# I Stopped Managing Notes and Started Using a Second Brain (Claude + Obsidian)
**作者**: Asteri
**日期**: 2026-07-22T11:24:47.000Z
**来源**: [https://x.com/Asteri_eth/status/2079890396688110001](https://x.com/Asteri_eth/status/2079890396688110001)
---
Most Obsidian vaults eventually become graveyards.
You keep adding notes, PDFs, articles, and ideas, but after a few months you can barely find anything useful. The graph looks impressive, yet the connections are weak, and every important question feels like starting from zero again.
Notion locks your data behind a subscription. Plain Obsidian gives you freedom, but all the organizing work stays on you.
There’s now a much better approach.
By combining Claude with Obsidian using the LLM Wiki pattern (popularized by Andrej Karpathy), you can build a knowledge base that reads sources, creates meaningful links, maintains itself, and actually answers your questions all while keeping everything in plain Markdown files you fully own.
## The Core Idea: LLM Wiki Pattern
Instead of treating every question as a new search across raw files, the system gradually builds a clean, interconnected wiki of Markdown pages.
- Raw sources stay untouched
- The AI creates and updates dedicated pages for concepts, entities, summaries, and comparisons
- New information gets properly integrated instead of just added
- Contradictions get flagged automatically
The result? Knowledge compounds over time instead of staying scattered.
## claude-obsidian: The Practical Implementation
The most popular open-source tool built on this pattern is claude-obsidian (GitHub repo with nearly 10k stars).
It turns Obsidian into a self-organizing second brain. You simply drop in sources, and Claude handles the rest: creating linked pages, updating the index, maintaining context, and answering questions with proper citations.
Quick Start (Takes 5-10 Minutes)
Recommended way:
```
git clone https://github.com/AgriciDaniel/claude-obsidian
cd claude-obsidian
bash bin/setup-vault.sh
```
1. Open the folder as a vault in Obsidian.
2. Open Claude Code in the same folder.
3. Type /wiki and answer the one question it asks.
You can also install it as a Claude Code plugin or add it to an existing vault.
## Daily Workflow
Once set up, the process is very simple:
- Drop a PDF, article, or transcript into the .raw/ folder
- Type: ingest filename.pdf
- Claude automatically creates 8–15 linked wiki pages
Later, just ask natural questions:
what do you know about [topic]?
It reads your wiki, synthesizes an answer, and cites the exact pages.
Other useful commands:
- /save - Save the current conversation as a structured note
- /autoresearch [topic] - Let Claude research the web and file the results
- lint the wiki - Weekly health check (finds orphans, dead links, and contradictions)
Pro Tips for Best Results
- Choose a methodology mode early (PARA, LYT, or Zettelkasten) using the setup script
- Install kepano/obsidian-skills so Claude writes proper Obsidian syntax
- Use the Obsidian Web Clipper to save articles directly into .raw/
- Point all your Claude projects to the same central wiki vault
- Run lint the wiki once a week
- Use /think when making important decisions it runs structured reasoning over your knowledge base
## What Changes After Using It for 30-60 Days
- You stop searching Google for things you’ve already read
- Questions get answered in seconds with citations from your own notes
- The graph view becomes genuinely useful
- Your knowledge base gets smarter the more you use it
- You spend much less time organizing and more time thinking
## Honest Limitations
This system works best if you already use Obsidian and Claude Code regularly. It’s not fully automatic you still need to feed it good sources. The quality of your second brain depends on what you put into it.
It’s also not magic. The AI handles the tedious maintenance work, but strategic direction still comes from you.
Final Thoughts
The biggest shift isn’t the AI itself it’s removing the constant friction of organizing and connecting notes.
When your system starts maintaining itself, everything changes. Research becomes faster. Writing becomes easier. Thinking becomes clearer.
You stop managing notes and start using a living knowledge base.
Ready to try it?
- GitHub repo: github.com/AgriciDaniel/claude-obsidian
- Karpathy’s original LLM Wiki pattern: gist.github.com/karpathy/442a6bf555914893e9891c11519de94f
Bookmark this article so you can come back to it when you’re ready to set everything up.
The sooner you start feeding your vault, the faster your second brain will start working for you.
> Follow me @Asteri_eth for more posts on Claude, AI tools, Prediction markets
## 相关链接
- [Asteri](https://x.com/Asteri_eth)
- [@Asteri_eth](https://x.com/Asteri_eth)
- [5.6K](https://x.com/Asteri_eth/status/2079890396688110001/analytics)
- [github.com/AgriciDaniel/claude-obsidian](https://github.com/AgriciDaniel/claude-obsidian)
- [gist.github.com/karpathy/442a6bf555914893e9891c11519de94f](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f)
- [@Asteri_eth](https://x.com/@Asteri_eth)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [7:24 PM · Jul 22, 2026](https://x.com/Asteri_eth/status/2079890396688110001)
- [5,627 Views](https://x.com/Asteri_eth/status/2079890396688110001/analytics)
- [View quotes](https://x.com/Asteri_eth/status/2079890396688110001/quotes)
---
*导出时间: 2026/7/23 11:12:48*
---
## 中文翻译
# 我停止了管理笔记，转而使用第二大脑 (Claude + Obsidian)
**作者**: Asteri
**日期**: 2026-07-22T11:24:47.000Z
**来源**: [https://x.com/Asteri_eth/status/2079890396688110001](https://x.com/Asteri_eth/status/2079890396688110001)
---
大多数 Obsidian 仓库最终都会变成坟墓。
你不断地添加笔记、PDF、文章和想法，但几个月后，你几乎找不到任何有用的东西。关系图看起来很令人印象深刻，但连接却很微弱，每一个重要的问题感觉都要从零开始。
Notion 将你的数据锁定在订阅服务后面。普通的 Obsidian 给了你自由，但所有的整理工作仍然落在你身上。
现在有一个好得多的方法。
通过使用 LLM Wiki 模式（由 Andrej Karpathy 普及）将 Claude 与 Obsidian 结合，你可以构建一个知识库，它阅读源文件，创建有意义的链接，自我维护，并真正回答你的问题，同时将所有内容保存在你完全拥有的纯 Markdown 文件中。
## 核心理念：LLM Wiki 模式
系统不再将每个问题视为对原始文件的新搜索，而是逐步构建一个整洁、相互关联的 Markdown 页面 Wiki。
- 原始源文件保持不变
- AI 为概念、实体、摘要和比较创建并更新专用页面
- 新信息被正确整合，而不仅仅是添加
- 矛盾会被自动标记
结果是什么？知识随着时间推移而复利增长，而不是保持分散。
## claude-obsidian：实际实现
基于此模式构建的最受欢迎的开源工具是 claude-obsidian（GitHub 仓库拥有近 1 万颗星）。
它将 Obsidian 变成了一个自我组织的第二大脑。你只需放入源文件，其余的由 Claude 处理：创建链接页面、更新索引、维护上下文，并通过适当的引用回答问题。
快速开始（需要 5-10 分钟）
推荐方式：
```
git clone https://github.com/AgriciDaniel/claude-obsidian
cd claude-obsidian
bash bin/setup-vault.sh
```
1. 在 Obsidian 中将该文件夹作为仓库打开。
2. 在同一文件夹中打开 Claude Code。
3. 输入 /wiki 并回答它询问的一个问题。
你也可以将其作为 Claude Code 插件安装，或添加到现有的仓库中。
## 每日工作流
设置完成后，过程非常简单：
- 将 PDF、文章或转录文件放入 .raw/ 文件夹
- 输入：ingest filename.pdf
- Claude 会自动创建 8–15 个链接的 wiki 页面
稍后，只需提出自然语言问题：
what do you know about [topic]?
它会阅读你的 wiki，综合答案，并引用确切的页面。
其他有用的命令：
- /save - 将当前对话保存为结构化笔记
- /autoresearch [topic] - 让 Claude 研究网络并将结果归档
- lint the wiki - 每周健康检查（查找孤立页面、死链接和矛盾）
获得最佳效果的专业技巧
- 尽早使用设置脚本选择一种方法论模式（PARA、LYT 或 Zettelkasten）
- 安装 kepano/obsidian-skills，以便 Claude 编写正确的 Obsidian 语法
- 使用 Obsidian Web Clipper 将文章直接保存到 .raw/
- 将你所有的 Claude 项目指向同一个中央 wiki 仓库
- 每周运行一次 lint the wiki
- 在做出重要决定时使用 /think —— 它会基于你的知识库运行结构化推理
## 使用 30-60 天后的变化
- 你停止在 Google 上搜索你已经读过的内容
- 几秒钟内就能根据你自己的笔记引用回答问题
- 关系图视图变得真正有用
- 你的知识库用得越多就越聪明
- 你花在整理上的时间大大减少，而花在思考上的时间增加
## 诚实的局限性
如果你已经经常使用 Obsidian 和 Claude Code，这个系统效果最好。它不是全自动的 —— 你仍然需要为它提供好的源文件。你的第二大脑的质量取决于你投入的内容。
它也不是魔法。AI 处理繁琐的维护工作，但战略方向仍然来自你。
最后的思考
最大的转变不是 AI 本身 —— 而是消除了整理和连接笔记的持续摩擦。
当你的系统开始自我维护时，一切都会改变。研究变得更快。写作变得更容易。思维变得更清晰。
你停止管理笔记，开始使用一个活生生的知识库。
准备试试了吗？
- GitHub 仓库：github.com/AgriciDaniel/claude-obsidian
- Karpathy 的原始 LLM Wiki 模式：gist.github.com/karpathy/442a6bf555914893e9891c11519de94f
收藏这篇文章，以便你在准备好设置所有内容时回来查看。
你越早开始向你的仓库投喂内容，你的第二大脑就能越快开始为你工作。
> 关注我 @Asteri_eth 获取更多关于 Claude、AI 工具、预测市场的帖子
## 相关链接
- [Asteri](https://x.com/Asteri_eth)
- [@Asteri_eth](https://x.com/Asteri_eth)
- [5.6K](https://x.com/Asteri_eth/status/2079890396688110001/analytics)
- [github.com/AgriciDaniel/claude-obsidian](https://github.com/AgriciDaniel/claude-obsidian)
- [gist.github.com/karpathy/442a6bf555914893e9891c11519de94f](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f)
- [@Asteri_eth](https://x.com/@Asteri_eth)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [7:24 PM · Jul 22, 2026](https://x.com/Asteri_eth/status/2079890396688110001)
- [5,627 Views](https://x.com/Asteri_eth/status/2079890396688110001/analytics)
- [View quotes](https://x.com/Asteri_eth/status/2079890396688110001/quotes)
---
*导出时间: 2026/7/23 11:12:48*