Obsidian as a Second Brain for AI Agents

Complete research guide covering frameworks, AI integration workflows, plugins, capture systems, templates, and common pitfalls. Updated March 2026.

Pro tip: For a better voice, tap select all below then use iOS Speak Selection (Settings > Accessibility > Spoken Content > Speak Selection).

On Mac, click Select Section next to any heading, then right-click and choose Speech > Start Speaking. macOS will read just that section using your system voice (set in System Settings > Accessibility > Spoken Content).

The idea is simple: your Obsidian vault becomes a persistent knowledge base that AI agents (Claude Code, ChatGPT, Gemini CLI) can read, search, write to, and reason over. Instead of starting every AI conversation from scratch, you give it access to everything you've ever captured. This guide covers how to set that up well.

Part 1

01 Second Brain Frameworks

There are four major organizational frameworks people use in Obsidian. The consensus in 2025-2026 is that a hybrid approach works best, not picking one system rigidly.

PARA Method (Tiago Forte)

Organizes everything into four top-level folders: Projects (active, with deadlines), Areas (ongoing responsibilities), Resources (reference material by topic), and Archive (inactive stuff). It is action-oriented and beginner-friendly.

Intuitive folder structure, easy to start
Great for managing active work and projects
Can become folder-heavy if over-applied
Weak at capturing deep knowledge connections

Zettelkasten (Niklas Luhmann)

Atomic notes with unique IDs (like 202603131430) connected via links. Every note captures one idea. Non-hierarchical by design. The power comes from linking ideas across domains, which is exactly what makes it good for AI retrieval.

Excellent for building deep, interconnected knowledge
AI loves atomic notes: easier to embed, search, and retrieve
Steeper learning curve, can feel disorienting at first
Maintenance overhead grows with vault size

Maps of Content (MOCs)

A MOC is simply a note that serves as a curated index or entry point to a cluster of related ideas. Think of it as a dynamic table of contents that you build bottom-up as your vault grows. Created by Nick Milo (Linking Your Thinking). You do not need MOCs when your vault is small, but they become essential once you pass a few hundred notes.

Flexible, emergent structure that grows with you
Great navigation layer for both humans and AI
Requires active curation to keep useful
Can overlap confusingly with folders if not intentional

Johnny Decimal

A strict numbering system (10-19 for Finance, 20-29 for Projects, etc.) that works well for file systems and reference-heavy vaults. Less popular in Obsidian because it fights against the link-first philosophy, but useful as a mental model for your top-level folder structure.

Recommendation: Use PARA as your top-level folder structure (it gives notes an obvious home) combined with Zettelkasten-style atomic linking within those folders. Add MOCs as your vault grows past 200+ notes. This is the "dual-engine" approach that most experienced users converge on.

Part 2

02 Obsidian + AI Agent Workflows

The biggest development in 2025-2026 is connecting Claude Code directly to your Obsidian vault. Since Obsidian stores everything as local plain-text Markdown files, any CLI agent with filesystem access can read and write to your vault natively.

The Claude Code + CLAUDE.md Approach

This is the most popular and powerful setup right now. You open a terminal in your vault folder, run claude, and the agent has full access to every note. The key file is CLAUDE.md at your vault root, which acts as persistent instructions that Claude reads at startup every session.

What goes in CLAUDE.md:

Example CLAUDE.md for a vault:

# Vault Instructions for Claude

## Structure
- /Projects    — Active projects with deadlines
- /Areas       — Ongoing areas of responsibility
- /Resources   — Reference material by topic
- /Archive     — Completed/inactive items
- /Inbox       — Unsorted captures, process weekly
- /Templates   — Note templates
- /Daily       — Daily journal notes (YYYY-MM-DD.md)

## Note Conventions
- Filenames: kebab-case (e.g., meeting-with-client-name.md)
- Every note needs YAML frontmatter with: tags, created, status
- Use [[wikilinks]] for internal links, not markdown links
- One idea per note when possible (atomic notes)

## Frontmatter Standard
---
tags: []
created: YYYY-MM-DD
status: seedling | growing | evergreen
source: ""
related: []
---

## Rules
- Never delete notes. Move to /Archive instead.
- When creating notes, always use the appropriate template.
- Add [[MOC - Topic]] links when a note relates to an existing MOC.
- Daily notes go in /Daily/YYYY/MM/ subfolder structure.

MCP (Model Context Protocol) Integration

MCP allows Claude (and other AI tools) to interact with Obsidian through a standardized protocol. Plugins like Claudian and obsidian-ai-agent embed Claude Code directly inside Obsidian's interface, giving you agentic capabilities (file read/write, search, bash execution) without leaving the app.

What AI Agents Can Actually Do With Your Vault

Making Your Vault AI-Friendly

The structure decisions that matter most for AI retrieval:

Part 3

03 Essential Plugin Stack

The community consensus is to standardize on 8-12 vetted plugins. Plugin sprawl fragments your workflow. Here is the recommended stack for an AI-integrated second brain:

Tier 1: Must-Have Foundation

Templater

Far more powerful than core Templates. Supports dynamic dates, file creation hooks, conditional logic, and JavaScript execution. Essential for consistent note creation. AI agents can reference your Templater templates when creating notes.

Dataview

Treats your vault like a database. Write queries to create dynamic lists, tables, and dashboards from your frontmatter metadata. Example: TABLE status, created FROM #project WHERE status != "archived" SORT created DESC. This is what makes consistent frontmatter pay off.

Calendar

Visual calendar for daily notes. Click a date, get a note from your daily template. Pairs with Templater for auto-populated daily journals.

Tier 2: AI Integration

Smart Connections

The most popular AI plugin. Uses embeddings to find semantically related notes (not just keyword matches). Chat with your entire vault using RAG. Free, local-first by default, indexing happens on-device. Stores workspace in .smart-env/ folder.

Copilot for Obsidian

In-vault AI assistant with chat-based vault search, web and YouTube support, and agentic capabilities. Use your own API key. Good for asking questions across your notes in natural language. Complements Smart Connections (use both).

Claudian / Agent Client

Embeds Claude Code directly inside Obsidian. Full agentic capabilities: file read/write, search, bash execution, sub-agents. This is the bridge between Obsidian's GUI and Claude Code's CLI power. Agent Client also supports Codex and Gemini CLI.

AI for Templater

Extends Templater with AI chat commands using the OpenAI client library. Lets you call LLMs directly from template commands. Example: auto-summarize a meeting note on creation.

Tier 3: Quality of Life

Tasks

Manages tasks across your vault with due dates, recurrence, and Dataview-queryable metadata. Checkboxes become a real task management system.

Periodic Notes

Extends daily notes with weekly, monthly, quarterly, and yearly note templates. Good for review cycles.

Linter

Auto-formats your notes on save: fixes YAML frontmatter, consistent heading levels, trailing spaces. Keeps your vault clean for both you and AI parsers.

Quick Add

Create notes from templates with a single hotkey or command. Supports multi-choice templates, macros, and capture workflows. Faster than navigating to file > new note > pick template.

Text Generator

Versatile plugin for generating text using various AI providers (OpenAI, Anthropic, Google, local models). Good for in-note generation tasks.

Part 4

04 Capture Workflows (Especially iPhone)

The hardest part of any second brain is getting information into it. Here are the best approaches ranked by friction level, with a focus on mobile capture.

iOS Quick Capture Options

Quick Draft (New in 2025)

Companion app for Obsidian that supports text, voice notes with transcription, photos, and attachments. Sends directly to your vault. Currently the most complete mobile capture solution.

Quick Capture for Vault Notes

iOS app that handles voice recordings with automatic transcription, document scanning, and file/photo sending to your vault. Available on the App Store.

iOS Shortcuts (DIY)

Build a Shortcut that pops up a text input and appends to a capture file in your vault. If you have an iPhone 15 Pro or later, bind it to the Action Button for one-tap capture. The shortcut writes to iCloud or uses Obsidian's URL scheme (obsidian://new).

Funnel

iPhone app designed specifically for quick capture into Obsidian. Minimal friction, designed for speed.

Obsidian Mobile App

The official app works but is not optimized for quick capture. Opening it, navigating to the right folder, creating a note, and typing is too many steps for a fleeting thought. Use it for reading/editing, not capturing.

Capture Processing Workflow

The key insight: separate capture from processing. Everything goes into an /Inbox folder first, tagged with a date. Then during a weekly review (or let Claude do it), process inbox items into proper notes with templates and frontmatter. A Dataview query in your daily note can show unprocessed inbox items:

```dataview
TABLE file.ctime as "Captured"
FROM "Inbox"
SORT file.ctime DESC
```

Web Clipping

Use the Obsidian Web Clipper browser extension (official, released 2024) to save web pages as Markdown directly into your vault with proper frontmatter. On iOS, share sheet integration sends articles to your vault. Readwise is another popular option that syncs highlights from Kindle, articles, podcasts, and tweets into Obsidian automatically.

Voice Notes

Record voice memos, then have Claude Code transcribe and process them. Workflow: capture audio with Quick Draft or Apple Voice Memos, then run a Claude Code command that transcribes the audio and creates a structured note from it. Some users run Whisper locally for transcription.

Part 5

05 Template Examples

Good templates have three things: useful frontmatter that AI can query, clear structure for humans, and Templater variables for automation. Here are practical examples.

Daily Journal

---
tags:
  - daily
created: <% tp.date.now("YYYY-MM-DD") %>
week: <% tp.date.now("WW") %>
---

# <% tp.date.now("dddd, MMMM D, YYYY") %>

## Focus
- [ ] Top priority:
- [ ] Secondary:
- [ ] If time:

## Log
-

## Reflections


## Inbox to Process
```dataview
LIST FROM "Inbox"
WHERE file.ctime >= date(today) - dur(1 day)
```

Meeting Notes

---
tags:
  - meeting
created: <% tp.date.now("YYYY-MM-DD") %>
attendees: []
project: ""
status: active
---

# Meeting: <% tp.file.title %>

**Date:** <% tp.date.now("YYYY-MM-DD HH:mm") %>
**Attendees:**
**Context:** [[]]

## Agenda
1.

## Notes


## Action Items
- [ ]

## Decisions Made


## Follow-up
- Next meeting:

Client / Contact Profile

---
tags:
  - person
  - client
created: <% tp.date.now("YYYY-MM-DD") %>
company: ""
role: ""
email: ""
phone: ""
status: active
last-contact: <% tp.date.now("YYYY-MM-DD") %>
---

# <% tp.file.title %>

## Overview
-

## Key Details
- **Company:**
- **Role:**
- **How we met:**
- **Working on:**

## Interactions
```dataview
TABLE created, tags
FROM [[]]
WHERE contains(tags, "meeting")
SORT created DESC
```

## Notes

Project Brief

---
tags:
  - project
created: <% tp.date.now("YYYY-MM-DD") %>
status: planning | active | paused | completed
deadline: ""
stakeholders: []
related: []
---

# Project: <% tp.file.title %>

## Objective
What are we trying to accomplish?

## Background
Why does this matter?

## Scope
- In scope:
- Out of scope:

## Key Milestones
- [ ]

## Resources & Links


## Updates
### <% tp.date.now("YYYY-MM-DD") %>
-

Research Note

---
tags:
  - research
created: <% tp.date.now("YYYY-MM-DD") %>
source: ""
status: seedling
related: []
---

# <% tp.file.title %>

## Key Takeaway
One sentence summary of the insight.

## Details


## Source
- URL:
- Author:
- Date accessed:

## My Thoughts


## Connected Ideas
- [[]]

Quick Idea / Fleeting Note

---
tags:
  - idea
created: <% tp.date.now("YYYY-MM-DD HH:mm") %>
status: inbox
---

# <% tp.file.title %>

<% tp.file.cursor() %>

Part 6

06 Common Mistakes to Avoid

Most people who abandon Obsidian do so because they over-engineer the system before using it. Here are the specific traps, based on community experience across thousands of users.

Mistake 1: Too Many Folders

The number one mistake. People create a folder for every conceivable category before writing a single note. Then they spend more time deciding where a note goes than actually writing it. Users with 6000+ notes report that reducing to 4-5 "pillar" folders eliminated the decision fatigue. Obsidian's search, backlinks, and tags surface what you need without folder hierarchies.

Fix: Start with 5-7 top-level folders max. Use tags for categorization. Let structure emerge from use.

Mistake 2: Perfecting the System Before Using It

Spending weeks choosing between PARA and Zettelkasten, configuring 30 plugins, designing the perfect template system, and never actually taking a note. The system only becomes clear through use.

Fix: Start with daily notes and an inbox folder. Add structure only when you feel pain from not having it.

Mistake 3: Plugin Overload

Installing every plugin that looks interesting. Each plugin adds cognitive overhead, potential conflicts, and update maintenance. The community recommends 8-12 vetted plugins maximum.

Fix: Install the core stack (Templater, Dataview, Calendar). Add others only when you have a specific problem to solve.

Mistake 4: Inconsistent Frontmatter

This one kills AI integration. If half your notes have tags: in YAML and half use inline #tags, if dates are sometimes MM/DD/YYYY and sometimes YYYY-MM-DD, if some notes have frontmatter and some do not, then Dataview queries break and AI has unreliable metadata to work with.

Fix: Define your frontmatter standard in CLAUDE.md and templates. Use the Linter plugin to enforce it. Run periodic Claude Code sweeps to fix inconsistencies.

Mistake 5: Writing Long Notes Instead of Atomic Notes

Dumping everything about a topic into one massive note. This makes linking, retrieval, and AI embedding all worse. A 5000-word note about "Marketing" is less useful than twenty 250-word notes about specific marketing concepts that link to each other.

Fix: One idea per note. If a note is getting long, split it. Link the pieces together.

Mistake 6: Not Processing the Inbox

Capturing everything into an inbox folder and never processing it. After a month you have 200 unprocessed captures and the system feels broken.

Fix: Process inbox items in your weekly review. Or set up a Claude Code skill that helps you process them: "look at my inbox, suggest where each note should go, add proper frontmatter."

Mistake 7: Ignoring Backlinks

Using Obsidian like a file system instead of a knowledge graph. The entire point of Obsidian over a folder of text files is bidirectional linking. If you are not creating [[wikilinks]] between notes, you are missing the core value.

Fix: Every note should link to at least one other note. Build this habit from day one.

Part 7

07 Real Examples & Notable Guides

These are specific people and resources doing Obsidian + AI agent workflows well, with links to their work.

Blog Posts & Guides

Obsidian Plugins & Tools

Framework & Method Resources

Key YouTube Channels

Part 8

08 Quick-Start Action Plan

If you want to go from zero to a working AI-integrated second brain, here is the sequence that avoids over-engineering.

Day 1: Foundation

  1. Create a new Obsidian vault
  2. Create these folders: Inbox, Projects, Areas, Resources, Archive, Templates, Daily
  3. Install Templater, Dataview, and Calendar plugins
  4. Create your daily note template and meeting note template (use examples from Part 5)
  5. Set up daily notes to auto-create with your template

Day 2: AI Integration

  1. Open your vault folder in terminal, run claude (or claude /init)
  2. Create a CLAUDE.md file describing your vault structure (use the example from Part 2)
  3. Install Smart Connections plugin for in-vault semantic search
  4. Test: ask Claude Code to create a note using your template, search your vault, generate a MOC

Day 3-7: Capture Pipeline

  1. Install Quick Draft or Quick Capture on your iPhone
  2. Set up an iOS Shortcut for one-tap text capture to your Inbox folder
  3. Install the Obsidian Web Clipper browser extension
  4. Start using it. Take notes. Capture ideas. Do not reorganize yet.

Week 2+: Iterate

  1. Do a weekly review: process Inbox items into proper notes
  2. Notice what feels clunky. Add plugins or templates only to solve real friction.
  3. When you have 50+ notes, create your first MOC
  4. Have Claude Code audit your vault: "find notes missing frontmatter" or "suggest connections between these notes"
  5. Build Claude Code skills/SOPs for your repeating workflows

The core principle: Your vault is a living workspace, not an architecture project. The best system is the one you actually use every day. Start simple, let AI handle the maintenance and organization tasks, and add complexity only when you feel the need.

Ready