Graphify + Obsidian is INSANE: Build an AI Second Brain That Never Forgets

This is an AI second brain that never forgets. Point it at a folder and your whole project turns into a graph your assistant can query because right now your coding assistant has amnesia. Every new chat it has read nothing. Ask it how your login flow works and it opens every file from scratch to figure it out. Do that across 200 files on every single question and you pay a token tax each time. And stuffing your entire repo into a giant context window just makes the answers worse. On April 3rd, Andre Carpathy posted a fix and the dev world copied it in a single weekend. Stop making the model rebuild your knowledge every query. Give it a wiki it keeps up to date itself. 2 days later, a developer shipped exactly that as a tool called Graphify. From nothing, it blew past 90,000 GitHub stars in about 3 months. Pair it with Obsidian, and that graph becomes plain Markdown you own. So, here's the full picture. How graphify builds the graph, why Obsidian is its home, and the four ways it falls apart if you get lazy. And this was not some quiet launch. It turned into a genuine wave. That claude code plus Obsidian idea pulled over 50,000 likes on X in a single week. One post declaring notetaking dead hit 54,000 on its own. Carpathy's little gist alone picked up 14,000 stars and,900 forks. First, the actual problem Graphify is solving. The usual fix for a big code base is what's called rag. Chop your files into chunks, turn them into vectors, and search for the closest matches when you ask a question. That works great for pros. For code, it's shaky because the link between a function and the things that call it isn't about similar words. It lives in the call graph. You end up paying to load the whole library just to find one paragraph. Graphify flips that around. Instead of searching text on every question, it does the expensive reading once up front and squeezes your project into an explicit graph of entities and the relationships between them. After that, answering a question means walking the graph instead of rereading the files.