Teaching it your business

Your root file: methodology, voice, standards

Your root file is a plain text file called CLAUDE.md that lives at the top of your project folder. Claude Code reads it automatically at the start of every session. It is the single most important file in your entire system.

Open Claude Code in your project folder. You are going to create this file right now. Start with who you are: your name, your role, what you value, and how you want to be supported. Then add your business: what you sell, who you serve, your methodology or framework, your pricing.

Next, add your voice. How do you write? Are you formal, casual, direct, warm? What words do you never use? What tone do you want in emails, posts, and client communication? Then add your standards: what you never compromise on. Your boundaries, your ethics, your scheduling rules, your non-negotiables.

The file does not need to be perfect on day one. A good starting root file is about one to two pages. Over time, it grows to include daily protocols, file locations, and integration details. You will add to it every week as you discover what matters.

Here is the fastest way to start. Open Claude Code and paste this:

Then have a real conversation about it. Claude will ask you questions and draft sections. You edit, refine, and keep going.

When you are done, you will have a file that contains your identity, your business, your voice, and your boundaries. Every future session starts by reading this file. That means every conversation already knows who you are before you say a word.

Why it gets smarter the longer you use it

Every time you have a conversation with Claude Code, it can learn something about you. Your preferences, your clients, your schedule, the way you like things written. But here is something important to understand: Claude Code does not automatically remember between sessions.

When you close the terminal and open it again, it starts fresh. That is actually a feature, not a bug. It means your data stays private and nothing leaks between sessions unless you choose to save it.

So how does it remember? Through files. Your root file is read every session. Your fact files store permanent truths. Your session learnings capture what happened recently. The more files your system has, the more context Claude starts each session with.

None of this is magic. It is structured files that Claude reads at the right time. The more you use it, the more files it has to draw from. Everything you teach it compounds instead of disappearing.

Memory: how it remembers across sessions

The memory system has five tiers, from temporary to permanent. Your repository already has the folder structure. Here is what each tier does and where it lives:

Scratchpads are quick notes Claude writes during a session to hold complex thinking. You do not need to manage them. Session Learnings capture what Claude learned after each session: mistakes, things that worked, preferences discovered. Patterns emerge when the same learning shows up multiple times. Facts are permanent truths you approve before saving. History is one file per day, your immutable record of what happened.

The folders are already in your repository. The files will fill themselves as you use the system. But the intervals are yours to decide. How often should patterns be reviewed? Weekly? Monthly? How long should session learnings stay active before archiving? These are your choices.

Open Claude Code and ask it to walk you through the setup:

Claude will explain each tier and help you set intervals that match how you work. There is no single right answer. Someone who works daily might review patterns weekly. Someone who works a few times a week might prefer monthly. You decide, and you can change it anytime.