Claude memory setup guide

Stop re-explaining yourself. Set up all 3 layers in 5 minutes.

by @cindiezhu

Part 1: The 3-layer setup (for everyone)

Layer 1: Chat memory (the automatic one)

What it does: Claude automatically summarises your conversations and builds a profile about you: your name, job, preferences, tools you use. This updates every 24 hours.

How to set it up:

1. Step 1 Go to Settings > Capabilities
2. Step 2 Toggle on "Generate memory from chat history"
3. Step 3 Toggle on "Search and reference past chats"
4. Step 4 Go to Settings > Memory to see what Claude already knows about you
5. Step 5 Review what's there, edit or delete anything that's wrong or outdated
6. Step 6 Tell Claude specific things to remember by saying "Please remember that I'm a [your role] working on [your project] and I prefer [your preferences]." It updates instantly, you don't have to wait for the 24-hour sync.

Layer 2: Chat search (find past conversations)

What it does: You can ask Claude to search through your entire chat history and pull relevant context into your current conversation.

How to use it:

1. Step 1 Make sure "Search and reference past chats" is toggled on in Settings > Capabilities
2. Step 2 In any new chat, just ask naturally: "What did we discuss about my website last week?" or "Find our conversation about the marketing strategy" or "What did you recommend for my email sequence?"
3. Step 3 Claude will search your past conversations and bring the relevant context into your current chat

Layer 3: Project memory (isolated context per project)

What it does: Each Claude Project gets its own separate memory space. Your work project doesn't bleed into your personal chats. This is the most powerful layer for people who use Claude for different purposes.

How to set it up:

1. Step 1 Go to Claude.ai and click "Projects" in the left sidebar
2. Step 2 Create a new project (e.g. "Content Strategy" or "My Business")
3. Step 3 Click on the project, then click the settings/gear icon
4. Step 4 Find "Project Instructions," this is your permanent briefing document
5. Step 5 Paste in your key context. Here's a template:

Who I am: [your name, role, what you do]
What I'm working on: [current projects, goals]
My preferences: [how you like responses, tone, format]
Key context: [anything Claude should always know when helping you]
Tools I use: [your tech stack, platforms]
What not to do: [things you don't want Claude to suggest]

• Step 6: Every conversation inside this project now starts with Claude already knowing all of this. You never have to re-explain it.

Part 2: What memory still misses (and how to fix it)

Claude's native memory is a profile, not a notebook. It captures your preferences and general facts, but it doesn't capture decisions you've made, reasoning chains, or detailed project context. Here's how to fill the gaps:

Fix 1: Tell Claude what to remember (in real time)

Don't wait for the 24-hour automatic sync. During any conversation, you can say:

• "Remember that we decided to use Webflow for the website"
• "Remember that my target audience is non-technical workers aged 22-40"
• "Remember that I prefer Averia Serif font for my brand"

Fix 2: Use project instructions as your permanent briefing

(Covered in Layer 3 above, but worth emphasising: this is the fix for the biggest gap in Claude's memory. Native memory is shallow and general. Project instructions are deep and specific.)

Fix 3: Import your ChatGPT memories

If you've been using ChatGPT and it already knows a lot about you, you can bring all that context over to Claude in about 60 seconds.

1. Step 1 Go to Settings > Memory
2. Step 2 Find "Import memories" option
3. Step 3 Select "Import from ChatGPT" (also supports Gemini and Grok)
4. Step 4 Follow the prompts, it will ask you to paste an export from ChatGPT
5. Step 5 Wait about 24 hours for processing
6. Step 6 Check Settings > Memory to see what was imported and edit anything that's wrong

Part 3: For power users (developer tools)

If you use Claude Code (the terminal coding tool), there's a whole ecosystem of memory plugins that go way deeper than the native web features.

claude-mem (46K+ GitHub Stars)

What it does: Gives Claude Code persistent memory across coding sessions. It automatically captures what Claude does during your sessions, compresses observations with AI, and injects relevant context into future sessions.

Why it matters: Cuts "rehashing context" by 80-90%. Instead of re-explaining your codebase every session, Claude Code remembers your project structure, recent decisions, and ongoing bugs.

How to install:

npx claude-mem install

Link: github.com/thedotmack/claude-mem

memsearch

What it does: A lightweight long-term memory library that stores daily session summaries as plain markdown files. Simpler than claude-mem: everything is human-readable and editable.

Why it matters: You can open any memory file in a text editor and see exactly what Claude remembers. Full transparency and control.

Link: Built by the Milvus team. Search "memsearch claude code plugin" on GitHub.

Claude Code's built-in auto memory

What it does: Claude Code now has native auto memory (since v2.1.59+). Claude saves notes for itself as it works: build commands, debugging insights, architecture notes, code style preferences.

How to check it: Type /memory in any Claude Code session to see what it's stored. Everything is plain markdown you can read, edit, or delete.

CLAUDE.md files

What it does: A markdown file you place in your project folder that Claude Code reads at the start of every session. It's like a permanent briefing document (similar to Project Instructions on claude.ai, but for the coding environment).

How to use it: Create a file called CLAUDE.md in your project root. Put your project context, coding conventions, preferences, and key decisions in it. Claude Code loads it automatically.

Quick reference checklist

5-minute setup (everyone):

• Toggle on memory in Settings > Capabilities
• Toggle on chat search in Settings > Capabilities
• Review and edit your memory in Settings > Memory
• Tell Claude 5 specific things to remember about you
• Create at least one Project with detailed instructions
• Import ChatGPT memories if you're switching over

Power user setup (developers):

• Install claude-mem: npx claude-mem install
• Check your auto memory: type /memory in Claude Code
• Create a CLAUDE.md file in your project root
• Review what auto memory has captured and edit as needed