An Actress Built an AI Memory Tool. Here's Why It Might Actually Matter for Your Work.

You know the frustration. You had a good conversation with ChatGPT or Claude a few weeks ago, something genuinely useful, a contract clause you needed, a pricing structure you worked out, advice that actually helped. And now that session is gone. You can’t find it. You’re starting from scratch again.

That problem is what this story is about.

Milla Jovovich, the film actress, known for The Fifth Element and Resident Evil, spent several months building an AI tool. She quietly released it on GitHub, and then something interesting happened: the AI community immediately started testing every claim she made. Within two days, they found several errors in how the tool was described. Most companies would have gone quiet at that point. Jovovich and her co-developer did the opposite, they published a full, public correction and explained exactly what was wrong with their original claims.

That matters. We’ll come back to it.

What Is MemPalace, Plainly?

MemPalace is a system that gives your AI assistant a memory that doesn’t disappear when you close the chat.

Here’s why that distinction actually matters. When you use ChatGPT or Claude, each conversation starts with a blank slate. The AI doesn’t remember your previous sessions. It doesn’t know that six months ago you asked about pension contributions, or that you run a plumbing business, or that you decided to price your services differently. This is fine for one-off questions. It becomes a real problem when AI becomes part of how you actually work.

The usual solution is for the AI company to build memory into their product. OpenAI has this. Google has this. These work, to a point. The limitation is that your conversation data is stored on their systems. For many people that’s fine. For others, particularly in professional contexts, it isn’t.

MemPalace works differently. Instead of your memory living on the AI company’s servers, it lives on your own computer. No subscription. No data going anywhere else. You own the memory, not the AI company.

What This Means for Your Business

If you’re self-employed, run a small business, or do meaningful work with AI, here is the practical version.

You use an AI assistant to help with client proposals, contract questions, pricing decisions, marketing copy, compliance queries, staff handbooks. Each conversation is useful in the moment, but over time the value evaporates because you can’t find the earlier context. You end up re-explaining your situation every session. Or you screenshot important exchanges and file them in a folder you’ll never look at again.

MemPalace is designed to fix that. It runs in the background, stores everything you’ve discussed with your AI assistant, and can retrieve the relevant context when you need it, even if the original conversation was months ago. The claim the project makes is a 96.6% retrieval accuracy rate on a standard benchmark, independently verified. You can look at the methodology on their GitHub page if you want to verify it yourself.

The cost comparison they publish: roughly 70 pence per year to run this locally, versus around £400 per year for a comparable cloud memory service. That cost differential is worth noting if you’re paying for AI subscriptions already.

Should You Actually Care?

Whether this is relevant to you depends on a few things.

You should pay attention if: You use AI assistants regularly for work decisions, you’ve lost useful context between sessions, or you’re not comfortable with your conversation data being stored on someone else’s infrastructure. Teachers managing student data, consultants with client confidentiality obligations, small business owners with trade secrets, the local-only nature of this tool addresses concerns that cloud memory services simply can’t.

You probably don’t need this if: Your AI usage is mostly one-off questions with no need for continuity, or you’re comfortable with the major AI companies holding your memory state.

The setup currently requires some technical comfort, it’s a command-line tool, not a point-and-click application. That’s a legitimate barrier for non-developers. The team is working on making it more accessible, but if you’re not comfortable installing and running Python tools, it’s worth bookmarking this and coming back in a few months.

Why the Honesty Is the Actual Story

I promised I’d come back to this.

When the AI community found errors in MemPalace’s original claims, the team could have deflected. They didn’t. They published a full correction, acknowledged the specific errors, explained what was still valid, before the community could force the issue. That’s uncommon in AI announcements. Most companies announce, defend, and hope the noise dies down.

A team that publishes its own errors is more trustworthy than one that doesn’t. It’s also a useful signal when you’re evaluating any AI tool: not what they claim works, but how they behave when those claims are challenged.

The Bottom Line

AI assistants are becoming genuinely useful for work. The missing piece for many people is continuity, the system remembering who you are, what you do, and what you’ve already figured out. MemPalace is an attempt to solve that with a local, open-source approach rather than relying on the AI companies to build it into their products.

Whether it becomes mainstream depends partly on whether it gets easier to set up. The architecture is interesting and the cost economics are compelling. For now, it’s worth understanding the direction this is heading, because it is heading there.

MemPalace is on GitHub at github.com/milla-jovovich/mempalace. The benchmark methodology is published and reproducible. The team’s correction is worth reading on its own, it’s a reasonable example of how AI announcements should work.