RefObsidian + Claude Code: Building a 24/7 Personal Operating System

Obsidian + Claude Code: Building a 24/7 Personal Operating System

Obsidian + Claude Code: Building a 24/7 Personal Operating System

"Markdown files are the oxygen of LLMs. The more you write, the more context your agents have."

Greg Isenberg Obsidian Tweet

Summary

Greg Isenberg outlines a workflow for using Obsidian (markdown-based note-taking) + Claude Code to build a "24/7 personal operating system" that turns your notes into leverage for AI agents. The key insight: by writing everything in interconnected markdown files, you create a knowledge vault that agents can read, understand, and act on — eliminating the need to re-explain context every session.

"Vin uses AI like a thinking partner wired into his life's work. 99.99% of people won't do this because it requires reflection + setup. But once the vault exists, the agent stops being generic. It starts thinking in your voice."

The 10-Step Workflow

1. Write Everything in Markdown

Daily notes, projects, beliefs, people, meetings — all in plain markdown files in Obsidian.

Use wiki-links ([[note-name]]) to mirror how your brain actually thinks. Build a graph of interconnected ideas.

3. Install Obsidian CLI

Enable Claude Code to read your entire vault + the relationships between notes. The agent gains access to your full knowledge graph.

4. Use Reference Files, Not Re-Explanations

Stop reexplaining projects every session. Point Claude to reference files in your vault.

5. Build Custom Slash Commands

  • /context → Load your full life + work state
  • /trace → See how an idea evolved over months
  • /connect → Bridge two domains you've been circling
  • /ideas → Generate startup ideas from your vault
  • /graduate → Promote daily thoughts into real assets

6. Rule: Human Writes, Agent Reads

Human writes the vault. Agents read it, suggest, execute.

This boundary keeps you as the source of truth while letting agents act on your behalf.

7. Let Claude Surface Patterns

The agent can identify patterns you've been "unconsciously circling for years" by analyzing the vault.

8. Delegate From Inside Notes

"One sentence in Obsidian → agent handles the rest."

Example: Write "Draft investor update for Q1" in your daily note → agent reads project context, financials, recent wins from vault → generates the update.

9. Treat Writing as Leverage

The more you write, the more context your agents have. Your notes become a force multiplier.

10. Markdown Files Are the Oxygen of LLMs

Plain text, structured, interconnected, and portable. LLMs consume markdown better than any other format.

Why This Matters

The Personal Knowledge Graph as Agent Context

Most people use AI in stateless sessions — each conversation starts from zero. This workflow inverts that:

  • Your vault is the state
  • The agent reads it before acting
  • Context persists across all sessions

This is the difference between "ChatGPT" and "an AI assistant that knows everything about you."

Writing as Infrastructure

In this model, writing isn't just documentation — it's the substrate AI agents operate on. Every note you write becomes:

  • Reference material for future sessions
  • Training context for understanding your voice/style
  • Source material for pattern recognition

The "Reflection + Setup" Barrier

"99.99% of people won't do this because it requires reflection + setup."

Most people want instant gratification from AI tools. This approach requires:

  • Discipline to write consistently
  • Time to link notes and build the graph
  • Willingness to structure knowledge

But once built, the agent "starts thinking in your voice."

Markdown as the Standard

Greg's emphasis on markdown aligns with the broader "Markdown Manifesto" movement — plain text files are:

  • Portable (not locked in proprietary formats)
  • Future-proof (readable for decades)
  • Version-controllable (git-friendly)
  • AI-friendly (LLMs trained on markdown)

How This Connects to Existing Patterns

AGENTS.md Convention

This workflow is conceptually similar to skill-vs-agents-vs-claude-md — using markdown files to provide persistent context to agents. Obsidian becomes a personal-scale version of project-level AGENTS.md files.

Second Brain / PKM Movement

Obsidian is a leading tool in the "Personal Knowledge Management" (PKM) space, building on ideas from:

  • Zettelkasten (Niklas Luhmann's note-taking system)
  • "Building a Second Brain" (Tiago Forte)
  • Evergreen notes (Andy Matuschak)

The AI integration transforms PKM from "information retrieval" to "augmented execution."

Claude Code Skills

This workflow essentially treats your Obsidian vault as a personal skill — persistent context that Claude Code loads to understand your projects, preferences, and patterns.

The Design Repo Pattern

Similar to design-repo-pattern (draft post on keeping design inspiration separate from code), this approach advocates for a structured, version-controlled knowledge base that AI can query.

Practical Implementation

Tools Mentioned

  • Obsidian — markdown-based note-taking with graph visualization
  • Obsidian CLI — command-line interface for programmatic access
  • Claude Code — AI coding agent that can read vault via CLI
  • Custom slash commands — scripted workflows triggered by shortcuts

Example Workflow

  1. Morning: Write daily note in Obsidian
  2. Link to relevant projects, people, ideas using [[wiki-links]]
  3. Add a task: "Prepare pitch deck for Series A"
  4. Claude Code reads:
    • /context → loads company info, financials, team
    • /trace pitch-strategy → sees how pitch evolved over months
    • /connect fundraising + product-market-fit → bridges domains
  5. Claude generates draft deck based on vault context
  6. Human reviews, edits, publishes

Quotes

"how to use obsidian + claude code to build a 24/7 personal operating system and build your startup"

"stop reexplaining projects every session. use reference files instead."

"human writes the vault. agents read it, suggest, execute."

"let claude surface patterns you've been unconsciously circling for years."

"markdown files are the oxygen of llms."

"once the vault exists, the agent stops being generic. it starts thinking in your voice."

"99.99% of people won't do this because it requires reflection + setup."

Limitations & Considerations

Privacy

Giving an agent access to your entire vault (daily notes, beliefs, meetings) raises questions about what gets sent to API endpoints. Need to ensure sensitive info isn't leaked.

Maintenance Overhead

Requires consistent discipline:

  • Writing daily notes
  • Linking notes properly
  • Organizing vault structure
  • Updating reference files

If the vault gets stale, context degrades.

Tool Lock-In (or Not?)

Obsidian stores notes as markdown files in a local folder — no lock-in. But the custom slash commands and CLI integrations are Obsidian-specific. Migrating to another tool means rebuilding those workflows.

Context Window Limits

Even with a well-organized vault, there's a limit to how much an agent can load into context. Need strategies for:

  • Selective loading (only relevant notes)
  • Summarization (compress vault into key points)
  • Incremental updates (agent remembers recent sessions)

Added: February 25, 2026

Edit this page