I was consuming a ton — and forgetting almost all of it. So I built a second brain that captures what I learn and feeds it back when I'm doing real work. Now my AI references the people I actually learn from and my own takes — not a random internet thread.
No email, no gate. Scroll down — it's all right here.
If you don't put what you learn into action, you forget it. Monday's podcast is gone by Friday.
Your AI gives generic answers because it's working from the internet's average opinion — not yours.
Newsletters, episodes, voice memos — all of it scattered, unsearchable, basically lost.
No new app to learn. It's a folder and a few instructions — working with the tools you already have.
The newsletters and podcasts you subscribe to get pulled in and filed automatically. No copy-paste, no tabs.
Hear something good? Send a note or a voice memo. It's transcribed privately on your own machine and filed in seconds.
Everything's organized by the people you follow, the sources, and the topics — so your AI answers from your judgment, not a stranger's.
Everything to build yours tonight. No code, no subscription. Copy the prompts, make the folders, done.
Make one folder for your brain — in Obsidian, or any plain folder your AI can read. Inside it, these six. That's the whole structure.
Drop anything in inbox/, then paste this. Your AI reads it, transcribes if needed, summarizes, and files it — so you never have to categorize.
You are my second-brain librarian. My vault has these folders: inbox/, people/, sources/, topics/, log/, and index.md. Go through everything in inbox/ (and anything I've pasted). For each item: 1. Read it. Transcribe audio first; fetch and read any link. 2. Write a tight note in sources/: a 3-sentence summary, the key ideas with WHY each matters to me, any quotes worth keeping, and 1-3 ways I could use it. 3. Create or update a note in people/ for anyone notable it references (who they are, what they believe, why I follow them). 4. Create or update the relevant topics/ note with my evolving take. 5. Link everything with [[wikilinks]], add simple frontmatter (type, tags, date), update index.md, then delete the raw inbox item once filed. Be high-signal. Don't create noise. Never make me categorize — that's your job.
For a single podcast, article, or newsletter — paste the link or name it.
Process this into my second brain: [paste a link, or "the latest [podcast/newsletter] episode"]. Read the whole thing (transcribe if it's audio). Then write a source note in sources/ with: a tight TL;DR, the key ideas and why they matter to me, quotes worth keeping, the people mentioned (as [[links]]), and the topics it touches. Add or update the people/ and topics/ notes it connects to, and update index.md. Capture the signal, not every link.
This is the difference between generic answers and answers that sound like you at your most informed. Paste it before any real work.
Before you answer me on anything substantive, use my second brain as your reference point — not generic internet knowledge. First, check index.md and the relevant notes in people/, sources/, and topics/. Ground your answer in what I've actually saved: the people I follow, my opinions, and my own takes on what they've said. Cite my own notes where they apply. If my brain doesn't cover something, say so before falling back to general knowledge. The goal: answers that sound like me at my most informed — not the internet's average opinion.
Point your AI at the newsletters in your inbox and the podcasts you follow, and have it pull and file new ones for you. A scheduled task each morning, or just ask. Paste this to set the routine:
Each morning, pull anything new from the feeds I follow — my newsletters (search my email) and these podcasts: [list them]. Also grab anything I emailed myself or dropped in inbox/. Run it all through the "process my inbox" steps, update index.md, and give me a 3-line digest of what came in.
| A link or a thought | drop it in inbox/ (or a quick-capture note) |
| A file — PDF, screenshot, image | drag it into inbox/ |
| Audio or a voice memo | drop it in inbox/ — transcribed locally & privately |
| Something on your phone | email it to yourself, your AI sweeps it |
| Then… | say "process my inbox" |
Bonus, on a Mac, for private local transcription of any audio: install whisper.cpp + ffmpeg (one line with Homebrew), grab a model, and your AI can transcribe voice memos without anything leaving your machine.