So we’ve been poking around a GitHub repo this week.
Not unusual for our team. What is unusual: the name on the commits.
Milla Jovovich. Yes, that one. Leeloo from The Fifth Element. Alice from the Resident Evil franchise.
She just shipped an open-source AI memory system called MemPalace.
(I had to read that twice too.)
Here’s the short version.
Jovovich started using ChatGPT and Claude heavily in late 2025. Thousands of conversations. Business decisions, creative projects, debugging sessions, the whole thing. Then she hit the wall every serious AI user eventually hits.
You start a new session. The AI has amnesia.
All that context, all those decisions, all that thinking-out-loud. Gone.
She tried the existing tools (Mem0, Zep). They didn’t work for her, because they used AI to summarize what was worth keeping. The reasoning. The trade-offs. The “why” underneath the decisions. All of it discarded.
So she partnered with developer Ben Sigman, spent months building inside Claude Code, and on April 5, 2026, dropped the whole thing on her own GitHub. MIT licensed. Free. Runs locally. Zero API costs.
It crossed 7,000 stars in 48 hours. It’s now past 50,000.
The architecture borrows directly from an ancient mnemonic technique called the method of loci. Information gets organized as wings, rooms, and drawers. Verbatim, not summarized. Searchable by structure, not just keywords.
Is the project perfect? No. There are honest critiques in the developer community about benchmark methodology and code quality (the Medium write-up by Ewan Mak walks through them carefully, and is worth your time if you’re curious).
But that’s not why I’m bringing it to you.
I’m bringing it to you because of what the project represents.
Because the actual story underneath MemPalace isn’t really about a Hollywood star who learned to code.
It’s about a problem she ran into.
A problem you and your organization run into every single day.
Which brings me to the actual point
Most nonprofits don’t need an AI memory palace.
Not today, anyway.
For most organizations, it would feel like every other shiny new thing. Another layer of tech bolted onto a stack that already feels too complex, too fragmented, too under-resourced to manage properly.
(The phrase “we just got Salesforce working” comes to mind.)
And yet.
The idea underneath it is worth your attention.
Because it quietly points at a problem nearly every nonprofit already has.
Not an AI problem. A memory problem.
You already know what this looks like
Spend enough time inside any organization and you start to see it everywhere.
The donor history that lives partly in the CRM, partly in someone’s inbox, partly in a three-year-old spreadsheet nobody owns.
The campaign insight that was definitely documented somewhere (was it the Q3 retro? the quarterly deck? a Google Doc now buried in someone’s archived folder?).
The institutional knowledge that walks out the door every time a key staffer moves on.
We’ve built systems to store data.
We haven’t really built systems to remember.
So what’s a memory palace, actually?
It’s an old idea. Ancient, actually.
The technique goes back to Greek and Roman rhetoric. Orators would mentally place ideas inside an imagined building, then “walk through” it to recall their speeches in order. Modern incarnations show up in books like Moonwalking With Einstein.
The point isn’t the spatial trick.
The point is structure over storage.
Don’t just dump information somewhere. Organize it so it can be pulled back meaningfully. Not by keyword. By relationship, context, and meaning.
Now apply that to AI. Which is exactly what Jovovich did.
Where this is actually going
Today’s “AI memory” systems (MemPalace included) are still early.
Most of them are closer to enhanced search than genuine understanding. Embeddings, vector similarity, retrieval pipelines (powerful stuff, but not magic).
The bolder claims floating around right now don’t really hold up under scrutiny yet. Perfect recall. Lossless compression. Fully structured organizational knowledge.
(We are not there.)
But that’s not the interesting part.
The interesting part is the direction.
AI is shifting from:
Stateless to stateful.
Answering questions to remembering context.
Generating outputs to participating in ongoing work.
That shift is real. And for nonprofits, the implications are real too.
The question worth sitting with
Forget the tools for a second.
Ask this instead:
If we had a system that truly remembered everything we’ve done (every donor interaction, every campaign insight, every decision your team ever made), what would that change?
Here’s the thing.
Most nonprofits aren’t limited by effort. Your team is working hard. Probably too hard.
They’re limited by fragmentation.
Imagine the version where it works
Your AI assistant doesn’t just draft a stewardship email. It understands the full arc of that donor’s relationship with your organization.
Your campaign planning doesn’t start from a blank page. It builds on structured memory from every campaign you’ve run before.
Your team doesn’t lose knowledge when someone leaves. It compounds it.
That’s the real promise here.
Not better tools. Better continuity.
Now, the catch
Before any of this works, your underlying data has to actually be usable.
And for most nonprofits, that’s where the wheels come off.
Disconnected systems. Inconsistent naming conventions. Half-filled records. Manual workarounds duct-taped to legacy platforms (you know exactly which ones).
You don’t need to read a single research paper on AI memory to feel this problem.
You just need to try pulling a clean report.
The boring work is the strategic work
So maybe the practical takeaway isn’t “go adopt MemPalace tomorrow.”
It’s prepare for what’s coming.
Because the organizations that get the most out of the next wave of AI memory systems won’t be the ones chasing the newest tool. (Or the celebrity-endorsed one.)
They’ll be the ones doing the quieter, less glamorous work right now.
Cleaning their data.
Standardizing their systems.
Thinking carefully about how information flows between people, tools, and decisions.
(None of that ends up on a conference panel. All of it compounds.)
Signal, not solution
That’s why a project like MemPalace is worth paying attention to. Not as a thing to deploy on Monday. As a thing to read.
A glimpse into a near future where:
Memory isn’t scattered.
Knowledge isn’t lost.
And AI doesn’t just respond.
It remembers.
The organizations that will be ready for that future are the ones putting in the unglamorous work today.
Not because it’s exciting.
Because it’s the foundation everything else gets built on.
Yours in fundraising,
Erik
