Does Claude Code remember between sessions?

Built in

What Claude Code remembers out of the box

Claude Code has two kinds of memory, and both are plain markdown files on your machine. It can also reopen an old conversation, and it keeps some things through a long session’s summary.

1. CLAUDE.md: instructions you write

A CLAUDE.md file is where you write down what you’d otherwise re-explain: build and test commands, coding standards, how the project is laid out, “always do X” rules. Claude reads it at the start of every session. You can put one in four places, from broadest to most specific:

• Claude Code loads every CLAUDE.md from the folder you start in and every folder above it, and adds them together rather than letting one replace another. A CLAUDE.md in a subfolder loads when Claude reads files there.
• A CLAUDE.md can pull in other files with @path/to/file imports. Imported files still load at the start, so they help you organize but don’t save space.
• For bigger projects, files in .claude/rules/ can be scoped to paths like src/api/**/*.ts, so they load only when Claude works on matching files.
• Run /init and Claude writes a starting CLAUDE.md from your codebase. Anthropic suggests keeping each file under 200 lines: longer files use more context and are followed less reliably.
• If your repository uses AGENTS.md and has no CLAUDE.md, recent versions of Claude Code read the AGENTS.md instead.

2. Auto memory: notes Claude writes for itself

Auto memory is on by default in local sessions. As it works, Claude decides what would be useful next time and saves it as a short note. There are four kinds: who you are and how you like to work, corrections you gave it and approaches you confirmed, ongoing project work and decisions it can’t see in the code, and where to find things outside the project (an issue tracker, a dashboard). It skips anything it can work out from the code, and anything your CLAUDE.md already says.

• The notes live in ~/.claude/projects/<project>/memory/: a MEMORY.md index plus one file per note. All worktrees of the same git repository share one folder.
• At the start of every session Claude Code loads the first 200 lines or 25KB of MEMORY.md, whichever comes first. The notes themselves aren’t loaded; Claude opens them when it needs them.
• Saying “remember that the API tests need a local Redis” saves it to auto memory. When you see “Saved 2 memories” or “Recalled 2 memories”, that’s this folder being written or read.
• To turn it off, use the toggle in /memory, set autoMemoryEnabled to false in a project’s settings, or set CLAUDE_CODE_DISABLE_AUTO_MEMORY=1.

3. /memory: see and edit all of it

Type /memory in a session to list your CLAUDE.md files, open any of them in your editor, open the auto memory folder, or switch auto memory on or off. Everything is plain markdown you can read, edit or delete. To check which files actually loaded into the current session, run /context.

4. Resuming a conversation

Claude Code saves every conversation as you go, so you can come back to one. claude --continue reopens the most recent conversation in the current folder; claude --resume opens a list to pick from; /resume switches conversations from inside a session. A resumed session gets back the full history, including tool calls and results.

The transcripts are stored as JSONL files in ~/.claude/projects/<project>/ and deleted after 30 days by default (the cleanupPeriodDays setting). Auto memory files are left out of that cleanup.

5. What survives /compact

When a long session fills up, Claude Code summarizes the conversation to make room. After compaction, the project-root CLAUDE.md and auto memory are read again from disk. An instruction you only ever gave in the chat can get lost in the summary, which is a good reason to put it in CLAUDE.md.

Side by side

CLAUDE.md vs auto memory

In short: CLAUDE.md is what you tell Claude; auto memory is what Claude tells itself. The two work together, and Anthropic’s comparison comes down to this:

Neither is enforced. Claude treats both as context and tries to follow them, and vague or conflicting instructions get followed less reliably. For something that must happen every time, like a check before each commit, Anthropic points to hooks instead.

The edges

What Claude Code doesn’t remember

• Your earlier conversations, in a new session. A new session sees only what’s in CLAUDE.md and auto memory. The conversation itself comes back only if you resume it.
• Anything past the top of the index. Only the first 200 lines or 25KB of MEMORY.md load at the start. Claude Code reminds Claude to keep the index short, and anything past the limit is dropped from the next load.
• Anything on another computer. Auto memory is machine-local: it isn’t shared across machines or with cloud sessions. A project CLAUDE.md committed to git is the exception, because it travels with the repository.
• Old transcripts, after 30 days. By default, conversations older than that are deleted, so they can no longer be resumed.
• Nothing fades or grows on its own. Notes stay exactly as written until you or Claude edit or delete them. Keeping them current is up to Claude’s judgement and yours.
• It’s memory for the work. Auto memory’s four kinds of note are about you as someone to work with, the project, and where things are. It isn’t built to give Claude a memory of its own.

Add-ons

Memory plugins and MCP servers for Claude Code

Claude Code can connect to outside tools through MCP servers (added with claude mcp add), and plugins can bundle skills, hooks and MCP servers (installed with /plugin install). Memory systems use these to plug in. Here are the ones in our directory, each in a sentence or two, with what to keep in mind. Each one’s directory page goes further, including how it compares with the way human memory works.

Claude Code’s built-in memory has its own page in the directory too, and the comparison puts them all side by side.

By what you need

Which should you use?

Start with what’s built in. Most people asking how to give Claude Code memory need a good CLAUDE.md and auto memory left on.

• Claude should follow my project’s rules every time. Write a CLAUDE.md (start with /init) and commit it so your team gets it too. Put rules for one part of the code in .claude/rules/.
• I keep correcting the same thing. Leave auto memory on, or say “remember that …”. If it’s a rule, add it to CLAUDE.md.
• I want to carry on yesterday’s conversation. claude --continue, or claude --resume to pick one.
• I want notes on every past session, written for me. claude-mem does this automatically.
• I’m building an app or agent that needs memory of its users. These are developer building blocks: Mem0 plugs into almost any stack; Zep keeps track of when facts changed; Supermemory takes chats and files and keeps a profile of each person; Cognee builds a knowledge graph from documents and code and can run fully on your machine.
• I want a long-lived agent with a memory of its own. Letta, which is its own agent rather than an add-on to Claude Code.
• I want Claude to remember who it is with me. Counterparts, below.

FAQ

Questions people ask