Productivity · July 2026

Stop Re-Uploading the Same Files to AI. Build a Second Brain Instead.

Every time you paste a document into ChatGPT, it reads it fresh and forgets it the moment you close the tab. Here’s how to build a knowledge base that actually accumulates — using Claude Code and Obsidian.

You’ve almost certainly done this. You upload the same PDF to ChatGPT for the third time, ask a variation of the same question, get a decent answer — and then it all evaporates. Nothing sticks. Nothing connects to what you asked yesterday. Every session starts from an empty room.

That’s the core flaw in how most of us use AI for knowledge work: the tool has no memory of your world. The fix isn’t a better chatbot. It’s a second brain — a personal wiki that quietly gets smarter every time you feed it something new.

The idea in one sentence

Instead of dumping files into a chat and starting over each time, you keep a growing wiki. You drop raw material into one folder; an AI reads it, writes a clean summary page, and cross-links it to everything related. Ask a question later and it answers from the whole accumulated library — with sources.

The division of labor is the elegant part. You gather sources and ask good questions. The AI does the summarizing, linking, and organizing you never have time for. Think of Obsidian as the window you look through, and Claude as the librarian filing everything behind it. (The concept is adapted from Andrej Karpathy’s “LLM wiki” idea.)

What you need — just two tools

No code, no complicated stack. Two pieces:

• Obsidian — a free markdown editor. This is where your vault of notes lives, with linking, graph view, and instant search.
• Claude Code — the AI agent that reads your sources and builds the wiki. It runs in your terminal and works with a Claude subscription (no API key needed).

The setup, in three steps

1. Make an Obsidian vault with two folders inside: raw-sources/ for your untouched originals (the AI only reads these, never edits them) and wiki/ for the organized notes it writes.

2. Point Claude Code at the vault. Open the vault folder as your working directory and run it there. Now the AI can read and write inside your notes.

3. Add a CLAUDE.md rules file at the vault root — this is the most important part. It’s a plain-language instruction sheet telling the AI how your wiki is structured, how to write each page, and what to do when you add a source. Here’s the shape it defines:

Then it’s just three moves

Once the rules file is in place, using your second brain comes down to three repeatable actions:

• Build — the first time, tell Claude to read everything in raw-sources, create the wiki, cross-link it, and build the index.
• Ingest — drop in one new article or note and say “add this.” It writes a summary page and wires it into the index. One source can touch 10–15 existing pages.
• Query — once you’ve got a dozen sources in, start asking. This is where it earns its keep.

Claude reads the index, pulls the relevant pages, and answers with citations. And here’s the compounding trick: save the good answers back into the wiki as new pages. Every question you ask makes the next one sharper.

The honest caveats

Who this is for

If you’re a researcher, creator, consultant, or lifelong learner who’s tired of losing what you read, this is your system. Most people set it up, poke it once, and abandon it. The ones who push past that first week end up with something rare: a “second brain” that never loses your material again — and answers back.

We break down AI workflows that actually stick — tested, not hyped. That’s MindWiredAI. Come back for the next build.

Concept adapted from Andrej Karpathy’s “LLM wiki” idea. Based on a July 2026 how-to guide. Tool features change — verify current steps for Claude Code and Obsidian before you start.