Channel
AI26 min

A second brain with AI: Claude and Obsidian

Every AI conversation starts from zero, and you are the one carrying the context. In this video I show how to connect Claude or Codex to Obsidian and build a second brain: an assistant that reads your notes, remembers your decisions and tidies up while you work. The video and the kit are in Portuguese.

Video cover: PH Alves holding the Claude and Obsidian icons, with the title Segundo Cérebro

The video goes live soon. Follow the channel on YouTube

In this video

  • Why AI answers generically and how memory changes the answer
  • Why Obsidian: the files are yours, in plain text, and work offline
  • The three layers: instructions, MCP and automation
  • Three ready-made routines: capture, close the day and weekly review
  • The installation from scratch, hands-on

Chapters

  1. 0:00Opening
  2. 1:18The problem
  3. 4:51The difference
  4. 8:23Who it is for
  5. 10:32Why Obsidian
  6. 12:57Myths
  7. 15:12The three layers
  8. 17:37Routines
  9. 20:20Hands-on
  10. 22:34Mistakes and kit

What is in the kit

INSTALAR-COM-IA.md
Script for the AI to install and customize everything for you.
instalar.sh
Installs everything into your vault with one command, without overwriting your files.
camada-1/
AGENTS.md (the AI's manual), CLAUDE.md, note templates and the daily note setup.
camada-2/
How to connect the vault to Claude Desktop through MCP.
camada-3/
Context hook, ready-made commands and routines.
PROMPTS.md
The prompts from the video and the prompt that writes your AGENTS.md.

What you need

  • Obsidian, free, with a vault created.
  • Claude Code (paid Claude subscription) or Codex (paid ChatGPT subscription).
  • Node.js, only if you install through npm or use layer 2.

Step by step

  1. Download and inspect the kit

    Unzip it. Before running anything, open the AI in the folder and ask it to read the files and tell you what they do. That goes for this kit and for anything you download.

    Read every file in this folder and tell me what each one does. Does anything touch files outside my vault folder?
  2. Let the AI install it

    In the Terminal, inside the kit folder, run claude or codex and send the prompt below. It finds your vault, shows the plan and asks for your ok before changing anything.

    Read INSTALAR-COM-IA.md and set up my second brain.
  3. Or install it by hand

    The installer creates the missing folders, copies the manual, the templates and the automation, and creates a git history for the vault. If one of your files already exists, the new version is saved next to it with the .segundo-cerebro suffix.

    chmod +x instalar.sh
    ./instalar.sh "/path/to/your/vault"
  4. Teach the AI who you are

    Open AGENTS.md and adjust the "Quem sou eu" section and the folder map. Or use the interview prompt further down: the AI asks up to 8 questions and writes the file with you.

  5. Run the first test

    Open the AI inside the vault and ask "good morning, what's on for today?". It should state the right date and what is pending without any explanation from you. Then run /capturar and watch the Inbox empty out.

    cd "/path/to/your/vault"
    claude

The three layers

Use only the first one or all three. Start with layer 1.

  • Layer 1

    Instructions

    One file, AGENTS.md, teaches the AI who you are and how the vault works. This layer alone already solves most of it.

    5 minutes

  • Layer 2

    MCP

    The socket that lets the chat app read and write in the vault, with tools to search, read and edit notes.

    10 minutes

  • Layer 3

    Automation

    Context hook, ready-made routines and git history. The AI opens the conversation already knowing the date, the last daily note and what is pending.

    15 minutes

Routines

/capturar
Empties the Inbox and sends each item to the right place. Originals go to the archive, nothing is deleted.
/fechar-dia
Logs in today's daily note what was done and decided in the conversation, along with what is pending.
/revisao-semanal
Builds the review of the last 7 days with decisions, stalled items and up to three focuses for next week.

Prompts from the video

Ask the vault

They work in Claude Code and Codex opened inside the vault folder, after installing AGENTS.md.

  • What did I decide about the app in the last few days, and why?
  • What did I promise someone and have not delivered yet?
  • What did I write about the most this week?
  • Which of my ideas connect with the book Atomic Habits?

Write to the vault

  • Note that I decided to launch the beta to 20 people on January 15, because I want feedback before charging.
  • I just had a call with Marina: she delivered the prototype, the stats screen is missing and she is back on Thursday. Log it.
  • Create a person note for Lucas, who referred Marina to me.

Write your own AGENTS.md

Paste it into the AI, inside your vault folder.

  • Read the structure of this vault (folders and a few notes from each) and interview me, one question at a time, to write an AGENTS.md with: who I am, the folder map with what goes in each one, and writing rules (search before creating, frontmatter, links between notes, where decisions and tasks live, never delete, always use the real date). At most 8 questions. At the end, show me the file before saving.

Five mistakes I already paid for

  1. Memory without an address is not memory

    The AI created a new note every time. Rule: search before creating.

  2. The AI makes up dates

    Decisions landed on the wrong day. A real-date rule, reinforced by the hook.

  3. Free-form tags turn into junk

    English, Portuguese, plural, singular. A closed list of tags or none at all.

  4. Connecting is not remembering

    Without the routine of logging, nothing stays. The brain only grows if you use it every day.

  5. Everything it reads goes to the model's server

    Confidential data stays out of the vault the AI reads, or use a local model.

Cautions

  • Everything the AI reads goes to the model's server. Passwords, client data and confidential information stay out of the vault.
  • Do not expose the vault to the internet just to connect it to an app.
  • The automation lets the AI edit the vault without asking for confirmation. That is why the git history comes with it.