Claude Code + Obsidian: Build a Local AI Knowledge Base – Step-by-Step Guide

These days, many people store their materials like this:

• Collect articles in WeChat public account folders.
• Dump PDFs into the Downloads folder.
• Save web links in browser bookmarks.
• Write course notes in different software.
• Scatter AI chat logs across various platforms.
• Keep notes in Obsidian, but they get messier over time.

This article explains a beginner-friendly solution: Use Obsidian to store materials, Claude Code to organize them, and Markdown to build your own AI knowledge base.

1️⃣ What Exactly Is This Combination?

Think of it this way:

Obsidian – stores materials

It acts like a local knowledge base shell.

Your notes, articles, summaries, and indexes can all be saved as Markdown files on your computer.

Advantages:

• Local and controllable
• Clear file structure
• Easy to migrate
• Suitable for long-term maintenance
• Easy for AI to read

Claude Code – organizes materials

It works more like an AI organizer.

After you put materials into folders, you can ask it to:

• Organize directories
• Summarize content
• Generate indexes
• Extract tags
• Find duplicates
• Create learning paths
• Answer questions based on your materials

Markdown – ensures long-term persistence

Markdown is simple – plain text format.

It isn’t locked into any specific software.

You can view it in Obsidian today, or open it with VS Code, Typora, GitHub, or other tools tomorrow.

Obsidian stores, Claude Code organizes, Markdown ensures long-term portability.

2️⃣ Who Is This For?

This method is especially suitable for:

✅ Content creators

Put your topic ideas, viral breakdowns, tool references, and article materials here. Later when you write, just ask:

“Which of the AI tools I saved before are suitable for a Xiaohongshu topic?”

✅ Students / Graduate students

Store papers, course notes, textbook summaries, and reading notes. Later when reviewing, ask: “Based on my knowledge base, create a review outline for this course.”

✅ Independent developers

Store product ideas, technical docs, competitor analysis, and user feedback. Later when building products, ask: “Based on my user feedback records, which features should be prioritized?”

✅ Working professionals

Store meeting minutes, project materials, SOPs, report templates, and client info. Later when writing weekly reports, proposals, or retrospectives, you don’t start from scratch.

✅ Obsidian users

If you already use Obsidian but your notes are scattered, this method is perfect for you.

If you regularly need to organize, reuse, and create content based on your materials, an AI knowledge base is valuable.

3️⃣ Step 1: Create an Obsidian Knowledge Base Folder

Don’t dump everything in at once.

Start by creating a new vault, for example:

AI-Knowledge-Base

Then create 6 folders inside:

00_Inbox
01_Raw
02_Notes
03_Wiki
04_Output
99_Archive

Each folder has a simple purpose.

00_Inbox – Temporary collection zone

Dump anything you come across here first.

For example:

• Web text
• Screenshot text
• WeChat public account articles
• Random ideas
• AI chat logs

No need to organize here – just collect.

01_Raw – Raw materials zone

Store unprocessed materials here.

For example:

• PDF summaries
• Original articles
• Course handouts
• Product documentation
• Research materials

02_Notes – Personal notes zone

Store your own understanding here.

For example:

• Reading reflections
• Practical logs
• Pitfall summaries
• Case breakdowns

03_Wiki – Structured knowledge zone

This is the most important folder.

Store organized knowledge pages here.

For example:

• What is an AI Agent
• How to use Claude Code
• Obsidian knowledge base workflow
• X platform growth methodology
• AI tool classification index

04_Output – Output zone

Store content produced from your knowledge base.

For example:

• Xiaohongshu posts
• WeChat long articles
• X long threads
• Course outlines
• Project proposals

99_Archive – Archive zone

Store outdated, rarely used, or processed materials.

Your knowledge base doesn’t mix everything together – it first separates: raw materials, notes, knowledge, and outputs.

4️⃣ Step 2: Write a CLAUDE.md to Tell Claude Code How to Work

Create a new file in your vault root:

CLAUDE.md

Its purpose is to tell Claude Code:

• What this knowledge base is for.
• How folders are organized.
• What rules to follow when organizing.
• What format to use for output.
• What actions to avoid.

You can copy this template directly:

# My AI Knowledge Base Working Rules

You are my local knowledge base organizer.

This knowledge base is managed with Obsidian + Markdown.

## Folder Descriptions

- **00_Inbox**: Temporary collection of unorganized materials
- **01_Raw**: Raw materials
- **02_Notes**: My personal notes and understanding
- **03_Wiki**: Structured knowledge pages
- **04_Output**: Articles, plans, scripts generated from the knowledge base
- **99_Archive**: Outdated or archived materials

## Working Principles

1. Do not delete files arbitrarily.
2. Before modifying a file, explain what you plan to change.
3. Do not directly overwrite raw materials.
4. When organizing, preserve the source and context.
5. Mark uncertain content with "(uncertain)".
6. Do not present speculation as fact.
7. Output should be well-structured and suitable for reading in Obsidian.
8. New knowledge pages should use Markdown headings, lists, tags, and double brackets whenever possible.

## Common Output Format

When organizing materials, always output:
- Core summary
- Key points
- Actionable steps
- Related concepts
- Extended questions
- Suitable output formats (e.g., articles, posts)

CLAUDE.md is like your knowledge base manual. The clearer you write it, the less likely Claude Code will mess up your organization.

5️⃣ Step 3: Start with a Few Materials – Don’t Import 1000 at Once

The most common beginner mistake:

As soon as you finish setting up the system, you try to dump everything in.

Result: you get buried by materials before your knowledge base even runs smoothly.

For the first version, only add 10 materials.

Place them in 00_Inbox.

Then ask Claude Code to organize.

Prompt example:

Please read the materials in 00_Inbox.

Do not move or delete files yet.

Please help me:
1. List the topic of each material.
2. Determine which folder it belongs to.
3. Extract the key ideas of each material.
4. Suggest which Wiki pages could be generated.
5. Output an organization plan. Wait for my confirmation before executing.

After it gives you a plan, confirm by saying:

I confirm to execute this organization plan.
Please organize the materials into structured knowledge pages in 03_Wiki,
while keeping the original materials in 01_Raw.

6️⃣ Step 4: Have Claude Code Generate Wiki Pages

Wiki pages are not ordinary notes.

They should be structured, reusable, and easy to query later.

For example, if you collected a bunch of Claude Code materials, you can ask it to generate:

03_Wiki/Claude_Code_Getting_Started.md

Prompt:

Based on the materials about Claude Code in 01_Raw and 02_Notes,
generate a structured Wiki page.

File name:
03_Wiki/Claude_Code_Getting_Started.md

Include:
1. What is Claude Code
2. Who is it for
3. What can it do
4. What can it not do
5. First steps for beginners
6. Common prompt templates
7. Important notes
8. Related links and extension topics

Requirements:
- Do not fabricate facts.
- If the materials don't provide clear evidence, mark as "(uncertain)".

This way, you don’t have to re-read a bunch of materials every time.

Just refer to this Wiki page directly.

7️⃣ Step 5: Create a Home Page Index

If your knowledge base keeps growing without an index, it will get messy again.

So it’s recommended to create a home page:

00_HOME.md

Content example:

# My AI Knowledge Base Home

## Core Topics

- [[Claude Code]]
- [[Obsidian]]
- [[AI Agent]]
- [[X Growth]]
- [[Content Creation]]
- [[Independent Development]]
- [[AI Tools]]

## Recent Organization

- 

## Pending Materials

- 

## Output Ideas

- 

## Common Prompts

- 

Then ask Claude Code to update it periodically:

Please scan 03_Wiki and 04_Output,
and update 00_HOME.md for me.

Requirements:
1. Categorize by topic.
2. List recently added knowledge pages.
3. Mark materials that haven't been fully organized.
4. Extract 10 topics that can be turned into articles.

8️⃣ Step 6: Use the Knowledge Base to Answer Questions

Later, you can ask directly:

Based on the content in 03_Wiki,
answer: I want to write an article about Claude Code + Obsidian knowledge base.
What angle would be the best to start from?

Or:

Based on my knowledge base,
organize an outline for a Xiaohongshu image-text post.

Topic: How to build a local AI knowledge base
Requirements:
1. Beginner-friendly
2. Conversational tone
3. No exaggeration
4. Step-by-step instructions
5. Pitfall warnings

You can also ask:

Please find all materials in my knowledge base related to "AI content workflow",
and summarize them into a 1500-word article outline.

The real value of a knowledge base is that your old materials continuously participate in new outputs.

9️⃣ Step 7: Maintain It Weekly – Don’t Set It and Forget It

It’s recommended to do maintenance once a week.

Prompt:

Please help maintain this knowledge base.

Tasks:
1. Scan 00_Inbox for unprocessed materials.
2. Find duplicate or similar topics.
3. Mark outdated content.
4. Update 00_HOME.md.
5. Summarize new knowledge points from this week.
6. Suggest 10 output topics for next week.

Note:
- Do not delete files.
- For moves and archiving, first give suggestions and wait for my confirmation.

This step is important.

A knowledge base that isn’t maintained will eventually turn into just another collection.

🔟 About “Offline” and “Privacy” – Must Be Clear

Obsidian files can be saved locally, but Claude Code usually needs to connect to a model service – it is not the same as a fully offline local model by default.

So, if you’re dealing with:

• Internal company documents
• Client privacy
• Contracts
• Financial data
• Undisclosed projects
• Personal sensitive information

Do not casually hand them over to any cloud AI tool.

The safer approach:

1. Anonymize first.
2. Only store publicly shareable materials.
3. Keep important content backed up locally.
4. Keep sensitive files out of the AI workflow.
5. For offline needs, use a true local model solution.

1️⃣1️⃣ Minimum Viable Version for Beginners

Step 1

Install Obsidian. Create a vault:

AI-Knowledge-Base

Step 2

Create 4 folders:

00_Inbox
01_Raw
03_Wiki
04_Output

Step 3

Create:

CLAUDE.md

Write clear organization rules.

Step 4

Add 10 materials to Inbox.

Step 5

Ask Claude Code to first produce an organization plan – don’t let it modify directly.

Step 6

After confirmation, generate Wiki pages.

Step 7

Use a Wiki page to generate an article or study outline.

Once you complete these 7 steps, you are no longer just collecting materials – you are building your own AI knowledge system.

Final Summary

Building a knowledge base with Claude Code + Obsidian is not just about “adding another tool.”

It is a workflow:

• Obsidian stores materials
• Markdown ensures portability
• CLAUDE.md governs organization rules
• Claude Code organizes and generates structure
• Wiki pages solidify knowledge
• Home index manages the whole picture
• Weekly maintenance drives continuous iteration

The core truth is: A collection is a warehouse; a knowledge base is a system.