Giving Claude a Memory That Actually Remembers You

Every new Claude Code session used to start the same way: re-explaining what I was working on, how I like feedback, what I'd already tried. None of it was hard to repeat — it just meant the first five minutes of every session went into rebuilding context instead of using it. This is the fix I've been running for a while now, and the exact steps to set up your own copy — written so a non-technical friend could follow it too.

Credit where it's due before I get into it: Kiyoraka built and open-sourced the whole thing. I didn't design this system, I just use it daily and wanted to make the setup easier to get into. Genuine thanks for putting it out there for free.

What this actually is

It's called AI MemoryCore, and there's no database or plugin underneath it — just three plain text files Claude already knows how to read. Think of it like a computer booting up: one file is firmware (barely changes, just tells Claude what to load), one is long-term memory (who you are, how you work, grows slowly over time), and one is RAM (whatever you're doing right now, cleared back to a short recap once the session ends).

Nothing about it is clever engineering — that's sort of the point. You can open every file it uses in Notepad and read exactly what it's doing.

What you'll need

• Claude Code already installed on your computer
• About 10 minutes
• No coding experience — every step below is copy, click, or type-and-answer

Get the files

Open the GitHub page in your browser, click the green Code button, then Download ZIP. No account or git knowledge needed. Unzip it — right-click and choose Extract All on Windows, double-click on a Mac — and move the folder somewhere easy to find, like Documents. Rename it to ai-memorycore if you like.

Open Claude Code right inside that folder

A terminal is just a plain window where you type one line instead of clicking. On Windows, open the folder in File Explorer, click into the address bar, and type cmd. On a Mac, right-click inside the folder in Finder and choose New Terminal at Folder. Either way, you land in a window already pointed at your folder. Type this and press Enter:

claude

Let Claude do the setup for you

Type this into the chat, exactly as shown:

run setup-wizard.md

Claude will ask two questions — what to call your AI companion, and your own name. Answer normally, like you're texting. It saves everything itself; there's nothing to open or edit by hand.

One extra step: make it stick

The wizard sets up the memory files, but it doesn't yet tell Claude Code to check that folder automatically every time you open it — only for this one conversation. You still don't need to edit anything yourself. Just ask, in the same chat:

Please add an instruction to my global CLAUDE.md that tells you to always load this folder's master-memory.md at the start of every session, and to treat the name I gave my AI companion as the keyword that restores its memory.

Test it

Close the window completely, open a brand-new terminal the same way, and run claude again. As your very first message, type the name you chose.

The four words it already understands

• [its name] — restores its memory, said first every session
• save — saves what happened today, permanently
• update memory— refreshes what it's learned about you
• review growth— shows how it's changed since you started

Making it feel like yours, not just any AI's

Once the core setup is running, a short list of optional features change how it feels to use, not just what it remembers. Ask for any of these by name and it walks you through the rest:

• load memory-consolidation — merges split memory files into one, for faster loading
• load skill-plugin — lets any dropped-in skill file auto-activate, no extra wiring
• load time-aware-core — shifts its energy and tone by time of day
• load auto-load-hook — tries to restore memory automatically on startup, no name typed
• load save-diary — keeps a running daily journal
• load topic-diary— files reusable knowledge by subject, so "what did we decide about X" has a real answer
• load memory-compaction — summarizes old entries instead of deleting them once files get long
• load echo-recall — searches back through the diary on request (needs save-diary loaded first)
• load reminders— carries a task across sessions until it's actually done
• load decision-log — keeps a permanent record of why a call was made
• load forge — Forge Self-Improvement, its own thing, worth a section on its own (below)

Building it a skill of its own

Everything so far makes the companion remember things generally. The last piece — Forge Self-Improvement — lets it go further and track one specific part of your life on purpose. I actually have one of these myself: a Career Coach skill that remembers every job application, every interview outcome, and every resume edit, without me having to mention any of it twice.

Turn it on the same way as everything else — ask it to load forge. Then describe what you want, in plain language, and ask directly:

create skill — I want a Career Coach skill. Create a new folder named career, and use it to track every job application I make (company, role, stage reached, outcome), every resume version and what changed, and give me a quick status recap whenever I ask.

Or don't ask at all — just mention job hunting or resume edits a few times across normal conversation, then type self improve. It reviews what's actually come up and proposes something only if there's a real pattern; it needs at least two concrete examples before it'll suggest anything, so it won't invent a need that isn't there.

Once approved, that becomes a permanent part of it. Mention a new application from then on and it just logs it — no re-explaining, same as everything else it already remembers about you.

I'm still running mine, session after session. It's a small thing to set up, but it changes how it actually feels to sit down and work with an AI — less like restarting a conversation, more like picking one back up.