Building an AI Second Brain

Table of content

Your brain handles insight. Your AI handles everything else: past decisions, forgotten context, half-formed ideas. This is the second brain model.

What “second brain” means

Traditional second brains (Notion, Obsidian) require you to:

Manually file everything
Remember where you put it
Search and retrieve yourself

An AI second brain inverts this:

Traditional	AI-powered
You organize	AI organizes
You search	AI retrieves relevant context
You connect ideas	AI surfaces connections
You remember to check	AI proactively reminds

The AI becomes the interface to your own knowledge.

The cycle

[Capture] → [Organize] → [Retrieve]
    ↑                        ↓
    └────── [Compound] ──────┘

Phase	What happens
Capture	Raw thoughts go in, instantly
Organize	AI categorizes, tags, connects
Retrieve	Query in natural language
Compound	Retrieved context improves future captures

Each interaction teaches the system about you.

Required tools

Episodic memory

claude mcp add episodic-memory

This stores conversation history locally. Query past sessions:

claude "What did we decide about the API architecture last week?"

Quick capture

# Add to ~/.zshrc
cap() {
  claude "Quick capture: $*
Remember this. Respond only: 'Captured.'"
}

Now capture takes 2 seconds:

cap "meeting: agreed to ship v2 by March"
cap "idea: use webhooks instead of polling"

Note storage

Option	Command	Best for
Episodic memory	Built-in	Conversation history
Memory file	`~/.claude/memory.md`	Explicit facts
Obsidian	`claude mcp add obsidian`	Existing vault
Notion	`claude mcp add notion`	Structured data

Building a context library

Your second brain needs food. Feed it:

# After meetings
cap "meeting: stakeholders want mobile-first, deadline Q2"

# After decisions
cap "decision: chose Postgres over MongoDB (need joins)"

# After research
cap "learned: rate limit is 100/min, not 1000"

# After problems
cap "solved: timeout was DNS, not API"

Weekly, grow the semantic layer:

claude "Review this week's captures. Update my memory with:
- Recurring themes
- Important decisions
- Facts I should remember"

Searching your history

Episodic memory enables queries like:

claude "When did I last work on authentication?"
claude "What were my concerns about the migration?"
claude "Show me all decisions about the database"

The AI searches, synthesizes, and returns relevant context.

Search strategies

Query type	Example
Temporal	“What did I capture last Tuesday?”
Topical	“Everything about Project X”
Decision-based	“Why did we choose React?”
Problem-based	“How did I fix the CORS issue?”

Example: Research project with AI memory

Starting a new project:

# Day 1: Initial research
cap "evaluating: Stripe vs Square for payments"
cap "stripe: 2.9% + $0.30, better docs"
cap "square: 2.6% + $0.10, better POS"

# Day 3: More data
cap "talked to Sarah: they had Stripe integration issues"
cap "found: Stripe has webhook reliability problems"

# Day 7: Decision time
claude "I need to choose a payment processor.
What have I learned? What are the tradeoffs?"

The AI synthesizes everything:

Your captured notes
Previous conversations
Relevant decisions

You decide with full context, not foggy memory.

From chatbot to operating system

A chatbot answers questions. A second brain:

Chatbot	Second brain
Stateless	Remembers everything
Generic	Knows your context
Reactive	Proactively surfaces info
Tool	Extension of your mind

The shift: stop treating AI as someone to ask, start treating it as somewhere to think.

Next: What is a Personal OS?