Articles

Claude memory explained: what each kind remembers and who can read it

Q: Where does Claude Code store its memory?

Auto memory lives in ~/.claude/projects/ /memory/ , with a MEMORY.md index, on your own computer. Your CLAUDE.md files live in the project, in your home directory, or at the organization level.

By Matthew Firth, founder of ModelBrain · Sep 14, 2026 · Updated Oct 16, 2026 · 6 min read

Claude does not have one memory. It has four, and they differ in ways that only show up when you switch app, project or computer. Here is our own proposal, not an industry standard, for comparing them: the reach test. For any memory store, ask who writes it, how far it reaches (one chat, one project, one machine or one account), and which other app can read it.

Four dark glass domes of increasing size, each holding one glowing amber capsule, the smallest on the left and the largest on the right

The four stores, checked October 16, 2026

Claude's memory stores by who writes them, how far they reach and which other apps can read them. From the vendors' own pages, checked October 16, 2026. The last row is ours.
Store	Who writes it	Reach	Other apps can read it	Source
Chat memory (and Projects)	Claude saves topics on its own; you can ask it to remember something	Your account. Each Project keeps its own memory, and Claude uses a Project's memory plus your non-project account memory, not other Projects' memory	No direct sharing. An experimental import and export feature moves memories between assistants	Claude Help Center
CLAUDE.md	You	Project, your user account, or your organization. Loaded every session	Indirectly. Claude Code can read or import `AGENTS.md`, which Cursor's rules page lists. That page does not mention `CLAUDE.md`	Claude Code docs, Cursor docs
Auto memory (Claude Code)	Claude	One git repository, shared across its worktrees, on one machine. The first 200 lines or 25 KB of the `MEMORY.md` index load at session start	No. Files are not shared across machines or cloud environments	Claude Code docs
An MCP memory layer such as ModelBrain	You, or the assistant through tools	Everything you connect to it, not one app or one repository	Any assistant you connect. ModelBrain works with Claude Desktop, Claude Code, Cursor and ChatGPT	How it works

What the table shows

Only instruction files travel. The files you write yourself are the part that other tools can read. The notes Claude writes about you stay inside Claude: chat memory belongs to your account, and Claude Code's auto memory sits in a folder on one computer.

Chat memory and Claude Code memory are separate. The Help Center page describes memory inside Claude's apps and does not say Claude Code reads it. The Claude Code page describes only its own files. Neither page says the other store feeds it. If you told Claude in the app that you prefer tabs over spaces, do not assume Claude Code knows.

Auto memory does not follow you to a second computer. The docs say it is machine-local. A new laptop starts with no auto memory unless you copy the files yourself, and the docs describe no sync.

Deleting a chat does not delete what Claude learned from it. The Help Center says memory entries made from a conversation stay until you delete them under Settings, then Memory.

Where each one stops in practice

Switching to another assistant. Chat memory and auto memory do not cross. A CLAUDE.md can, but only if the other tool reads it or reads a file it imports. We tested the Claude Code half of that in a short page on AGENTS.md.

Switching projects. Claude Code scopes auto memory to the repository, and chat memory scopes a Project's notes to that Project. A fact you want in both has to be written twice, or live somewhere that does not depend on the project.

Switching computers. Your account follows you, so chat memory does. A CLAUDE.md in git follows the repository. Auto memory does not.

Where a memory layer fits, and where it does not

A memory layer is a separate store that sits outside any one assistant. ModelBrain keeps memories on your own computer and any assistant you connect can recall them. It does not read or replace Claude's own memory files, and it adds one more thing to maintain.

If everything you want remembered was said inside Claude, and you are happy for Claude to be the only place it can be read, you do not need one. A layer starts to earn its place when the same facts matter in more than one app, or when you want to open the files and see what is there. Plenty of people reasonably use both: Claude's own memory for how a conversation flows, and a separate record for what they would be upset to lose. How we tag where each memory came from is in how ModelBrain tags what it remembers.

Sources

Use Claude's chat search and memory to build on previous context, Claude Help Center
How Claude remembers your project, Claude Code docs
Rules, Cursor docs

Common questions

Does Claude remember previous conversations? Yes, when memory is on. Claude saves topics about you and your work and uses them in later chats. It is on by default for Free, Pro and Max, and Team and Enterprise owners must enable it. You can view, edit or delete topics under Settings, then Memory.

Where does Claude Code store its memory? Auto memory lives in ~/.claude/projects/<project>/memory/, with a MEMORY.md index, on your own computer. Your CLAUDE.md files live in the project, in your home directory, or at the organization level.

Can Claude Code and Claude chat share memory? The vendor pages do not describe any sharing between Claude Code and the Claude apps. Treat them as separate stores, and put anything both must know in a file or a separate memory layer that both can read.

Last checked October 16, 2026. Settings and tool behavior change often, so confirm a detail on the vendor's own page before relying on it.