What Claude remembers
Every conversation with Claude starts fresh. Close it, open a new one, and Claude doesn’t remember the last one happened. That’s usually fine — but it means anything genuinely worth keeping would vanish unless you wrote it into a CLAUDE.md yourself.
So Claude has a way to hold onto the important stuff on its own.
Memory is a set of notes Claude keeps about you and your work that carry over from one conversation to the next — written by Claude, stored privately on your machine, and fully under your control.
What kinds of things land in memory
Not everything — just the durable, worth-remembering things. Over time Claude might note:
- About you — “prefers TypeScript,” “is new to the backend side of this repo,” “likes a plan before big changes.”
- About the work — “the team is mid-migration off the old auth system,” “deploys are frozen until next week.”
- Feedback you’ve given — “asked me to stop adding comments to every function.”
The throwaway details of a single task don’t get saved. The things that would still be true next month do.
You’re always in the driver’s seat
This is the part worth internalizing: memory isn’t a black box. It’s just files, and you can look at them whenever you want.
Run /memory and Claude shows you exactly what it’s holding onto. You can read it, fix anything that’s wrong or out of date, and delete anything you don’t want kept. Nothing is hidden, and nothing is permanent unless you leave it there.
Type
/memoryanytime to see, edit, or clear what Claude remembers about you.
And it stays on your machine — private to you, not shared with teammates and not the same as the project’s CLAUDE.md.
Why this matters
The first few times, memory feels invisible. Then one day you start a brand-new conversation and Claude already knows you prefer small commits, or that you asked it not to touch a certain folder — and you never said it this time. It carried over.
That’s the unlock: the relationship gets better the more you use it, without you having to re-explain who you are every morning.
What’s next
Now you’ve got two ways to give Claude lasting context: a CLAUDE.md you write, and a memory Claude keeps. They sound similar, and people mix them up constantly. The next lesson draws the line — when to write it down yourself, and when to let Claude remember.