WhyVault
Sign indeStart for free

AI second brain

A second brain your agents help keep.

The idea of a second brain is old and good: what you think and learn goes into one place where you can find it again. It usually failed on upkeep. AI agents change exactly that. Here are the three ways to build one, what the structure looks like, and how it works day to day.

Reading time: 11 minutesUpdated:

01

What a second brain is

The term was made popular above all by Tiago Forte, with his course and book Building a Second Brain. The idea: your head is for thinking, not for storing. What you read, hear and decide goes somewhere outside it, into a system you trust, so you can find it later and put it to use.

Forte describes four steps, CODE: Capture, Organize, Distill and Express. For organizing, he suggests PARA: Projects with a goal and an end, Areas you’re responsible for on an ongoing basis, Resources for topics and material, and an Archive for anything inactive. PARA doesn’t sort by topic, but by how close something is to action. It’s simple, and it works in almost any app.

The honest weak spot is familiar to anyone who has tried it: upkeep. Capturing is quick, but distilling, linking and tidying up get left for later. After a few months, the second brain is a full inbox of good intentions.

02

What changes with AI agents

An agent with access to your notes isn’t a better search box. It writes along with you.

–Classic second brainAI second brain
Who files thingsYou, by hand, after every conversation and every sourceThe agent writes along while you work; you review
DistillingOften left for later, because it takes timeThe agent summarises and extends existing notes
LinkingWhen you remember toThe agent searches before it writes and adds [[links]]
Finding things againYou search and read yourselfThe agent looks things up before it answers, and names the notes
What you doCapture, organize, distill, tidy upDecide what stays, and sharpen the rules
Who files things
Classic second brainYou, by hand, after every conversation and every source
AI second brainThe agent writes along while you work; you review
Distilling
Classic second brainOften left for later, because it takes time
AI second brainThe agent summarises and extends existing notes
Linking
Classic second brainWhen you remember to
AI second brainThe agent searches before it writes and adds [[links]]
Finding things again
Classic second brainYou search and read yourself
AI second brainThe agent looks things up before it answers, and names the notes
What you do
Classic second brainCapture, organize, distill, tidy up
AI second brainDecide what stays, and sharpen the rules

The method stays the same. What’s new is that someone else does the tedious half. For that, you need two things a classic second brain didn’t: rules the agent reads, and a history where you see what it changed.

03

Three ways to build an AI second brain

All three work. They differ in which AI writes along and where your notes live.

  • 1 · Note app with built-in AI

    Apps like Notion come with their own AI. Pro: no setup, the AI sits right next to your notes. Con: the app sets the AI and the structure. Some apps now let Claude or ChatGPT in via MCP too – Notion, for example – but the format usually isn’t plain Markdown, and moving out takes more work.

  • 2 · Local folder or Obsidian + MCP

    Markdown files on your computer, an MCP server and an agent like Claude Desktop or Claude Code. Pro: full control, endlessly extensible, free. Con: setup with config files, reachable only from that machine, and you only get a history with Git. Connect Obsidian to Claude

  • 3 · Hosted vault with MCP

    A Markdown vault online with a built-in MCP server, such as WhyVault. Pro: one URL for Claude, ChatGPT, Cursor and Claude Code, in the browser and on your phone too, with a history per note. Con: no scripts or plugins of your own, and WhyVault deliberately has no AI of its own – you bring your agent.

04

Which way fits you?

A quick way to decide:

  • The note app is enough if …

    you already live in one app, its AI is good enough for you, and you don’t need a second assistant that sees the same notes.

  • Local fits if …

    you enjoy tinkering, work on one machine, feel at home in the terminal and want your own scripts or plugins.

  • Hosted fits if …

    you use several AI tools, on the go and in the browser too, want to start without setup, and want to see and undo what an agent changed.

05

Local with MCP or hosted?

Ways 2 and 3 share what matters most: it’s your Markdown, and any agent that speaks MCP can work with it. The difference is the plumbing around it.

–Local folder / Obsidian + MCPHosted vault (WhyVault)
SetupA plugin or MCP server, plus an entry in your agent’s configCreate an account, connect your agent with https://mcp.whyvault.io
Reachable fromThe computer the folder lives on, while it’s runningClaude on web, desktop and phone, ChatGPT, Cursor, Claude Code
Rules for agentsOne file per tool: CLAUDE.md, AGENTS.md, Cursor rulesNotes in 9_System that every agent reads at the start
Seeing changesWith Git, if you commit firstAutomatic history per note, with the agent’s name
DeletingThe agent can delete any file it can reachOff by default; if you allow it, only into the trash
SearchWhatever you set up: grep, a local search engine, embeddingsFull-text search built in, no vector search
FreedomUnlimited: your own scripts, plugins, automationsFixed tools, no scripts, no background jobs
Where it livesOn your computerEU-hosted (Frankfurt), GDPR-compliant; what an agent reads, its provider processes too
CostFree, apart from your AI planFree up to 20 notes, Pro €9.99 a month
Setup
Local folder / Obsidian + MCPA plugin or MCP server, plus an entry in your agent’s config
Hosted vault (WhyVault)Create an account, connect your agent with https://mcp.whyvault.io
Reachable from
Local folder / Obsidian + MCPThe computer the folder lives on, while it’s running
Hosted vault (WhyVault)Claude on web, desktop and phone, ChatGPT, Cursor, Claude Code
Rules for agents
Local folder / Obsidian + MCPOne file per tool: CLAUDE.md, AGENTS.md, Cursor rules
Hosted vault (WhyVault)Notes in 9_System that every agent reads at the start
Seeing changes
Local folder / Obsidian + MCPWith Git, if you commit first
Hosted vault (WhyVault)Automatic history per note, with the agent’s name
Deleting
Local folder / Obsidian + MCPThe agent can delete any file it can reach
Hosted vault (WhyVault)Off by default; if you allow it, only into the trash
Search
Local folder / Obsidian + MCPWhatever you set up: grep, a local search engine, embeddings
Hosted vault (WhyVault)Full-text search built in, no vector search
Freedom
Local folder / Obsidian + MCPUnlimited: your own scripts, plugins, automations
Hosted vault (WhyVault)Fixed tools, no scripts, no background jobs
Where it lives
Local folder / Obsidian + MCPOn your computer
Hosted vault (WhyVault)EU-hosted (Frankfurt), GDPR-compliant; what an agent reads, its provider processes too
Cost
Local folder / Obsidian + MCPFree, apart from your AI plan
Hosted vault (WhyVault)Free up to 20 notes, Pro €9.99 a month

Both are plain Markdown with folders, frontmatter and [[wikilinks]]. So you can switch in either direction: import an Obsidian vault as a zip, or export your WhyVault as a zip and open it in Obsidian.

06

The structure of an AI second brain

An agent needs clear places, or it files everything somewhere. The WhyVault template separates four things: bounded rooms, timeless knowledge, dated notes and the rules.

0_Inbox/                    everything unsorted, emptied weekly
1_Knowledge/                timeless, flat, shared
  People/  Notes/  Sources/  Maps/  Principles/
2_Calendar/                 dated, thin
  Days/
  Reviews/
3_Rooms/                    everything bounded, with context
  Work/
    Client project/
      00 Overview – Client project.md
      01 Decisions/
  Private/
  Ventures/
9_System/
  Identity/                 USER.md, SOUL.md, IDENTITY.md, CONTEXTS.md
  Templates/
  Second brain rules.md     tells every agent where things go
3_Rooms/
Everything with a history: a client, a venture, your family. A new room starts with one file, the overview. 01 Decisions/ and 02 Meetings/ only appear when you need them.
1_Knowledge/
What belongs to no room: people, concepts, sources. A person exists only once, even if they show up at work and in private life. How an agent grows this into a wiki is in the LLM wiki guide.
2_Calendar/
Daily notes and reviews, kept short. This is where the weekly review goes that the agent writes with you.
9_System/
Who you are, how things should sound and how the agent should work. Every connected agent reads this folder at the start via vault_overview. More on the structure
9_System/Second brain rules.md
# Second brain rulesYou help me keep my second brain. I decide, you write along.## Before you answer- Look in the vault before you answer anything about my projects,  people or decisions. Name the notes you used.- If you find nothing, say so before you use general knowledge.## Where things go- Unsorted and raw material: 0_Inbox.- People, concepts, sources: 1_Knowledge. If the note exists, extend it.  Never create a second note for the same thing.- Daily notes and reviews: 2_Calendar.- Anything about a project or area of life: its room under 3_Rooms.- No tasks or deadlines in the vault. They belong in my task tool.## How you write- Decisions get their own note in the room’s `01 Decisions/`:  what, why, what we rejected, which assumptions hold.- Extend an existing note rather than creating a new one.- Link with [[Title]], freely inside a room, outside it only to 1_Knowledge.- Answer first, then details. Short notes, clear titles.## What you don’t do- Don’t delete or move anything. Tell me what should be tidied up.- At the end, tell me in one line what you created or changed.

07

The rules note to copy

The heart of an AI second brain isn’t the app, it’s a short note that tells every agent how to write along. Copy it and adapt it.

  1. Read first, then answer. The most important line. Without it, the agent treats your vault as storage, not as memory.

  2. Fixed places. Each kind of note has exactly one home. Then the next agent finds what the last one wrote.

  3. Extend, don’t multiply. Duplicate notes are the most common reason a second brain gets messy, with or without AI.

  4. One line of feedback. That way you always know what happened, and the history shows you the details.

08

Four workflows for every day

This is what an AI second brain looks like in a normal week. Every prompt works in any connected agent.

  1. 1

    The weekly review

    Fridays, ten minutes. The agent reads what happened in your rooms this week and drafts a review. You add what mattered to you, and it files it in 2_Calendar/Reviews/.

    Read what changed in my rooms this week and draft a weekly review: what was decided, what was learned, what’s still open. Then ask me three questions and save the result in 2_Calendar/Reviews.
  2. 2

    Capturing decisions

    Right after you’ve decided something with the agent. It writes the note with the reasons and the rejected alternatives into the right room. What a good decision note looks like is in the decision log guide.

    Capture what we just decided as a decision note in the right room: what, why, what we rejected, and which assumptions it depends on.
  3. 3

    From reading to knowledge

    Drop an article, meeting notes or a podcast transcript into the inbox, and the agent works it in: a source note, and the concepts land in existing notes in 1_Knowledge instead of gathering dust as another bookmark.

    Work the new note in the inbox into my knowledge: create a source note, extend the concept notes it touches, and tell me where the source contradicts older notes.
  4. 4

    Project context in every tool

    Before you carry on in Cursor, ChatGPT or a fresh chat: the agent reads the room’s overview and latest decisions. You don’t have to explain yourself again.

    Read the overview and the latest decisions in the room “Client project” and sum up in five sentences where we stand. Then let’s continue.

09

What it feels like

Monday morning, a new chat, and the agent still knows what’s going on:

Claude

  1. You: Where are we with the client project?

  2. Tool: read_note · “00 Overview – Client project”

    Claude: According to the overview, the relaunch runs until the end of November. Last week you decided to push the shop integration into a second phase, because the client’s API isn’t ready yet. Hosting is still an open question.

  3. You: We settled hosting on Friday: we’re staying with the current provider, moving isn’t worth it before launch. Write that down.

  4. Tool: create_note · “01 Decisions/Staying with the current host”

    Claude: Saved, with the reasons and the move as the rejected alternative. I’ve marked the open question in the overview as answered and linked it to the note.

Tomorrow you ask ChatGPT about hosting and get the same answer.

10

How to start your AI second brain in WhyVault

From zero to your first entry, without any code:

  1. 1

    Create a vault

    Sign up and pick the WhyVault template. If you’d rather think in projects, take the PARA template. Both come with folders and system notes.

  2. 2

    Connect your agent

    Add the URL https://mcp.whyvault.io in Claude, ChatGPT, Cursor or Claude Code and sign in once. Guides for every agent

  3. 3

    Let it get to know you

    The interview prompt in 9_System/Setup/ has your agent ask you questions and drafts USER.md, SOUL.md and the other identity notes.

  4. 4

    Add the rules note

    Create 9_System/Second brain rules.md with the text above, or ask your agent to. From then on, every connected agent reads it at the start.

  5. 5

    Check the history

    Every change an agent makes shows up in the history with its name. If something’s off, undo it with one click. History and export

A second brain doesn’t fail at capturing. It fails at upkeep. Get help with that part.

?

Frequently asked questions

What is an AI second brain?

A second brain where AI agents don’t just search, they write along: they read your notes before they answer, and file decisions, summaries and reviews themselves. You decide what stays, and sharpen the rules.

Do I need Tiago Forte’s method or PARA?

No, but they help. What matters is that every kind of note has a fixed place and the agent knows those places. When you create a vault, WhyVault offers the WhyVault template with rooms and a PARA template. The structure in detail

Can I use my existing Obsidian vault?

Yes, in two ways: locally with an MCP server that only your computer can reach, or by importing the folder into WhyVault as a zip. Folders, frontmatter and [[wikilinks]] are kept. Connect Obsidian to Claude

Which AI agents work with WhyVault?

Any agent that speaks MCP: Claude on the web, on desktop and on your phone, ChatGPT in developer mode, Cursor, Claude Code and other MCP clients. They all use the same URL. Connect agents

Does WhyVault have its own AI?

No, on purpose. WhyVault is the memory, not the assistant. You work with the agent you already use, and switch whenever you like. The notes stay the same.

What if the agent writes something wrong?

Every change is a version with the agent’s name. You undo it with one click, or roll back everything one connection changed since a point in time. Deleting is off by default.

Where are my notes stored?

WhyVault is hosted in the EU, in Frankfurt, and GDPR-compliant. What an agent reads from your vault is also processed by its provider, such as Anthropic or OpenAI. You can export everything anytime as a zip of plain Markdown. History, export and privacy

All guides

Your second brain, every agent.

Create a vault, connect your agent, add the rules note. Free up to 20 notes, no credit card.

Start for free