
    end

    A1 --> B1
    A2 --> B1
    A3 --> B1
    A1 --> B2
    A2 --> B2
    B1 --> C1
    B1 --> C2
    B1 --> C3
    B2 --> C1
    B2 --> C2
    C1 --> D1
    C2 --> D1
    C3 --> D1
    C4 --> D1
    C1 --> D2
    C2 --> D2
    C3 --> D2
    C1 --> D3
    C2 --> D3
    D3 --> D4
    D3 --> D5
    C1 --> D6
    C2 --> D6
Loading

Features

Feature Description
BM25 Search 中文分词 + BM25 全文检索，离线索引，毫秒级查询
Qwen Batch Ingest 基于通义千问 API 的批量 ingest，无需 Claude 上下文
Ralph-Loop 自动化循环 ingest，逐文件处理整个目录
D3.js Knowledge Graph 交互式力导向图，支持 GitHub Pages 部署
Statistics Dashboard Chart.js 统计看板：类型分布、置信度、标签频率、增长时间线
Static Wiki Viewer Wiki 页面 HTML 化，支持搜索、导航、元数据卡片
Hook System ingest 后自动触发 BM25 重建、图谱更新、lint 检查
Unified Search BM25 + maps topic expansion + graph traversal with RRF fusion
Format Conversion markitdown 批量转换 PDF/DOCX/PPTX/XLSX → markdown
Index Integrity snapshot_index.py 验证 index.md 完整性，--slim 模式压缩至 ~35 行统计表
Per-Topic Maps build_maps.py 从 topic-to-wiki.json 生成分主题 map（含概述），支持 --topic 按主题过滤 ingest 上下文
Custom Dictionary jieba 自定义词典从 wiki 元数据自动构建，提升领域术语分词精度
4-Layer Memory Working → Episodic → Semantic → Procedural 知识生命周期
QA Integration QA 对话数据批量导入，自动提取洞见并双向链接
CI/CD GitHub Actions 自动部署 graph + statistics + wiki 到 GitHub Pages

Quick Start

Prerequisites

  • Python 3.10+
  • Obsidian desktop app
  • Claude Code CLI (npm install -g @anthropic-ai/claude-code)
  • (Optional) DASHSCOPE_API_KEY for Qwen batch ingest

Setup

# 1. Clone the repo
git clone https://github.com/1998x-stack/llm-wiki-plugin.git
cd llm-wiki-plugin

# 2. Install Python dependencies
pip install -r requirements.txt

# 3. Open vault/ as an Obsidian vault
#    Obsidian → Open folder as vault → select vault/

# 4. Start Claude Code in the vault directory
cd vault
claude

# 5. Ingest your first source file
/wiki:ingest raw/articles/your-file.md

# 6. (Optional) Start auto-ingest loop
bash scripts/setup-ingest-loop.sh raw/articles/

# 7. (Optional) Build knowledge graph
python3 scripts/build_graph.py

Commands

All commands are invoked via Claude Code's wiki: prefix (e.g. /wiki:ingest).

Wiki command dependencies

Command Usage Description
ingest /wiki:ingest <path> 源材料 → wiki 页面，更新 index/log
query /wiki:query <question> BM25 + maps + graph 统一搜索并综合回答
check /wiki:check 只读健康诊断（不修改文件）
lint /wiki:lint 健康检查 + 自动修复（调用 check 后修复）
consolidate /wiki:consolidate 记忆晋升 + 置信度衰减
crystallize /wiki:crystallize 会话探索 → 结构化摘要
journal /wiki:journal <type> 写日记（daily/reflection/judgment）
review /wiki:review <scope> 分形回顾（weekly/monthly/quarterly）
qa-import /wiki:qa-import <path> QA 数据批量导入
ingest-loop /wiki:ingest-loop <dir> [--engine=qwen] Ralph-loop 批量 ingest（默认 Claude，--engine=qwen 用 Qwen API）
build /wiki:build 构建所有静态产出: graph + statistics + wiki HTML
reindex /wiki:reindex 验证 index 完整性 + 生成主题 maps + 精简 index
relink /wiki:relink 自动链接 wiki 中未链接的术语提及（最长匹配优先）
maintain /wiki:maintain 一键维护: relink → check → lint → build
convert-to-markdown /wiki:convert-to-markdown [dir] markitdown 批量转换 PDF/DOCX → markdown

Vault Structure

vault/
├── .claude/commands/wiki/   # 15 Claude Code commands
├── .obsidian/               # Obsidian settings
├── _schema/                 # 系统规则 (CLAUDE.md + 类型定义)
│   ├── CLAUDE.md            # Master schema
│   ├── entity-types.md      # 实体类型定义
│   ├── relationship-types.md # 关系类型定义
│   └── quality-rules.md     # 质量规则
├── _memory/                 # 4-layer memory system
│   ├── working/             # 当前会话观察
│   ├── episodic/            # 会话摘要
│   ├── semantic/            # 跨会话事实
│   └── procedural/          # 工作流与模式
├── raw/                     # 不可变源材料 (LLM read-only)
│   ├── articles/            # 文章、论文、分析（嵌套主题分类）
│   ├── books/               # 书籍章节
│   ├── assets/              # 代码资产、技能文件
│   ├── qa/                  # QA 对话数据
│   ├── re-map.json          # 文件夹重分类映射
│   └── raw-wiki-map.json    # raw → wiki 页面映射
├── wiki/                    # LLM 生成的知识页面
│   ├── entities/            # 人物、工具、项目
│   ├── concepts/            # 概念、理论、方法
│   ├── syntheses/           # 跨领域综合分析
│   └── qa-insights/         # QA 提取的洞见
├── journal/                 # 个人日记系统
│   ├── daily/               # 每日笔记
│   ├── reflections/         # 反思
│   ├── judgments/            # 判断记录
│   └── growth/              # 成长追踪
├── qa/                      # QA 数据存放
├── index/BM25/              # BM25 搜索索引
├── templates/               # 5 套模板
├── maps/                    # 主题分类索引（含概述，由 build_maps.py 生成）
├── scripts/                 # Python & Shell 自动化脚本
├── index.md                 # 内容目录（精简版：统计表 + 全局名称列表）
├── log.md                   # 操作日志
├── log.hook.md              # Hook 执行日志
├── graph.json               # 知识图谱数据
└── dashboard.md             # 仪表盘

Static Site

Three interconnected pages deployed to GitHub Pages:

Page Description
graph.html Interactive D3.js force-directed knowledge graph
statistics.html Chart.js dashboard (type distribution, confidence, tags, growth)
wiki/index.html Searchable wiki page viewer with metadata cards

Build locally:

cd vault
python3 scripts/build_graph.py --output ../static/graph.json --full
# Serves graph.json, graph-statistics.json, and wiki/*.html
python3 -m http.server 8080 --directory ../static

Documentation

Document Description
USERGUIDE.md 完整用户指南（中文）
docs/wiki.md Wiki 系统技术文档
docs/CHANGELOG.md 版本变更记录
llm-wiki-v1.md LLM Wiki v1 设计文档 (Karpathy)
llm-wiki-v2.md LLM Wiki v2 设计文档 (agentmemory)

Contributing

  1. Fork the repository
  2. Create your feature branch (git checkout -b feature/your-feature)
  3. Commit your changes (git commit -m 'Add your feature')
  4. Push to the branch (git push origin feature/your-feature)
  5. Open a Pull Request

For bug reports and feature requests, please use GitHub Issues.

License

This project is licensed under the MIT License. See LICENSE for details.

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages