Skip to content
Claude Memory Explained: How It Works in the Claude App and in Claude Code
← All postsAI Tools

Claude Memory Explained: How It Works in the Claude App and in Claude Code

November 17, 2026·FixMyPrompt Team·7 min read
Share

Claude has two different memories. The Claude app remembers you across chats, and Claude Code remembers your project through CLAUDE.md and auto memory. How each one works, how to turn it on or off, and what neither one keeps.

#claude memory#claude ai memory#claude code memory#claude.md#auto memory#ai memory

"Claude memory" means two different things depending on where you use Claude. In the Claude app on the web, desktop or phone, memory is what Claude learns about you from your chats. In Claude Code, memory is a set of files on your computer that tell Claude about your project. They work differently, they are stored in different places, and one does not read the other.

This post covers both, using Anthropic's own documentation, and then where a tool like ours fits.

Memory in the Claude app

Is Claude memory on by default?

Yes, for individual plans. Anthropic's help center says memory is on by default for Free, Pro, Max and Ultra plans on the web, Claude Desktop and Claude Mobile. On Team and Enterprise plans it depends on when the organization was created. In most organizations created on or after October 31, 2026, memory starts on for each member. In older ones it stays off until each person turns it on.

What Claude remembers

Claude now adds things to memory while you chat, rather than summarizing a conversation after it ends. Anthropic's example is a project deadline: mention that it moved to September, and the next conversation already knows. Memory leans toward work context such as your role, your projects and how you like answers written.

Some topics are left out on purpose. Claude does not save subjects Anthropic treats as sensitive, such as health or beliefs, unless you turn that on in Memory settings. When it is on, you see a notice each time Claude saves something on one of those topics.

If you use Projects, each project gets its own memory space, so a client project and your personal chats do not bleed into each other.

Memory and chat search are separate

Claude can also search your past conversations when you ask it to, and it shows citations that link back to the original chats. That is a lookup on demand. Memory is what Claude carries into a new chat without being asked.

How to see, edit or turn off Claude's memory

  • See and edit it. In Memory settings you can view what Claude remembers topic by topic, and edit or delete any of it. You can also just tell Claude in a chat what to remember or forget.
  • Turn it off for your account. Go to Settings, then Your Claude (Settings, then Memory in the older layout), and switch off "Generate memory from chats."
  • Turn it off for one chat. Open a new chat, click the "+" menu and turn off Memory. That chat will not use or add to your memory, and your other chats are not affected. Incognito chats also stay out of memory and history.
  • Move it. Anthropic says you can transfer memory between Claude and other AI services, which helps if you are coming from ChatGPT.

The same memory is shared with Claude Cowork when Cowork runs in the cloud. Cowork sessions that run locally on your computer do not use it.

Memory in Claude Code

Claude Code is the coding agent that runs in your terminal or editor, and its memory is built for a codebase rather than for you as a person. The Claude Code memory docs describe two systems that work together. Both are plain files on your machine.

CLAUDE.md: instructions you write

A CLAUDE.md file holds the instructions you want Claude to read at the start of every session, like how to run the tests or which folders not to touch. Claude Code looks for them in a few places, from broadest to most specific:

  • a managed policy file set by your company, if there is one
  • ~/.claude/CLAUDE.md for your own preferences across every project
  • ./CLAUDE.md or ./.claude/CLAUDE.md in the project, usually committed so the team shares it
  • ./CLAUDE.local.md for your personal notes on one project

If a project has no CLAUDE.md, Claude Code reads AGENTS.md instead. A CLAUDE.md can pull in other files with an @path import, such as @AGENTS.md on its own line. Anthropic suggests keeping each file under 200 lines, because a longer file costs more context and gets followed less reliably. Our study of 1,431 real CLAUDE.md and AGENTS.md files found 28% of them run past that.

Auto memory: notes Claude writes itself

Auto memory is Claude Code saving its own notes as it works with you. Each project gets a folder at ~/.claude/projects/<project>/memory/ with a MEMORY.md index and separate topic files. The notes come in four kinds: facts about you, corrections and approaches you confirmed, ongoing project decisions that are not in the code, and pointers to outside resources such as a dashboard or issue tracker.

At the start of a session Claude Code loads the first 200 lines of MEMORY.md, or the first 25KB, whichever comes first. Auto memory is on by default in local sessions. To turn it off, open /memory in a session and use the toggle, set autoMemoryEnabled to false in settings, or set CLAUDE_CODE_DISABLE_AUTO_MEMORY=1.

For a longer walkthrough, including what goes in each file, see our post on Claude Code memory.

Does Claude Code use your Claude app memory?

No. The Claude Code docs describe only CLAUDE.md files and auto memory, and both live on your computer. What the Claude app has learned about you in chat is not part of either. If you use both, you will explain some things twice.

What neither one keeps well

Both kinds of memory are good at preferences and background. Neither is built to keep a procedure: the exact steps, commands and checks that finally fixed a deploy after an hour of trial and error. The app's memory is a summary of you. Auto memory is a short index that gets cut off at 200 lines. A CLAUDE.md that collects every fix soon runs past the size where Claude follows it well.

They also stay where they were made. Auto memory belongs to one project on one computer. A teammate's Claude does not see it, and neither does Codex or Cursor if you use those too.

Where FixMyPrompt memories fit

This is the gap we built FixMyPrompt for, so weigh that when you read this part. When your AI finally works something out, you say "save how you fixed that to memory," and it stores the steps, the commands, the check that proves it worked and the mistakes to avoid. A secret check runs first, so keys and passwords are caught before anything is saved. Memories are filed under a project named after the folder you are in.

Next time, you ask in plain words, such as "use fmp to find how we deploy staging," and your AI reads back the memory that fits and follows it. On Pro the same memories work in Claude Code, Codex and Cursor, and on the Biz plan a teammate's AI can find what you saved. The honest limit is that your AI does not look memories up on its own yet. You ask, or run one as a slash command.

It works alongside Claude's own memory rather than replacing it. Keep using CLAUDE.md for the rules every session needs, and let the app remember who you are. Use a FixMyPrompt memory for the hard-won fix you never want to work out again. The free plan includes 25 memories, and setup takes one command.

Try it on the prompt that sent you here

Paste the prompt that misbehaved and see what is weak in it.

The weak spots and fixes are tuned to this goal.

0 of 5,000 characters

No prompt handy? Try one:

Found this useful? Share it.

Share

Read next

Run a free QA on your own prompt

Get a structured score, specific issues, and a rewritten prompt in seconds.

Run free QA