Skip to content
Go back

The Vault That Thinks With You

20 January, 2026

5 min read

•

285 views

Obsidian Vault

Your notes are lonely.

That brilliant insight from last Tuesday? Buried in a file you'll never open again. The connection between your API architecture decision and that article about performance tradeoffs? Lost forever because the two notes never met.

Most note-taking is archaeology. You dig through folders hoping to find something useful. But what if your vault could think with you, surfacing connections, maintaining context, and evolving alongside your understanding?

This is a practical guide to making that happen using Obsidian and Claude Code, combining the best local-first knowledge base with an AI that can read, understand, and help you navigate your thinking.

The Shift: Writer to Curator

Traditional note-taking puts you in the writer's seat. You capture, you organize, you maintain. It's exhausting, which is why most systems collapse under their own weight.

AI-powered note-taking inverts this. You become the curator, deciding what matters, what connects, what stays. The AI handles the grunt work: finding related notes, suggesting links, maintaining consistency.

Your job shifts from "write everything down" to "judge what's valuable." The same way a senior engineer reviews code instead of writing every line, you review and refine what the AI produces.

Teaching Your AI: The CLAUDE.md Deep Dive

Claude doesn't know your system. You have to teach it.

The CLAUDE.md file is your teaching document, placed at your vault's root, it explains how your system works. Here's a practical structure:

# Vault Philosophy

This vault tracks projects, thinking, and reference material.
Core principle: capture quickly, structure deliberately.

## Structure (PARA-inspired)

- `01_Inbox/` - Capture zone. Everything lands here first.
- `02_Projects/` - Active work with clear outcomes
- `03_Areas/` - Ongoing responsibilities (health, finances, home)
- `04_Resources/` - Reference material organized by topic
- `05_Archive/` - Completed or inactive items

## Language Rules

- Default to English for technical content
- Swedish acceptable for personal notes
- Keep titles in English for consistent linking

## Frontmatter Standards

Every note needs:
---
created: YYYY-MM-DD
tags: [relevant, tags]
status: active | completed | archived
---

## Linking Expectations

- Link inline, not at the bottom. Write "because [[async beats sync]]"
  not "Related: async-beats-sync"
- Every note should have at least one incoming link
- Hub pages link to all related notes in a topic

## Naming Conventions

Notes are claims, not topics:
- *Don't*: "API design thoughts"
- *Good*: "GraphQL works best for client-driven queries"

The title should complete the sentence "I believe that..."

## Overrides

- Never create folders without asking
- Always preserve existing frontmatter when editing
- Suggest links but don't add them automatically

## Vocabulary

- WIP = Work in Progress (notes still being developed)
- PR = Pull Request (not Public Relations)
- MVP = Minimum Viable Product (not Most Valuable Player)
- POC = Proof of Concept (not Point of Contact)

The Daily Dance: Workflows

Morning Capture

Dump everything into 01_Inbox/. No structure, no formatting, just capture.

Inbox Processing (10 minutes)

Ask Claude to process your inbox:

"Review 01_Inbox/. For each note: suggest a destination folder, recommend links to existing notes, and flag if it should be split into multiple notes."

Review the suggestions. Move what makes sense. Delete what doesn't matter.

The Automation Layer

For repetitive patterns, create slash commands. Example for processing meeting notes:

# /process-meeting

1. Read raw meeting notes in 01_Inbox/
2. Extract: key decisions, action items (with owners), open questions
3. Create structured note in 02_Projects/[relevant-project]/Meetings/
4. Link action items to responsible person's [[People/Name]] notes
5. Update project hub with decisions and next steps
6. Flag any items needing calendar follow-up

This turns scattered notes into actionable structure in seconds.

Note Architecture

Before: The Isolated Note

# Team Sync Jan 20

Discussed the API refactor. Sarah thinks we should use GraphQL.
Mike mentioned performance concerns. Need to follow up.
Also talked about the release timeline.

This note will die alone. No links, vague title, no structure.

After: The Connected Hub

---
created: 2026-01-20
tags: [meeting, api-refactor, team-sync]
status: active
---

# API Refactor Discussion - Team Sync

Weekly sync covering GraphQL migration decision and Q1 timeline.

## Key Decisions

- Moving forward with [[GraphQL migration]] for new endpoints
- Keeping REST for legacy clients (see [[API versioning strategy]])
- Timeline: complete by end Q1 2026

## Action Items

- **[[People/Sarah]]**: Draft GraphQL schema by Jan 27
- **[[People/Mike]]**: Performance benchmarks for GraphQL vs REST by Feb 3
- **[[People/Me]]**: Update [[02_Projects/API Refactor/Timeline]] with new dates

## Open Questions

- Caching strategy for GraphQL queries → discuss in [[Architecture Review Feb 1]]
- Client migration plan for mobile apps → needs [[People/Emma]]'s input

**Status:** In progress
**Next sync:** Jan 27, 2026

The Backlink Footer

Always close with navigation:

---

**Back to:** [[02_Projects/API Refactor]] | [[Team Meetings Hub]]

This helps both you and Claude understand the note's place in the hierarchy.

Frontmatter for Tracking

Use frontmatter for state that changes:

---
created: 2026-01-20
status: action-items-pending
meeting_type: team-sync
action_owner: Sarah
due_date: 2026-01-27
---

Claude can query this: "Show me all meeting notes where status: action-items-pending and due_date is within the next 3 days."

Common Mistakes

Over-structuring too early

You don't need 15 folders on day one. Start with inbox, active work, and archive. Add structure when you feel friction, not before.

Using AI as search instead of thinking partner

  • Bad: "Find my note about the API refactor"
  • Good: "We're deciding between GraphQL and REST for the new API. What concerns and insights from past architecture decisions should inform this choice?"

Search finds files. Thinking partners synthesize.

Not reviewing Claude's work

AI makes mistakes. It will:

  • Create duplicate notes
  • Misunderstand your conventions
  • Add links that don't make sense

Always review. Your judgment is the quality filter.

Orphan notes

A note without incoming links is invisible to the network. Every note should answer: "What other note would link here?" If nothing would link to it, it either needs to be split into linkable claims or it's not worth keeping.

Maintenance as Meditation

This isn't busywork. Regular maintenance is how you stay connected to your thinking.

Weekly (15 minutes)

  • Inbox at zero
  • Check orphan notes (Obsidian graph view helps)
  • Update one hub with recent connections

Monthly (30 minutes)

  • Archive completed projects
  • Review and update CLAUDE.md
  • Look for notes that should be merged

Quarterly (1 hour)

  • Philosophy review: Is the vault serving its purpose?
  • Structure audit: Do folders still make sense?
  • Link quality: Are connections meaningful or cluttered?

The Vault Remembers

Here's the magic: Claude has no memory between sessions. But your vault does.

When you teach Claude something, a convention, a connection, an insight, encode it in the vault. Write a note. Update CLAUDE.md. Add a link.

Next session, Claude reads those notes and picks up where you left off. The vault becomes shared memory between you and every future AI session.

Your notes stop being static files. They become a living system that thinks with you, remembers for you, and grows alongside your understanding.

The question isn't whether to build this system. It's what you'll do with all that recovered mental space.