Claude Memory: How to Make Claude Remember You Between Chats

Does Claude remember conversations?

By default, a fresh Claude conversation starts with a blank slate. Within a single chat, Claude remembers everything you've said. Open a new chat and that context is gone — Claude doesn't carry what you told it yesterday into today unless something explicitly feeds it back in.

Newer memory features and Projects narrow this gap by letting Claude reference earlier material, but they're built to recall facts from past chats, not to encode how you want Claude to behave. They remember that you mentioned a deadline; they don't reliably enforce "always answer in tight bullet points and never ask me clarifying questions." For consistent behavior across every conversation, you need a file that states it plainly.

The two kinds of memory

Recall memory — "What did we talk about?" Past facts, prior threads. This is what Claude's built-in memory and Projects address.
Operating memory — "Who am I and how should you work with me?" Your role, preferences, boundaries, standing instructions. This is what a CLAUDE.md provides — and it's the one that actually changes Claude's behavior every time.

Most people reach for recall memory when what they're missing is operating memory. You don't need Claude to remember a transcript; you need it to remember you.

How CLAUDE.md gives Claude persistent memory

A CLAUDE.md is a plain-text file Claude reads at the start of every conversation before it answers. Put your role, working style, and boundaries in it once, and Claude opens every chat already knowing them — no re-explaining, no copy-pasting context, no drift between sessions.

What belongs in it:

Who you are — your role, your domain, what you're working on right now.
How you want Claude to work — response length, format, tone, whether to ask before acting.
Boundaries — the "never do this without checking" rules.
Standing context — the projects, people, and shorthand Claude should already know.

Because Claude reads it every time, the file is the memory. Update it and the memory updates; nothing is forgotten between Tuesday and Wednesday.

Why most memory files quietly fail

Too long. Past roughly 2,000 tokens, adherence drops and the bottom half gets ignored.
Too vague. "Be helpful and professional" tells Claude nothing it didn't already assume.
No boundaries. The highest-leverage lines — what Claude should never do — are the ones people leave out.
Wrong location. It has to sit where Claude looks.

A memory file that hits these is worse than none, because you think Claude remembers when it doesn't.

Build memory Claude will actually use

The fastest way to get a memory file that works is to describe your role in plain English and let it be drafted and scored for you. CoworKit's free CLAUDE.md Builder turns a short description of who you are and how you work into an optimized CLAUDE.md, then scores it across clarity, specificity, format, voice, boundaries, and length. Already have a file? Paste it in and the Builder will optimize and trim it under the token ceiling so all of it actually gets read.

Bottom line

Claude doesn't forget you because it's limited — it forgets because nothing is feeding it your context each time. Built-in memory recalls past facts; a CLAUDE.md encodes how you want Claude to work and gets read at the start of every chat. Build and score one free.

FAQ

Does Claude remember previous conversations?

Not by default — a new chat starts fresh. Within one conversation Claude remembers everything; across conversations it doesn't, unless a memory feature or a CLAUDE.md feeds your context back in.

How do I make Claude remember who I am?

Put your role, working style, and boundaries in a CLAUDE.md file. Claude reads it at the start of every conversation.

What's the difference between Claude's memory feature and a CLAUDE.md?

Memory features recall past facts; a CLAUDE.md encodes how you want Claude to behave. The second reliably changes Claude's behavior every time.

Why does Claude ignore my memory file?

Usually because it's too long (over ~2,000 tokens), too vague, or stored where Claude doesn't look. Score it free to find the gaps.