You don't need to be technical to build this. You need Obsidian (free), Claude Code, and articles you want to remember. This guide tells you what to build, why each piece matters, and how to hand it off to Claude to do the actual work.

The Problem You Already Likely Have

You read constantly. Articles, newsletters, papers, podcasts. You highlight things, save links, maybe drop them into a folder or a Notion page.

When you actually need what you learned… nothing. You remember reading something about X three weeks ago but have no idea where it went.

ChatGPT and NotebookLM rediscover knowledge from scratch every session. Manual wikis collapse under maintenance burden. Folders break down when things fit multiple categories. Tagging gets inconsistent the moment you stop being disciplined.

Here's What Fixes It

Instead of asking an LLM to search your documents every time you have a question, you have it build and maintain a wiki from your documents — once, incrementally, permanently.

You drop in an article. Claude reads it, writes a summary, creates pages for key people, ideas, and themes, links everything to what's already in the wiki, and flags where the new source contradicts the old ones.

Later you ask a question. Claude reads from the compiled wiki — where the cross-references already exist, the contradictions are already flagged, and the synthesis already reflects everything you've read. The answer quality is dramatically better because the hard work happened at ingest time, not query time.

That's what we're building: a personal knowledge base in Obsidian where the LLM does all the writing, cross-referencing, and maintenance. You do the reading, the sourcing, and the thinking.

How It Works (The 60-Second Version)

Obsidian is a markdown editor with graph view and wikilinks. Your wiki is just a folder of .md files — no database, no lock-in.

Claude Code runs inside VS Code (or the terminal, or as a desktop app) and can read and write files directly. It creates wiki pages, updates cross-references, and maintains the index — all without you touching a file.

CLAUDE.md is a schema file at the root of your vault that tells Claude how the wiki is structured and what workflows to follow. Without it, Claude is a generic chatbot. With it, Claude is a disciplined wiki maintainer that follows consistent rules every session.

You clip articles into a raw/ folder using Obsidian Web Clipper. You tell Claude to ingest them. Claude builds the wiki. You browse it in Obsidian and ask questions against it in Claude Code.

That's all you need to understand. Claude handles the implementation.

The Architecture (What You're Building)

Your Vault/
├── CLAUDE.md          # Schema — the rules Claude follows
├── raw/               # Your source documents (articles, PDFs, notes)
│   └── assets/        # Images and attachments
└── wiki/              # Claude-generated content — Claude owns this entirely
    ├── index.md       # Catalog of every page with one-line summaries
    ├── log.md         # Chronological record of all activity
    ├── sources/       # One summary per ingested source
    ├── entities/      # People, orgs, places, products
    ├── concepts/      # Ideas, theories, frameworks, terms
    ├── topics/        # Broader theme overviews
    ├── analyses/      # Filed answers from your questions
    └── meta/          # Overview pages, open questions