Second brain
One folder of plain notes that every AI I use reads before it does anything: who I am, how I work, what each project is, and the rules for the AI. Every AI starts where the last one stopped.
- Why
- AI tools trap your context. I did not want to start from zero every time I switched tools, so the notes live in files I own, and each AI's own memory holds nothing but a pointer to them.
- How long
- Started 19 Aug 2026; about 700 commits by late October
How I built it
- I started from a friend's vault in Obsidian and rewrote the identity pages from a profile of myself.
- I put the folder under git with a private GitHub repo. A hook pushes every commit, and a nightly job commits anything left over.
- I added a README that tells every AI: read the rules folder first, then the task queue.
- I split memory like a brain: a short-term page of about 25 dated items, long-term pages per person and project, a daily journal, and an archive for anything untouched for two months.
- Every scheduled task is a pointer to a note in the vault, so changing the note changes the routine.
- To copy it you need
- Claude, or the AI of your choice, plus Obsidian (or any folder of text files) and a private GitHub repo. The gotcha: git hooks are not copied with the repo, so on a fresh copy they are silently off until you switch them on.
Try it yourself
Copy this prompt and paste it into ChatGPT or Claude to try it yourself.
I'm looking at Shweta Kamble's second brain on whatarewebuilding.ai: one folder of plain markdown notes that every AI reads first, with a README pointing to a rules folder, identity pages, one page per project and person, and a short-term page of about 25 dated items. The AI's built-in memory holds only a pointer to the folder. Help me start my own. Ask me what I do and which AI tools I use, then create the README, a rules page and three starter pages. The gotcha Shweta hit: a to-do list inside the folder grew to 13,400 words, so cap lists and rewrite them instead of appending. Ask me for anything you need.
Tools I used for this project
