How to Make Claude Remember Your Work Across Sessions

I used to spend the first 5 minutes of every Claude conversation explaining who I am and what my project does. Now I spend zero. Here's everything I set up.

What You'll Learn

• The four places Claude stores "memory" and which one matters most
• How to set up Chat Memory so Claude learns automatically
• How to import ChatGPT preferences into Claude in 60 seconds
• The one file that gives Claude Code persistent project context
• What Claude actually remembers vs. what it forgets

Layer 1: Chat Memory (Everyone Should Turn This On)

Go to Settings → Capabilities → Memory and toggle it on. Once enabled, Claude processes your conversations roughly every 24 hours and pulls out things worth remembering: your job, preferred language, tools you use, writing style.

It's extractive summarization — Claude decides what's "worth keeping" and stores a compact profile that loads into every future conversation.

What do you know about me from our past conversations? List everything.

💡 You can actively tell Claude to remember things. Say "Please remember that I prefer tables over bullet points" and it adds that to your memory profile.

⚠️ Memory updates aren't instant. After a conversation, it can take up to 24 hours for new memories to appear.

Layer 2: User Preferences (Your Global Rulebook)

Go to Settings → Profile. There's a text box for instructions that apply to every conversation. Instead of staring at an empty text box, paste this into a Claude chat:

I want to set up my Claude preferences. Ask me 5-8 questions about how I work and how I like AI responses. After I answer, generate a clean preferences block I can paste into Settings.

Claude interviews you, then produces something ready to copy-paste.

Layer 3: CLAUDE.md (For Claude Code Users)

This is the highest-impact thing you can do. CLAUDE.md is a plain text file in your project root that Claude reads at the start of every session.

What to put in it: your tech stack, build/test commands, naming conventions, architecture decisions, and common gotchas.

💡 The compounding effect: Every time Claude makes a mistake, ask — "Should this be in my CLAUDE.md?" If yes, add it. Over weeks, it becomes a precision instrument.

⚠️ Keep it under 100 lines. A bloated file wastes context tokens and dilutes important instructions.

Layer 4: Auto Memory (Claude Code 2.1.59+)

Claude Code now writes its own notes as it works — build commands, debugging insights, architecture observations. It's on by default. Type /memory in a Claude Code session to check what it's learned.

To guide it: say "Remember: always use bun instead of npm in this project" and Claude saves it.

Migrating from ChatGPT

1. Open ChatGPT and paste: "List every memory you have about me, formatted as structured sections."
2. Copy the output
3. Go to Claude's memory import page and paste

Claude processes it within about 24 hours.

What Claude Cannot Remember

❌ Anything from Incognito conversations

❌ Real-time session details — memory is synthesized after conversations

❌ Cross-session state in Cowork — each task starts fresh

❌ Sensitive data — passwords, SSNs, financial details are excluded

Getting Started

Do these three things today:

1. Turn on Chat Memory: Settings → Capabilities → Memory → ON
2. Write your preferences: Settings → Profile → add 5-10 lines
3. If you use Claude Code: create a CLAUDE.md in your project root

For details: Chat Memory docs | Claude Code Memory