Knowledge Base Setup
在 Mac Mini 上快速搭建本地知识库 + RAG 搜索系统。
快速开始
方法一:一键安装(推荐)
cd ~/.openclaw/workspace/skills/knowledge-base-setup/scripts
bash setup.sh <飞书用户ID>
setup.sh 会自动完成:
- 创建目录结构
- 安装 tesseract + Python 依赖
- 下载 nomic-embed-text 模型
- 部署分析脚本到 knowledge/.analysis/
- 更新 OpenClaw 配置(ollama provider + memorySearch)
- 重启网关
安装完成后手动注册定时任务(setup.sh 会打印具体命令)。
方法二:手动分步安装
按顺序执行以下步骤:
Step 1: 环境准备
brew install tesseract
pip3 install pytesseract pymupdf pdfplumber
# 安装 Ollama: https://ollama.com/download
Step 2: 下载 embedding 模型
ollama pull nomic-embed-text
Step 3: 创建目录结构
mkdir -p ~/.openclaw/workspace/knowledge/.analysis/summaries/archives
mkdir -p ~/.openclaw/workspace/knowledge/temp_docs
mkdir -p ~/.openclaw/workspace/knowledge/"Macro Financials"
touch ~/.openclaw/workspace/knowledge/文章目录.md
Step 4: 部署脚本(从 skill 目录复制)
cp ~/.openclaw/workspace/skills/knowledge-base-setup/scripts/run_analysis.py \
~/.openclaw/workspace/knowledge/.analysis/
cp ~/.openclaw/workspace/skills/knowledge-base-setup/scripts/generate_catalog.js \
~/.openclaw/workspace/knowledge/.analysis/
chmod +x ~/.openclaw/workspace/knowledge/.analysis/*.py
chmod +x ~/.openclaw/workspace/knowledge/.analysis/*.js
Step 5: 配置 OpenClaw
编辑 ~/.openclaw/openclaw.json,加入:
{
"models": {
"providers": {
"ollama": {
"baseUrl": "http://127.0.0.1:11434",
"api": "ollama",
"models": [
{"id": "nomic-embed-text", "name": "Nomic Embed Text"}
]
}
}
},
"agents": {
"defaults": {
"memorySearch": {
"provider": "ollama",
"model": "nomic-embed-text"
}
}
}
}
确保 tools 区块有:
"tools": {
"alsoAllow": ["exec", "process"]
}
然后重启:openclaw gateway restart
Step 6: 注册定时任务
# 22:00 分析任务
openclaw cron add \
--name "22:00分析新文档" \
--cron "0 22 * * *" \
--tz "Asia/Shanghai" \
--session isolated \
--timeout-seconds 300 \
--message "运行 run_analysis.py 和 generate_catalog.js" \
--announce --channel feishu --to "user:<飞书用户ID>"
# 08:00 发送任务
openclaw cron add \
--name "08:00发送文档摘要" \
--cron "0 8 * * *" \
--tz "Asia/Shanghai" \
--session isolated \
--timeout-seconds 120 \
--message "读取 summaries/ 目录发送摘要到飞书,完成后移动到 archives/" \
--announce --channel feishu --to "user:<飞书用户ID>"
迁移到新电脑
- 复制整个目录:
scp -r ~/.openclaw/workspace/knowledge user@new-mac:~/.openclaw/workspace/ - 在新电脑运行 setup.sh 或手动分步安装
- 重新注册定时任务(Job ID会变)
避坑指南
| 问题 | 原因 | 解决 |
|---|---|---|
| PDF提取乱码 | 自定义字体无ToUnicode | pymupdf+tesseract OCR |
| 定时任务超时 | 默认120秒太短 | --timeout-seconds 300 |
| 飞书无exec工具 | tools策略限制 | 添加 alsoAllow: [exec, process] |
| skill加载失败 | 导出名称错误 | CodeChunker → FileChunker |
| BGE-M3卡顿 | 16GB内存不足 | 继续用 nomic-embed-text |
| brew install ollama慢 | 网络问题 | 直接下载 dmg 安装 |
关键路径
- Skill目录:
~/.openclaw/workspace/skills/knowledge-base-setup/ - 知识库:
~/.openclaw/workspace/knowledge/ - 分析脚本:
~/.openclaw/workspace/knowledge/.analysis/ - 摘要输出:
~/.openclaw/workspace/knowledge/.analysis/summaries/