Automatic memory: view, correct, turn off

This lesson is still being tested: it improves with learners' feedback. If something is unclear, it is the lesson that needs fixing, not you.

By the end of this lesson, you will be able to

say what Claude Code notes by itself, and where it stores it;
view and read its memory;
correct it or delete part of it;
turn it off for all your projects, or for just one.

Why it matters to you

Claude Code takes notes about you and your work, without you asking. This is handy: it remembers your preferences from one session to the next. But a wrong or overly personal note can stay for a long time, and influence every session. Knowing how to read and correct it means keeping control.

Before you start


You need	Claude Code in the app, in Manual mode (or in the terminal), and your mon-espace-ia folder. If /memory is not available in the app: the terminal option from module 4 (lesson 2)
Cost	The agent subscription, the same as in module 4: Claude Pro at minimum, €18 per month excl. VAT, price shown in France on 24/10/2026
Duration	30 minutes
Reminder	If a note contains something from the “never” list, delete it

Understand it in 3 minutes

What Claude notes, according to the Claude Code documentation (24/10/2026): four kinds of notes, about your role and your preferences, about the corrections you have given it, about the work in progress and its decisions, and about where to find information.

It does not note everything: it chooses what seems useful for a future conversation, and leaves out what your CLAUDE.md already says.

Where it stores it:

in a folder specific to each project, on your computer: ~/.claude/projects/<projet>/memory/ (~ means your personal folder, the one named after your user name, on Mac as on Windows);
a MEMORY.md file serves as an index, one line per note; each note is a small text file;
the first 200 lines of MEMORY.md (or its first 25 KB) are read at the start of each conversation; the detailed notes, only when Claude needs them;
this memory is stored on this machine (what Claude reads from it at the start of each conversation is still sent to Anthropic with your requests): it is not shared with another computer, and it is not in your mon-espace-ia folder, so not in your copies of that folder (a backup of the whole computer, such as Time Machine, does include it).

It is turned on by default. When Claude uses it, you may see messages such as “Saved 2 memories” (2 notes saved) or “Recalled 2 memories” (2 notes read back).

“Remember” or “add to CLAUDE.md”? When you tell Claude to remember something, it saves it in its automatic memory. For a rule to go into CLAUDE.md, the file you read and control, say so: “Add this to CLAUDE.md”.

Do it together

1Step 1: Open the memory

In a session on mon-espace-ia, type:

/memory

According to the documentation, /memory lists your CLAUDE.md files, offers a switch for automatic memory, and an option to open its folder.

If the app answers that the command is not available here, use the terminal (the optional setup from module 4, lesson 2): in your folder, run claude --permission-mode manual, then type /memory.

If you have not installed it, skip this step: you can still read the memory by asking Claude (step 2).

2Step 2: Read what Claude has noted

Open the automatic memory folder, then MEMORY.md. You can also ask:

Show me what your automatic memory contains about me and about this project, without changing anything.

For each note: accurate? useful? nothing sensitive?

3Step 3: Correct or delete

The notes are simple text files: you can edit or delete them yourself. Or ask:

In your automatic memory, correct the note about [the subject]: [the right information]. Show me the change before writing it.

A note you want to be applied for sure? Move it into CLAUDE.md: “Add this to CLAUDE.md, then remove it from your automatic memory.”

4Step 4: Choose your rule

Your choice	How
Keep it, and read it again once a month	Nothing to change; note the review date in your journal
Turn it off for all your projects	The `/memory` switch: according to the documentation, it saves this choice in your personal settings (`~/.claude/settings.json`), which apply to all your sessions
Turn it off for a single project	This is possible with a project setting, `"autoMemoryEnabled": false`, in its hidden file `.claude/settings.json`. With the module 4 kit, this file already exists and sets Manual mode. Do not erase anything: the file must become `{ "permissions": { "defaultMode": "default" }, "autoMemoryEnabled": false }`, with a comma after the `permissions` block. If you have Claude do it, read the result before approving, and approve this once only, not for the whole session. Claude cannot change it without a separate permission: it is a protected settings file. In the app, you cannot edit it yourself in the file pane: the documentation advises opening it in another application. The simplest way at the start: turn off memory with the `/memory` switch for the duration of this project, then turn it back on

For a very personal project (family, confidential work), turning off memory limits what Claude keeps by itself.

But this changes nothing about what is sent to it: everything you write and everything it reads still goes to Anthropic, and the conversation is also recorded on your computer. Your “never” list (health, passwords…) therefore remains valid, whatever this setting.

Your turn

Read your automatic memory again, note by note.
Correct or delete at least what is wrong or too personal.
Choose your rule (keep and re-read, turn off everywhere, turn off for one project), and note it in your journal with its reason.

Check your work

It worked if:

you opened /memory (in the app or the terminal) and found the automatic memory folder, or, failing that, asked Claude to show you what its memory contains;
you read MEMORY.md and each note it lists;
no wrong or sensitive note remains;
your rule is chosen and noted in your journal.

Common mistakes:

thinking the memory is in your folder, and so in your copies: it is not there;
saying “remember” for an important rule: write it in CLAUDE.md instead;
turning off memory “for all projects” while thinking you are turning off only this one;
never reading it again: an old note can contradict your current situation.

Keep it and pick it up later

Note in your journal the date of your last review of the memory.
Once a month: /memory, read MEMORY.md, make corrections.

Going further

optional

The official documentation, in English: code.claude.com/docs/en/memory (the “Auto memory” section). It also describes an autoMemoryDirectory setting to store the memory elsewhere, and the CLAUDE_CODE_DISABLE_AUTO_MEMORY=1 variable to turn it off at launch.
Chat assistants (claude.ai, ChatGPT) have their own memory, with their own settings: module 2, lesson 5.

New words

4 words explained

Word	What it means
Automatic memory	The notes Claude Code takes by itself, from one session to the next
`MEMORY.md`	The index of automatic memory, one line per note, read at the start of each conversation
Personal settings	The `~/.claude/settings.json` file, which applies to all your sessions on this computer
Project settings	A folder’s `.claude/settings.json` file, which applies only to that folder

To be checked: /memory in the desktop app (screenshot to be taken).

Date, version, sources Written on 24/10/2026. Checks of 24/10/2026 on code.claude.com/docs/en/memory: the four kinds of notes, what Claude does not note, being on by default, the per-project and single-machine location, MEMORY.md (200 lines or 25 KB read at startup), the “Saved” and “Recalled” messages, the /memory command and its switch saved in ~/.claude/settings.json, turning it off for one project with autoMemoryEnabled, “remember” versus “add to CLAUDE.md”. .claude/settings.json file: in the app, it is read-only in the file pane (code.claude.com/docs/en/desktop); Claude asks for a separate permission to change it (test of 24/10/2026). Claude Pro price: claude.com/pricing.