Why this exists
I’ve always written notes on my phone whenever I get a new idea. Not long ago I heard about someone using AI to help inspire new projects from a second brain; that alone convinced me it was worth building.
The whole pitch is: dump everything in, do almost nothing to maintain it, and have it there when you need it.
What it does
The vault holds four kinds of source material, all in plain markdown:
- Notes I write in Obsidian on phone or PC, captured into an
Inboxand folded into curated topic notes - Books dropped in as PDFs, auto-extracted to text by a Python script
- Chats with ChatGPT and Claude Code, imported in bulk from each platform’s export format
- Web clips saved through the Obsidian Web Clipper
A wiki/ layer sits on top: synthesized concept and people pages built from the raw notes by a /process-inbox slash command. Sensitive material (credentials, PII) gets quarantined into a separate folder during processing so it never lands in the wiki.
How it works
The whole thing is plain markdown files on iCloud: no database, no proprietary format.
- Capture: Every new note lands in
00-Inbox/. Phone, PC, web clipper, and the import scripts all write there - Sync: iCloud handles file-level sync between Windows and iPhone, no extra service in the middle
- Process: A
/process-inboxcommand in Claude Code reads the inbox, synthesizes a wiki page (or extends an existing one), and quarantines anything sensitive - Query: Obsidian’s full-text search handles everyday lookup. Claude Code is the chat with your brain interface; it can read the entire vault as plain markdown and reason across notes, chats, and books at once
One choice that mattered: keeping the AI out of band. Claude doesn’t run on every read; it runs when I run a slash command, on data I see before it lands.
What’s next
- More intentional capture. Right now this is a broad dump of my own writing: notes, AI chats, books. The next move is curating for a topic: feeding it material on, say, semiconductors, so it can quiz me on what I haven’t internalized yet, rather than just summarize what I already know
What I learned
I have a lot of chats with AI models. Watching them all pile up in one place has made me more aware of my own behaviors around reading and writing, including the fact that most of my own notes are 1–2 sentence scribbles (AI use isn’t helping with this). It’s nudging me toward longer-form writing.
Status
Shipped. Notes flow in daily, the wiki layer grows when I run /process-inbox, and Claude Code in the vault is now my default way to talk to my brain.