AI powered Second Brain setup

Remember the last time you got really excited about using a computer?

A practical, budget-friendly AI setup that runs on your home computer.

Do you remember the last time you got really excited about using a computer or a piece of technology? I mean really, really excited?

I remember when I was a kid and upgraded my Commodore 64 to an Amiga 500. All the things I suddenly could do: connect a VHS camcorder and digitize images, edit them in Deluxe Paint, sample sounds, and make music in Soundtracker and similar magic. The next time I felt something similar was when I grasped how to create 3D models. All of a sudden I could build what I wanted, and computers had become so powerful that I no longer needed to leave mine running all night just to ray-trace an image of reflective balls.

Now I feel something similar, but about something much simpler to set up: basically, folders with text files.

An AI powered "second brain" - what is it?

An AI-powered second brain is your local digital workspace that captures, organizes, and executes tasks alongside you. Its memory and capabilities continuously evolve on your computer, adapting to any workflow whether you're managing finances, writing, doing research, vibe coding, creating 3D assets, or organizing digital archives.

Strengths

The key strength of the setup is its simplicity:

• Text files and folders: The skills and memory are stored in simple text files organized in folders. Transparency is total and you are not locked into any proprietary file format.
• AI-agnostic: Works with any AI solution that can read and write to your local hard drive.
• Local but accessible from anywhere: The brain's memory and skills are stored on your local hard drive in a human-readable format. Put the brain on, for example, Google Drive and you can access it from anywhere.
• Low effort to set up: Just create a folder and give the AI access to it.
• Low cost (even free): Works with paid AI modes (like Claude Code/Cowork) or with free tiers of, for example, Mistral or GitHub Copilot.
• Sharable / collaborative: Share subfolders with others and you get shared memory and shared skills. Add links to other people's Google Docs or similar documents in your source folders to automatically make use of material that others keep updated. Mirror your skills folder on SharePoint and you automatically have a shared brain with your colleagues at work.

The setup

A folder for your local brain

Create a folder on your computer where you will give the AI access to create and edit files.

• Source - a folder where you dump source material needed to carry out tasks. This can be images, transcripts, documents in various formats, or links to web pages.
• Output - a folder containing files you and the AI are working on and your finished output.
• Skills - a folder containing all your skills, i.e. specific instructions for the AI to follow when performing specific tasks.

You can organize this as you wish, for example with a subfolder per project or topic.

In the root folder, and in the project-specific subfolders, place these text files:

• Claude.md - a text file where you write, in plain human-readable language, global instructions for the AI always to follow. For example, language or tone of voice preferences.
• Skill.md - a text file in the skills folder where the AI stores special skills, such as how to create a certain type of document, structure a report, or carry out a similar task.
• Skills_changelog.md - a text file in the skills folder where I have instructed the AI to log whenever it makes a change to a skill. This might seem superfluous, but since the skills constantly evolve, I found it useful to keep a log of what changed, when, and by whom if you share the skills with others.

The content of these text files is written in plain language, not code or technical jargon. You can ask the AI to improve and clarify the instructions and the memory itself over time.

Sharing the skills - collective learning

The setup can be adapted for collaboration and shared learning. Sync your skills folder with your colleagues using a cloud or enterprise storage service, and you will collectively contribute to the brain's learning. If needed, store and track changes to the brain in GitHub for proper version control.

Choosing an AI and giving it access

You need an AI that can read and write files on your computer as well as access pages on the web. Claude Code or Cowork is popular and powerful, but in this example I have instead chosen Mistral, since you can register a free account that works well.

Mistral is entirely web-based and has no local client, but through a simple workaround you can let it read and write your local files. To do this:

1. Install VS Code, the free and simple code editor from Microsoft.
2. Inside VS Code, find and add the Mistral Vibe VS Code extension, then log in with your Mistral account.

If you don't like Mistral (the Mistral VS Code plugin can, for example, not transcribe images to text), you can register an account at GitHub and use the GitHub Copilot in the same way. To save tokens (i.e. expand your monthly free usage of Mistral or GitHub Copilot), you can create a new project in Google NotebookLM and upload your instruction and memory files and let Google's Gemini AI do the work. It requires manual upload/download of .md files between your 2nd Brain and NotebookLM, but it works well when the tasks require a lot of searching on the web and image analysis by the AI.

Choose a text editor

Since you and the AI mainly work with plain text files, you do not need anything other than the text editor that comes with your operating system, such as Notepad. Personally, I prefer Obsidian, which works with Markdown files.

A Markdown file, with the file extension .md, is an ordinary text document that uses simple special characters to format the text. The files are readable as they are and can be opened in any text editor. The AI is very compatible with Markdown files.

Examples

Example 1: Easy access to the knowledge in the books I have read

I really like keeping a reading diary, but I can't always remember every detail. That's where my Second Brain comes in. I use it to auto-index, tag, and categorize my reading list, transcribe key passages, and enrich my notes with web insights—building a comprehensive, searchable note for every single book.

The above may look fancy but it is still just text files—notes are .md files, databases are .yml, and settings are .json.

Example 2: Family meal planning

Here's an example of how to use the 2nd Brain to do meal planning and shopping lists based on what's at home, what we like, what we've eaten in previous weeks, and adapting to what else the family has going on during the week.

In the Source folder I have put:

• A Google Sheet containing dishes the family likes, grouped by protein. The ingredients are specified per dish, and each dish is categorized by whether it requires little, medium, or a lot of effort to cook. The document lives in Google Drive and is shared within the family, so anyone can add more dishes.
• A family schedule showing who has which activities on which evening.
• A list of staple items that should always be checked, regardless of the week's meal plan.
• A list of food items that are plentiful at home or need to be used up soon.

The instructions (Claude.md) describe:

A description of which sources should be used, how dishes should not be repeated, how they should be adapted to effort level and planned activities, which dishes should be prioritized, and how shopping lists should be written.

The memory (Memory.md) contains:

Details such as how many people are in the household, whether any family member is tired of a certain dish, and whether an ingredient should be dried or frozen.

When I ask the AI to suggest a meal plan, it puts together a proposal based on the above and saves it in the Output folder as an HTML file. Since the brain lives in Google Drive, I can easily pull up the meal plan and shopping list on my phone: