I Gave Every AI I Use the Same Memory: A Folder of Text Files

Key Takeaway: An AI second brain is not an engineering project. It is a folder of plain text files on your own computer that every AI tool reads before it starts working. Tools come and go. The folder stays yours.

Every New Chat Starts With a Stranger

If you have used any AI tool for more than a week, you have felt this. The assistant that knew your business yesterday has no idea who you are today. So you re-explain: your business, your customers, how you like things written, what you decided last month. Every session starts with a stranger.

And if you are the careful type, there is a second worry underneath: the more of yourself you pour into one AI product, the more trapped you are inside it. What happens to all that context if the price triples or something better shows up next year? That is not paranoia. That is good judgment.

Both problems have the same fix, and it fits in one ordinary folder.

An AI Second Brain Is Just a Folder of Text Files

I keep a private folder of plain text files on my own computer. Every AI tool I use, and I use several, reads that folder through one tiny "start here" file that points at everything else. That is the entire trick. One folder, one front door, and every assistant I work with shares the same memory.

The folder is backed up with full version history: every change is saved and reversible, like the edit history on a shared document. If a tool I use disappears tomorrow, I lose nothing. I point the next tool at the same folder and keep working.

The principle has a name, "File over AI," coined by Nick Milo: own your ideas and workflows in files you control, and rent the tool. Andrej Karpathy describes the same destination from the technical side: a second brain as a folder of plain text an AI works through the way you use a book: a table of contents, then the right page. Different worlds, same conclusion.

What Lives Inside Mine

Every file is something a human can read. No app, no database, no code. Here is the tour:

• me.md, the identity file. One file every assistant reads first: who I am, what I do, my values, my voice, and how I want to be advised. It sets the tone with "Call me Joseph." and spells out how I want advice: "Lead me with a ranked recommendation. Give me your pick first, labeled 'my pick,' then the reasoning. Do not hand me an unranked menu." I wrote those preferences once and have never repeated them since.
• A Vault Map. I call the folder the vault, and the map is its one-page manual: what lives where, so an AI finds the right note instead of trying to read everything. An AI can only read so much at once, so you hand it a map instead of a library.
• A Skill Map, plus skills. Repeatable processes written as plain instructions: how a briefing gets run, how emails get written. Any assistant can pick one up and follow it.
• Atlas. Durable knowledge and frameworks. The stuff that stays true.
• Calendar. Dated logs and decisions, so "when did we decide that?" always has an answer.
• Efforts. One note per active project, pointing at the real work.
• An AI-zone. A quarantine folder. Anything an AI writes lands here untrusted until I review it and promote it.
• A private area. Never leaves my machine, ever.

If that sounds like a lot, yours could be far simpler.

The Guardrails Are Most of the Value

A shared memory is only useful if you can trust what is in it. Four rules keep mine honest:

The core files are human-gated. An AI can propose an edit to my identity file or my curated knowledge; I approve it. Who I am never gets silently rewritten by software.

Secrets never go in. No passwords, no keys, no account numbers. Those live in a separate encrypted store built for that job. The folder holds knowledge, not credentials.

Every change is saved with history. Nothing is ever truly lost, and every assistant sees the same version of the truth.

Rewrites get audited. When an AI rewrites a curated file, the result gets compared against the previous version before it is accepted, so no specifics quietly vanish.

That last rule exists because of a real incident. During one rebuild, an AI rewrote a knowledge file and silently dropped 138 substantive items. Nothing looked broken. The audit caught it against the old version, and every item was restored. That scare is why the check now runs every time.

What You Actually Get

The payoffs are the point:

• You can switch AI tools freely. Nothing is trapped inside any one product, so you choose tools on merit, not on hostage math.
• Every assistant remembers you. A brand-new session starts already knowing your business, your voice, and your standards.
• Knowledge compounds. What an assistant learns today is filed and available tomorrow, instead of evaporating when the chat ends.
• You stay the editor. The AI drafts and proposes. You decide what becomes permanent.

None of this requires you to be technical. If you can keep a folder of documents, you can keep this. I have written before that the real work of AI is understanding your own business; this folder is that understanding, written down where any tool can use it. And if your instinct says keep it close to home, good: it lives on your computer, not inside someone else's product.

A Starter Version Is One Folder and Three Files

You do not need my whole setup. A starter second brain is one folder and three plain text files:

1. A "me" file. Who you are, what your business does, how you like to communicate, and what an AI should never do. One page.
2. A map. What lives where, even if "where" is just these three files for now.
3. A log. Dated decisions and lessons, newest first.

Paste the "me" file into any AI chat and you have already stopped introducing yourself from scratch. Point a tool that can read files at the folder and you have the real thing. Grow it only when the work asks for it.

One local note. If your business is on Staten Island, this philosophy hits print in Q4 2026: The Local Vibe is a premium 9x12 community card we mail free to neighborhoods across the island, and local businesses claim an ad slot and design it themselves with our AI Ad Builder. You build it, you own it.

Own the memory. Rent the tool. Everything else builds on that one move.

---

Joseph Philip Savino is the founder of Savino Marketing, based in Hazlet, NJ. He teaches startups and small businesses to adopt AI safely, with 9% of all proceeds going to charity.