Building a PKM with Obsidian and Claude: A Practical Guide

What This Is: A real-world guide to building a personal knowledge management system that actually works with AI and turns Claude from a chatbot into a genuine thought partner.

The Real Problem

I'm constantly synthesizing research, connecting patterns across projects, and developing frameworks that integrate insights from multiple domains. The challenge lies in building deeper knowledge - understanding what information means, how ideas connect, and what implications emerge from those connections.

I needed a system where Claude could understand my context, challenge my thinking, and help me see what I was missing. That required three things: teaching Claude about my work, giving it access to my knowledge base, and creating a framework for genuine intellectual partnership.

The Architecture That Actually Works

The system has three layers that work together. Miss any one of them and you're back to chatting with a helpful but context-free AI assistant.

First, Obsidian as a document storage system. Not Notion, not Roam, not a folder full of Google Docs. Obsidian stores everything in plain markdown files on your computer. This matters because Claude can read these files directly through the filesystem extension. Your entire knowledge base becomes accessible to your AI thought partner. Critically, storing it in markdown and in file storage makes this portable and easy to use with other tools in the future.

Second, the Claude Refs folder. This is where the magic happens. I created a dedicated folder in my vault specifically for Claude's reference materials. It contains three critical files that transform how Claude works with me: my personal context, writing style guide, and Obsidian syntax reference. These files teach Claude who I am, how I think, what I'm working on, and how to interact with my knowledge system.

Third, system instructions that actually work. I built a set of core system instructions that Claude loads at the start of every conversation. These instructions point Claude to the reference files and establish the rules of engagement. They define Claude as a critical thought partner - someone who tests my thinking, challenges unsupported positions, and provides alternative perspectives. A peer who makes my thinking better.

What is Obsidian?

Obsidian is a powerful knowledge base that works on top of a local folder of plain text markdown files. Unlike cloud-based tools like Notion or Roam, your notes stay on your computer as simple text files you can read with any editor. The magic comes from how Obsidian helps you link ideas together - internal links between notes create a web of connected knowledge, visualized through an interactive graph view. Think of it as your personal Wikipedia, where every note can reference and build upon others, creating a genuine knowledge network rather than isolated documents.

Setting This Up: The Practical Steps

Step 1: Set Up Obsidian

1a. Download and create your vault. Download Obsidian from obsidian.md and create a new vault. This vault is just a folder on your computer where Obsidian will store your markdown files.

1b. Get comfortable with the basics. Spend a day or a few hours taking notes on real topics. Learn how internal linking works with double brackets [[like this]], how to search your notes, and how the graph view reveals connections between ideas. The tool should feel natural before you add AI to it.

1c. Create your Claude Refs folder. Make a dedicated folder in your vault for Claude's reference materials. Keep it separate from your regular notes - these are training materials for Claude, teaching it how to work with your knowledge base.

Step 2: Set Up Claude Desktop and Connect Your Vault

2a. Install Claude Desktop. Download Claude Desktop from claude.ai and install it. The desktop version provides access to extensions that the web version doesn't have.

2b. Enable the Filesystem extension. In Claude Desktop, go to Settings → Connectors → Browse Connectors → Desktop Connectors and enable the Filesystem extension. This allows Claude to read and write files in designated folders on your computer.

2c. Grant access to your Obsidian vault. When enabling the Filesystem extension, you'll be prompted to select which folders Claude can access. Add your Obsidian vault folder. Claude can now read your notes, search your vault, and help create new content.

2d. Create a dedicated Claude project. In Claude Desktop, create a new project (I called mine "PKM"). Projects in Claude allow you to set custom instructions and maintain context across conversations. This project becomes your dedicated workspace for knowledge management.

Step 3: Build Your Context Files

3a. Build your personal context file. This is the most important file you'll create. Document who you are professionally, what you're working on, how you like to work, and what you need from Claude. Include your stakeholders, your goals, your communication preferences, and most critically - how you want Claude to challenge your thinking.

My context file includes my professional background, my work patterns (early morning for deep thinking, late afternoon for creative work), and my core values around evidence-based decision making. When Claude knows this context, every interaction becomes more relevant and useful.

3b. Create an Obsidian syntax reference. Claude needs to know how to write in Obsidian-flavored markdown. Build a comprehensive reference covering wikilinks, callouts, frontmatter, tables, and Obsidian-specific features. Include best practices for AI-assisted note creation - when to add internal links, how to organize folders, what frontmatter to include, and how to maintain the atomic note principle.

When Claude creates or modifies notes in your vault, they'll integrate seamlessly with your existing knowledge system.

3c. Write your system instructions. This master file ties everything together. It points Claude to your reference files, establishes your core behavioral principles, and sets the rules for how you work together. My instructions emphasize evidence-based approaches, thought partnership, and specific patterns for responding to different types of requests.

Be explicit about what you want. Tell Claude directly to challenge your thinking, provide multiple perspectives, and cite sources. AI trained on helpfulness won't spontaneously push back on your ideas unless you tell it that's what you need.

3d. Load instructions into your Claude project. In your Claude project settings, add your system instructions file. Claude will read these instructions at the start of every conversation in this project, ensuring consistent behavior and access to your context files.

Step 4: Test and Refine

Start using the system for real work. Pay attention to where Claude helps and where it falls short. Update your context file as your work evolves. Refine your system instructions based on what actually works. This is a living system - it improves through use.

The Method: What Actually Works

The system works because it solves three problems that plague most AI interactions: lack of context, lack of continuity, and lack of critical thinking.

Context comes from the personal context file. When you start every conversation with Claude having already read about your current projects, your stakeholders, your goals, and your preferences, the quality of responses transforms immediately. You skip explaining your situation from scratch. Advice accounts for your specific constraints. Claude starts from where you actually are.

Continuity comes from vault access. When Claude can search your previous notes and build on existing thinking, you build knowledge over time rather than starting over with each conversation. Ideas connect. Patterns emerge. Your thinking evolves.

Critical thinking comes from explicit instruction. If you want Claude to challenge your assumptions, test your logic, and provide alternative perspectives, you have to say so explicitly in your system instructions. AI trained on helpfulness won't spontaneously push back on your ideas. Tell it that's what you need.

I specify that Claude should "engage as peer, not assistant" and "be intellectually honest when data contradicts positions." That transforms conversations from "here's what you asked for" to "here's what you asked for, but have you considered these three problems with your approach?"

The integration pattern that works. When working on something complex, I follow a consistent pattern. First, I search my vault for related notes and read them myself - Claude can help with this search, but I need to understand the baseline. Second, I have Claude analyze the situation with access to those notes, challenging my initial thinking. Third, we develop the framework or solution together, with Claude creating properly formatted notes that integrate into my vault. Fourth, I review and refine, ensuring the new thinking builds coherently on the old.

The goal is using AI to make your thinking better, amplifying your capability rather than replacing it.

What I Learned the Hard Way

The personal context file is essential. I started by chatting with Claude about my work, assuming conversation history would provide enough context. Without a formal context file, Claude forgets important details, makes incorrect assumptions, and provides generic advice. The context file seems like extra work until you realize it eliminates explaining yourself repeatedly.

Be specific about how you want to be challenged. My first version of the system instructions asked Claude to be "helpful and thoughtful." I revised to specify exactly what kind of challenge I wanted: "test my thinking rigorously" and "challenge positions when facts don't support them - provide alternate interpretations backed by evidence." The quality of our conversations improved immediately.

Maintain the system actively. Your personal context file needs regular updates as your projects change, goals evolve, and stakeholders shift. I review mine monthly and update whenever something significant changes. An outdated context file means Claude operates on bad information.

Use the system for real work. Use it for actual projects immediately - messy, complex, uncertain projects where you genuinely need help thinking through hard problems. That's where you discover what works.

Keep the structure simple. I started with elaborate folder hierarchies and complex tagging systems. What matters is that Claude can find relevant information when needed, and that new notes integrate coherently. Simple folder organization by topic, consistent naming conventions, and good internal linking accomplishes this.

The atomic note principle matters more than you think. I initially created long, comprehensive notes on topics. This makes them hard for Claude to work with - it has to wade through everything to find relevant sections. Breaking knowledge into focused, linkable atomic notes means Claude can quickly find and connect exactly the pieces needed for current thinking. Each note should be about one idea, properly linked to related concepts.

Making It Sustainable

The system works long-term because it reduces friction rather than adding it. That's the key test: does this make your thinking easier or harder?

For ongoing projects, I create dedicated notes that Claude updates as thinking evolves. The note becomes the record of our shared thinking process, with each conversation building on the last. This means less time explaining context and more time on actual problem-solving.

For recurring patterns, I document frameworks and approaches that become reusable. When similar situations arise, Claude can reference these established patterns and adapt them rather than rebuilding from scratch.

For learning and synthesis, I use Claude to help create permanent notes from observations and research. Claude helps synthesize scattered thoughts into coherent knowledge that integrates with my existing understanding.

The maintenance overhead is minimal if you're using the system for real work. Updating context happens naturally as your work evolves. Creating new notes is what you're doing anyway - Claude just helps do it better.

When to Use This vs. Doing It Yourself

The system amplifies independent thought rather than replacing it. I still think through many problems entirely on my own and write plenty of content without Claude's help.

Use Claude when you need perspective. When I'm stuck in my own assumptions or too close to a problem, working with Claude helps me see it differently. The act of explaining my thinking to an AI that can challenge it reveals gaps I wouldn't notice alone.

Use Claude for synthesis across domains. When I need to connect insights from government transformation work with private sector consulting, or link organizational change principles to AI implementation, Claude's ability to search my entire knowledge base and find unexpected connections is invaluable.

Use Claude for creating well-structured knowledge. When I have rough ideas that need to become coherent frameworks, or scattered notes that should become integrated understanding, Claude helps impose structure while maintaining my voice and thinking.

Maintain independent capability. If you find yourself unable to think without AI assistance, something's wrong. The system should amplify your thinking. I regularly write, plan, and solve problems entirely on my own to ensure the capability remains strong.

The Honest Assessment

This system transformed how I work with AI. It requires initial setup time and ongoing maintenance. It works best for people who already think in writing and value systematic approaches.

What it provides in return is genuine thought partnership. A thinking partner that knows your context, challenges your assumptions, helps you see blindspots, and makes your ideas better.

For complex knowledge work - strategy, transformation, leadership, consulting - that's invaluable. For simple tasks or casual questions, it's overkill.

The test is simple: Does working with Claude help you think things you wouldn't have thought alone? Does it reveal gaps in your reasoning you wouldn't have seen? Does it connect ideas in ways that surprise you?

If yes, the system is working. If no, either the setup needs refinement or this approach isn't right for your needs.

Starting Tomorrow

If you're going to build this, start with the personal context file. Spend an hour documenting who you are professionally, what you're working on, and how you want Claude to work with you. Be specific. Be honest. Include the hard stuff - your constraints, your uncertainties, your need for critical feedback.

Then create a simple Obsidian vault and add the filesystem connection. Don't worry about perfect organization. Just start taking notes on real work and let Claude help you think through actual problems.

The system will evolve as you use it. The structure will clarify through use. The value will become obvious when Claude references your previous thinking, challenges a weak assumption, or helps you see a connection you'd missed.

That's when you'll understand the difference between AI as a tool and AI as a thought partner.

Key ResourcesClaude Obsidian Syntax Reference - Complete guide to Obsidian formatting and best practicesPaul Kruchoski - Personal Context - Example of a comprehensive personal context fileClaude Core System Instructions - The system instructions that tie everything together

Questions or Ideas?

This is how I built my system, but your needs will be different. The principles matter more than the specifics - build what works for your thinking, not what worked for mine. The goal isn't to replicate my system, it's to create one that makes your own thinking better.