Tech • AI • Robotics • Game
How to Build a Perfect Second Brain on Claude? (the simple system)
5/10TL;DR
A simple AI second brain built from well-organized text files, not complex databases, can give tools like Claude, Codex, and other agents persistent context, current project knowledge, and cleaner automation for non-technical users.
KEY POINTS
A reaction against overbuilt systems
Many non-technical users trying to build an AI workspace end up with overly complex setups stitched together from tutorials, often full of duplicated information and conflicting rules. The proposed alternative is deliberately minimal: a structured folder that stores core business and project knowledge in plain text, allowing AI to retrieve only what is needed instead of loading everything into each chat.
The core idea is a file-based RAG
The system uses retrieval-augmented generation in its simplest form: the AI searches files first, then answers from what it finds. Instead of relying on a vector database or custom infrastructure, the model scans folders for terms like price, rate, or euros, then reads the relevant file and responds with the most current information.
Why plain text beats a database here
Text files are readable by both humans and AI agents, easy to edit manually, and portable across platforms. A database often requires code, scripts, and server maintenance, while a file can be corrected by changing a single line. That also reduces lock-in: if a user switches from Claude to Codex or another agent, the same files still work.
A small folder structure does most of the work
The recommended structure is compact: a root brain folder, project folders, notes, and archives. Each project contains an about.md file that acts as an entry point, summarizing status, decisions, current context, and next steps. Rather than maintaining thousands of documents, the emphasis is on a few clear folders and a consistently updated overview file.
An agent file sets the rules for every chat
A separate agent.md file is read at the start of each new conversation. It contains essential background on the user, communication style, constraints, and operating rules. The goal is not to overload the model with everything known about the user, but to provide just enough stable context so the AI can ask fewer repetitive questions.
Decisions and reasons are treated as first-class data
The system stores not only what was decided, but why. That matters because AI tools often repeat past mistakes when they do not have access to the reasoning behind earlier choices. Keeping decisions, rationale, and next actions in the project file helps preserve continuity across sessions and improves consistency in outputs such as launch emails, plans, or client work.
Daily notes turn chat into memory
Notes and journals act as a running memory of weekly work, ideas, and to-dos. This allows the AI to answer practical questions such as what was done that week, what remains unfinished, or what ideas were captured during voice notes. The approach is also meant to protect against limited retention in chat tools, where conversations may only remain easily usable for about 30 days.
Tools should connect to live data, not duplicate it
The design argues strongly against copying data from systems like Gmail, Notion, Stripe, or scheduling tools into the brain. Instead, the AI should access those sources through connectors on demand. That keeps one source of truth in one place: if a price changes, the user edits one line, and the AI retrieves fresh operational data only when needed.
Skills replace repeated prompt copying
Reusable workflows are stored as skills, such as creating a project, cleaning up files, or managing a process. This is intended to replace the common habit of saving and re-pasting long prompts. By separating durable procedures from project knowledge, the setup stays cleaner and easier to scale.
The system also separates thinking from production
The brain folder is positioned as a workspace for reflection, business context, and project memory, not a dumping ground for every operational asset. A video editing workflow, for example, should live in its own folder with its own files and skills. Separating the AI brain from execution environments is presented as a way to improve retrieval quality and avoid clutter.
CONCLUSION
The model for an AI second brain is shifting away from elaborate stacks toward a lean file system built around agent.md, about.md, and live tool connections. The main stake is reliability: one clean source of truth that keeps AI useful across projects, sessions, and platforms.
Ask a question