👇 After this module

You understand what gets saved and where, what Claude can reach, how memory tiers work, how to teach your system every day, and how to feed it your real network.

What actually gets saved

Before we talk about memory, let's clear up something important. Your conversations with Claude are not lost when you close a session. Every conversation is saved as a file on your computer. You can go back to any past conversation, pick up where you left off, and keep working.

Conversations are saved locally

Every session you have with Claude Code gets saved as a file on your Mac. These files live in a hidden folder and they stay there indefinitely. You can browse past conversations in the sidebar, search through them, or resume any of them at any time.

To go back to a previous conversation, type /resume and pick from the list. Or type /continue to jump straight back to your most recent one. The full conversation loads back in, exactly as you left it.

Where conversations live: app vs terminal

If you use Claude on the web (claude.ai), your conversations are stored on Anthropic's servers and appear in a sidebar list. You can scroll through them, search them, and pick up where you left off.

In Claude Code (the terminal), your conversations are saved as files on your computer in a hidden folder called ~/.claude/projects/. Each session becomes a .jsonl file. They stay on your machine indefinitely. You can resume any of them with /resume or jump to your most recent with /continue.

In both cases, the conversations are processed through Anthropic's API. The difference is where the transcript file lives after the session ends.

But new sessions start fresh

Here is the important part. When you start a brand new session (not resuming an old one), Claude does not automatically know what happened in previous conversations. It does not read your old transcripts. It starts with a blank slate.

What Claude reads at the start of every new session falls into two groups: things Claude does natively (built into Claude Code itself), and things your Four Languages root system adds on top.

Claude native (built-in to Claude Code)

Those three files auto-load every single session without any action from you. Claude Code reads them before you type your first message. That is the entire automatic load.

Four Languages system (loaded by your morning protocol, not by Claude itself)

The morning protocol is the bridge between Claude Code's small native auto-load and your full Four Languages memory. If you skip /morning and just start typing, Claude only has CLAUDE.md, MEMORY.md, and rules to work from. That is why /morning matters: it pulls in the recent context from your last few sessions so Claude walks in knowing where you left off.

Zero conversation history loads from prior sessions on its own. Skills, content folders, and history all load on demand when the task requires them.

The difference that matters

Think of it this way. Your conversation transcripts are a tape recording of a meeting. Saved on your machine, playable anytime. Your root system files (CLAUDE.md, memory, fact files) are the meeting notes you bring to the next meeting. The tape recording exists, but Claude only reads the notes.

This is why your root system matters so much. Without it, you could resume old conversations, but every new session would start from zero. With it, Claude walks into every session already knowing who you are, what you are working on, and how you like things done.

What you build over time

Your root system does not get smarter on its own. It gets better because you teach it how you work. Every file you add, every correction you make, every preference you save is a deliberate choice. Here is what that looks like when you actively build it:

This is the difference between using an AI tool and building a system. The tool guesses. Your system knows, because you taught it.

Turn off training on your conversations

By default, Anthropic may use your conversations to improve future Claude models. You can turn this off. We recommend you do this before you start building your root system.

The setting applies to both claude.ai and Claude Code since they share the same account. How Anthropic handles your conversations depends on your plan, this setting, and their current policy: Anthropic's privacy policy is the place to check.

The toggle has limits worth knowing. It is a setting you can flip back at any time, so treat it as a preference rather than legal protection. And on a personal plan it comes with no data-processing agreement, so if you handle client personal data, the terms you run Claude under matter more than this switch. The Privacy & GDPR module in Get Set Up covers what that means for your business.

This is your first infrastructure decision. It takes 30 seconds and it matters.

Mark complete Next Lesson

What Claude can reach

The last lesson covered what gets saved. This one covers the other direction: what Claude can reach while you work. The two together are the full picture of your system's footprint.

The boundary is the folder you start in

When you start Claude Code inside your project folder, everything in that folder is within reach. Files outside it need your permission first, one request at a time. That is why you start Claude inside your system folder rather than your whole Documents folder: the folder you choose is the boundary.

Reach matters even for files Claude never opens. Under GDPR, having access already counts as processing, whether Claude reads the file or not. The Privacy & GDPR page in Get Set Up covers the legal side in full.

The boundary resets every time you start

Close the terminal and the session's access ends with it. Start Claude in a smaller folder tomorrow and yesterday's wider reach is gone. One-time "yes" answers to permission requests end with the session too. If you ever started Claude somewhere too broad, the fix begins at the very next launch.

The exception: saved permissions

Some permission answers stick. When Claude asks to do something and you choose a "don't ask again" option, that choice is written into a settings file inside your project folder, and it keeps working in every future session until you remove it. These rules accumulate quietly. A working system can gather hundreds of them in a few months without you noticing.

Integrations are reach too

Anything you connect to your system, your inbox, your calendar, your database, extends what Claude can reach. Set each connection to pull data only when you trigger it, and to ask before each call. You decide what is within reach, tool by tool, and everything else stays out.

This is the same kind of decision as the training toggle in the last lesson: a few minutes now, structural later. Decide what Claude can reach, audit it now and then, and let the boundary do the quiet work.

Mark complete Next Lesson

Your memory system

Your root system organizes memory into layers. Each layer has a different purpose and a different writer. Together, they make sure nothing important is lost while keeping your root system clean and current.

The working layers (plus history)

Scratchpads are working notes Claude writes during a session to hold multi-step thinking, build plans, or drafts-before-doing. You do not need to manage them. They live in context/pad/ until you run /review, where you decide together what to keep and what to delete.

Patterns are observations about you. Your wording, choices, emotional tells, repeated phrases, energy shifts. Claude appends one to patterns.md at every save. If nothing came up, it skips silently. When you run /review, Claude mirrors the patterns back to you in neutral observation format. You confirm, dismiss, or promote them to a structural fact. No opinions, no judgement, no coaching. You decide what is real.

Session Learnings are operational lessons. Mistakes that happened, things that worked, decisions that were made. Claude appends to learnings.md at each save. When you run /review, Claude shows you accumulated entries and you decide together which to keep, promote, or delete. Nothing gets deleted without your explicit approval. The save receipt flags when the file is getting large.

Facts are structural truths. Your pricing, your methodology, your preferences, your boundaries. They live in context/memory/fact/ as YAML files. Read on demand by skills and by the strategic thinking trigger. Audited when you run /review, or when Claude reads a fact file during a task and flags drift.

History is one file per day. Your immutable record of what happened. Never deleted. Never edited after the fact.

How cleanup actually works

Nothing expires automatically. Claude has no internal clock or timer. It cannot track "this file is N days old" on its own. What keeps the system clean is two commands working together: /save-session and /review.

• Every save: Claude appends to patterns.md and learnings.md. This happens automatically as part of /save-session.
• When you choose to clean up: Run /review. Claude walks through your patterns (confirm, dismiss, promote), your learnings (keep, promote, delete), your scratchpads (keep or delete), and your facts (still true?). You decide the rhythm. The save receipt flags when files are getting heavy if you want a signal.

Your only job when you run /review is to answer in plain language: keep, promote, delete, still true.

Mark complete Next Lesson

Teaching it more every day

Every interaction teaches your root system something new. Not because it is recording everything in some hidden database, but because the files in your root system grow and refine with use. The question is not whether it learns. It is how fast.

Daily check-ins

The simplest way to teach your root system is to start each session by telling it how you are. Energy level, what you are working on, what is on your mind. These get logged, and over time your root system builds a picture of your rhythms. You will set these up properly in the next module.

Voice refinement

When Claude writes something that sounds off, tell it. Be specific. The more direct you are, the faster your voice guide sharpens.

Every correction gets saved. After a week of daily use, the difference is real. After a month, Claude writes in your voice without being asked.

Saving your sessions

At the end of a work session, you want to save what happened so your root system remembers it next time. Just say "save" or type /save-session. Claude runs a multi-step save process. Every step is verified. Every failure is visible.

When the save finishes, Claude prints a receipt showing every step as confirmed, skipped, or FAILED. Always check it. If any step says FAILED, tell Claude to retry before you close the session.

The receipt is not just a summary. Asking Claude to account for every step is what keeps it honest: when it has to show its work, it stops quietly skipping a step it would otherwise drop without telling you. That silent skip is the failure mode the receipt is built to catch.

Your evening protocol (which you set up in Module 3) triggers this full save process automatically. But you can also save mid-session whenever you have done something worth keeping. The word "save" is the trigger. Say it, and the full process runs.

The receipt tells you exactly what landed, what was skipped, and whether your memory files are getting large. When file health shows warnings, that is your natural trigger to run /review and clean up. No scheduled days, no gates to remember.

Reflective practice

Your root system can support whatever grounds you. Journaling, meditation, gratitude practice, prayer, time in nature. You choose what matters. Tell your root system about your practice, and it will prompt you, log your reflections, and track your consistency.

The files your root system writes are stored locally on your computer. But anything you type in a conversation with Claude goes through Anthropic's servers for processing, the same as any other message. If your reflective practice is deeply personal and you want it fully private, keep it in a file outside the folder you start Claude in, because whatever Claude can reach can travel to Anthropic the moment it reads it. Your root system can still track consistency and prompt you without reading the content.

Mark complete Next Lesson

Your fact files

Inside your memory system, there is a folder called context/memory/fact/. This is where permanent truths live. Things that do not change session to session. Your name. Your boundaries. How you prefer to work. What you have decided and why.

Your starter kit comes with seven fact file templates, each in YAML format. YAML is just a simple way to store structured information that both you and Claude can read. You do not need to learn any syntax. You fill them in by talking to Claude, and it writes the files for you.

The seven fact files

How to fill them in

You do not need to write YAML yourself. Just talk to Claude. Tell it about your business, your preferences, your boundaries. It will organize everything into the right files.

Claude will interview you, one file at a time. It asks questions, you answer in plain language, and it writes clean YAML. You review it, approve it, and move to the next file. Most people finish all six in one session.

When fact files actually load

Important: fact files are not loaded into every session automatically. Only your CLAUDE.md and your rules in .claude/rules/ auto-load every session. Fact files are loaded on demand: when a skill references one, when your morning protocol asks for current state, or when Claude hits the strategic thinking trigger (any "what should I do about" question). This is deliberate. It keeps your context window light and lets fact files grow detailed without slowing every conversation down.

Fact files vs skills: know the difference

Fact files hold your permanent baseline. Things that are true regardless of what you are working on today. Skills hold the detailed instructions for specific tasks. When you run a skill like /content-calendar or /crm, that skill file loads with all of its specific rules and templates. When the task is done, that context goes away.

This means your fact files do not need to contain everything. If your brand guidelines only matter when you are creating content, those details belong in a content skill, not in your fact files. If your CRM rules only matter when you are logging contacts, those belong in your CRM skill. Fact files carry your baseline truth. Skills carry the operating manual for specific work.

The golden rule: if it does not describe WHO you are or WHAT you have decided, it probably does not belong in a fact file

A good test: if a skill could do its job without reading this line, it probably belongs in the skill instead. Your identity? Fact file. Your brand color hex codes? Skill. Your pricing decisions and the reasoning behind them? Fact file. Your carousel slide template? Skill.

Five lines. That is enough for Claude to know who it is working with. Everything else loads when it is needed.

Mark complete Next Lesson

Organize your knowledge: the map and the terrain

In Module 1, you brought your existing knowledge into your system. Files, frameworks, client data, notes. Now it is time to organize WHERE everything lives so your root system knows how to find it and when to load it.

This is the difference between having a pile of useful files and having a system that routes to the right information at the right time.

The map vs the terrain

Your root file (CLAUDE.md) is the map. It does not hold all the details. It tells Claude where the details live and when to load them. The terrain is everything else: your fact files, your skills, your rules, your content, your data.

When Claude opens a new session, it reads the map first. From the map, it decides what to load based on what you are doing. If you are writing content, the map points to your voice guide and content skills. If you are managing contacts, the map points to your CRM skill and contact data. The map stays small. The terrain can grow endlessly.

Where each type of knowledge lives

Here is a concrete guide for deciding where to put things. Go through the files you imported in Module 1 and ask: what type of knowledge is this?

The sorting exercise

Ask Claude to help you audit where your imported files ended up and whether they are in the right place. This is the prompt that makes the difference:

Common mistakes

Mark complete Next Lesson

Teaching it your network

The fastest way to teach your system is to give it real data. Your LinkedIn connections are one of the richest sources you already have. Hundreds or thousands of people you have met, worked with, or connected with over the years. Your root system can read through all of them and tell you who matters most for where you are going.

Export your LinkedIn connections

Go to LinkedIn Settings > Data Privacy > Get a copy of your data > Connections. LinkedIn will email you a CSV file (usually within 10 minutes). Download it and drop it into your project folder.

Let your system classify them

Run /import-linkedin and your root system will read through every connection. It uses the identity and business fact files you filled in during onboarding (Module 1) to figure out who in your network looks like the kind of person you want to work with. Your dream client profile, your industry, your offer. All of that comes from your own files, not from a generic template.

Analyze your network

Then run /network-analysis to get strategic recommendations. Who should you reconnect with? Where are the clusters of aligned people? What patterns show up in your network that you have never noticed?

Go deeper on your best matches

For your strongest matches, run /lead-enrichment to pull more context before you reach out. This gives you something real to say when you reconnect, not a generic "hope you are well" message.

See your data in the dashboard

Your dashboard is ready to show your imported network. Two ways to open it:

Click the Network tab to see your connections classified by industry and ICP fit.

Mark complete