YOU CAN BUILD AN AI SECOND BRAIN IN 15 MINUTES. No coding experience. No $1000 course. Here is the entire setup. Step 1: Download Claude Desktop. Step 2: Download Obsidian. Step 3: Create a new vault and start dropping .MD files into it. Step 4: Tell Claude Code to... connect to your vault using Karpathy's prompt: That is it. Your entire knowledge base becomes searchable, connectable, and queryable by the most powerful AI model on earth. Every note you have ever written. Every idea you have ever captured. Every resource you have ever saved. Claude reads all of it, finds connections you missed, and surfaces insights from your own thinking that you FORGOT you had. Most people use Claude as a search engine. The people building second brains use it as an INTELLIGENCE LAYER on top of everything they know. That gap is the gap between asking Google a question and having a research partner who has read everything you have ever written. Bookmark this. Build it tonight. Follow CyrilXBTshow more

Unfortunately, this will work for few hundreds of documents only, I write about it a while ago

the real issue is that claude code isn't designed to be a passive search engine. it's an active coding agent. if you point it at a vault of messy text files without a proper ignore file, it's going to spend half its token budget trying to "refactor" your raw notes and journal entries.

The concept is powerful, but doesn't the quality of your "second brain" depend entirely on how well you organize your notes?

Cooool!! Be honest I still have knowledge base but I haven’t looked at it for a long time

Looks nice does nothing.

Claud + Obsidian good combo in order to create an assistant for yourself in almost all areas.

The gap is not between search and intelligence layer. It is between "Claude read my notes once" and "Claude maintains a running model of what I have actually internalized versus what I only captured." Most second brain setups never cross that line.

@cyrilXBT honestly, this sounds cool but isn't Claude Desktop invite-only or something? how do I get in?

такая хуита помоему

Claude + Obsidian is a very powerful combo for working with ai

did that exact setup months ago. the 15min part is true, keeping it fed daily is the actual work nobody mentions

Systems >>> second brain.

"most people use claude as a search engine, the people building second brains use it as an intelligence layer on top of everything they know" — that line is the whole point right there

Can someone name something that you've done with this that couldn't have been done otherwise? What real-life value has this provided other than a cool looking graphic doing what a free ChatGPT account could do.

Why not take al the data you have put in a vector store and slap a vector search. YOu have a personal google.

What is the practical limit of size here? 2 million concepts? Nothing is unlimited

Obsidian plus Claude works because markdown is the closest thing to a universal LLM context format. The bottleneck is not the setup, it is what you actually feed the vault over time.

The most defensible conclusion is not: Small models have defeated frontier models. It is: A model’s apparent intelligence depends heavily on the system of planning, memory, validation, scheduling, and recovery in which it operates.

Step 3 hides a surprisingly difficult part: most people’s existing knowledge isn’t already in clean .md files. It’s scattered across PDFs, DOCX files, scans, and inconsistent exports.Preparing and reviewing those documents before they enter the vault is almost a separate workflow of its own.

thanks for the setup

Every note you've ever written — still waiting to be broken into atoms.

@grok how safe is this?

W share

Gonna try building this tonight

You don't even need Claude Desktop for this. Total setup time is closer to three minutes, and there's no Obsidian, no manual vault config, no wiring your own MCP connection by hand. Built an app for exactly this, The Curator. Point it at your sources, articles, notes, PDFs, transcripts, whatever you're feeding it, and it automatically writes the wiki pages itself. Entities, concepts, summaries, all interlinked, all cross-referenced, updating existing pages instead of duplicating them every time you ingest something new. Same underlying idea as the Karpathy setup, none of the manual bookkeeping. It goes further too. Beyond a personal second brain, you can spin up a Shared Brain, a collective wiki for a team or cohort. Everyone keeps their own private graph, opted-in domains sync into the shared one, and the group's knowledge compounds together instead of living in scattered notes nobody reads twice. It ships with a built-in MCP server as well, so any agent you run, Claude Code, Claude Desktop, whatever harness, connects directly and can read and write to the graph mid-session. That's the part most second-brain setups don't have, persistent context that survives past a single chat. Fully open source, MIT licensed, runs entirely local on your own machine.

Claude finds connections you missed. But the deepest ones aren't between notes. They're inside them. Same topic. Opposite conclusions. Both mine. Both buried.

A second brain is a cute name for a folder full of notes that still needs retrieval rules, source hygiene, and a human who knows when the answer is stale. Memory without provenance is just organized hallucination.

The second brain isn't abt storing more it's abt finding what u already know

great share

Bookmarked. I have been meaning to do this!

“But Claude keeps burning tokens on tool calls and attempted conversions of PDFs I drop in the raw folder, produces no output, and then tells me to upload the PDF to the chat, where it will burn more tokens to read it…!” Yeah, that really sucks, huh? Why not automate it? You drop in a pdf. Seconds later, Inbox Ingestor protects a clean md file in the raw folder next to the pdf. THEN you point Claude at the md and let him parse it at a tenth of the token cost you would otherwise have wasted. Runs in the background as a lightweight watchdog script tied to the raw folder (or wherever you point it). Best of all? It’s totally free.