Building a Personal Knowledge Base with AI

Got Notes?

#

Most note-taking systems are designed for humans. You write notes, organize them, and hope you can find them later. But what if your AI assistant could be a full participant—reading your notes for context, writing new ones, and helping you build a knowledge base over time?

This tutorial shows how to set up a MEMORY system: a simple folder structure that serves as shared memory between you and your AI.

The Problem with Traditional Notes

#

When you use Apple Notes, Notion, or even Obsidian, your AI assistant can’t easily:

• Read your existing notes for context
• Save insights directly to your knowledge base
• Search your notes to answer questions
• Build on previous research

You end up copy-pasting between your notes app and your AI chat. That friction adds up.

The Solution: A Shared Filesystem

#

The fix is simple: use a folder of markdown files that both you and your AI can access. No special app required—just files on disk.

~/Documents/PAI/MEMORY/

Your AI reads and writes markdown. You can open the same files in any text editor, VS Code, or even Obsidian if you want a GUI.

Directory Structure

#

Here’s the structure I use:

MEMORY/
├── research/       # Research session outputs
├── ideas/          # Brainstorm captures
├── learnings/      # Things I've learned
├── decisions/      # Decisions and their rationale
├── sessions/       # Session summaries
├── analysis/       # Deep-dive documents
├── Work/           # Active task working memory
└── State/          # Operational state (JSON)

Each folder has a clear purpose. When you tell your AI to “save this research,” it knows exactly where it goes.

Setting It Up

#

Create the directory structure:

mkdir -p ~/Documents/PAI/MEMORY/{research,ideas,learnings,decisions,sessions,analysis,Work,State}

Add a README so you remember what goes where:

cat > ~/Documents/PAI/MEMORY/README.md << 'EOF'
# MEMORY

Personal knowledge base.

| Folder | Purpose |
|--------|---------|
| research/ | Research outputs |
| ideas/ | Brainstorms |
| learnings/ | Things learned |
| decisions/ | Decisions + rationale |
| sessions/ | Session summaries |
| analysis/ | Deep dives |
| Work/ | Active tasks |
EOF

That’s it. You now have a knowledge base your AI can use.

Using It with Your AI

#

Once the structure exists, you can talk to your AI naturally:

Saving Content

#

"Save this idea: what if we used webhooks instead of polling?"

→ Creates MEMORY/ideas/2026-01-21_webhooks-idea.md

"Record this learning: Claude works better with specific examples"

→ Creates MEMORY/learnings/2026-01-21_claude-examples.md

"Save this research to memory"

→ Creates MEMORY/research/2026-01-21_[topic].md

Retrieving Content

#

"What ideas have I saved recently?"

→ AI reads from MEMORY/ideas/

"What did I learn about API design?"

→ AI searches MEMORY/learnings/

"Show me my recent research"

→ AI lists files in MEMORY/research/

Building On Previous Work

#

"Continue the research from yesterday on authentication patterns"

→ AI reads previous file, adds new content

"What decisions have I made about the database?"

→ AI searches MEMORY/decisions/ for relevant entries

File Naming Convention

#

I use this format for consistency:

YYYY-MM-DD_short-description.md

Examples:

• 2026-01-21_webhooks-vs-polling.md
• 2026-01-15_auth-architecture-decision.md
• 2026-01-10_fabric-patterns-research.md

This keeps files sorted chronologically and makes them easy to scan.

Integration with Fabric Patterns

#

This pairs well with Fabric patterns. When you run extract_wisdom on a podcast or article, save the output directly to your memory:

"Use extract_wisdom on this transcript and save it to research"

Now that insight is permanently captured and searchable.

The Work/ Directory

#

For active tasks, I use a separate Work/ directory with per-task folders:

Work/
└── Website-Redesign_2026-01-20/
    ├── Work.md           # Goal, status, notes
    ├── Output/           # Deliverables
    └── Learning/         # What I learned

When a task is done, learnings get promoted to the main learnings/ folder, and the work directory gets archived.

Tips

#

1. Be consistent with commands — Pick phrases like “save this idea” and stick with them
2. Date everything — Makes it easy to find recent vs. old content
3. Don’t over-organize — Five folders is plenty. Add more only when needed
4. Review periodically — Skim your ideas and learnings monthly
5. Let the AI summarize — Ask “summarize my learnings from this month”

Why Not Just Use Obsidian/Notion?

#

You can! This approach works alongside those tools:

• Obsidian: Point a vault at MEMORY/ and get graph view, search, and plugins
• Notion: Harder to integrate, but you could sync key files
• Apple Notes: Use for quick mobile capture, then migrate important notes to MEMORY/

The point isn’t to replace your favorite app—it’s to have a shared space where your AI can participate fully.

What You Get

#

After a few weeks of use, you’ll have:

• A searchable archive of your research
• A log of decisions and their rationale
• Ideas captured before they evaporate
• Learnings that compound over time
• An AI that knows your context

Your AI becomes less of a stateless chatbot and more of a collaborator with memory.

The best note-taking system is one you actually use. By making your AI a full participant, you remove the friction of maintaining it alone.