Milla Jovovich Co-Devs in Open Source MemPalace Project: An AI Memory System That Never Forgets

When you think of Milla Jovovich, you think of Alice mowing down zombies in Resident Evil or Leeloo saving the world in The Fifth Element, which is why I’m writing this, because I follow her on IG where this announcement came outta nowhere. Because you probably don’t think of her and open-source AI software development, but here we are.

 

View this post on Instagram

 

A post shared by Milla Jovovich (@millajovovich)

Together with tech collaborator Ben Sigman, Jovovich has co-developed MemPalace, an AI memory system that aims to solve one of the most persistent problems in AI: the fact that your AI assistant forgets everything the moment your conversation ends.

My friend Milla Jovovich and I spent months creating an AI memory system with Claude. It just posted a perfect score on the standard benchmark – beating every product in the space, free or paid.

It’s called MemPalace, and it works nothing like anything else out there.

Instead… pic.twitter.com/trAUZyMHIe

— Ben Sigman (@bensig) April 6, 2026

So What’s the Problem, Exactly?

Think of it this way. Every time you talk to an AI like ChatGPT or Claude, it’s like talking to someone with total amnesia. You could spend hours explaining your work, your preferences, your entire project history, and the next day, it’s gone. You start from zero every single time. Its better nowdays especially with ChatGPT but for in-house context, you’d understand the issue.

Some existing solutions try to fix this by having the AI pick out “important” bits to remember, like saving a sticky note that says “user likes Postgres.” But it throws away the entire conversation about why you chose Postgres, which is usually the part that actually matters.

MemPalace: Remember Everything, Find Anything

MemPalace takes a different approach. Instead of letting AI decide what’s worth keeping, it stores everything, then organizes it so your AI can find what it needs, fast.

The organizing principle is borrowed from an ancient memory technique used by Greek orators. They would memorize entire speeches by mentally placing ideas in different rooms of an imaginary building. Walk through the building, find the idea. MemPalace applies this same concept to AI memory:

• Wings, big categories, like a person or a project
• Rooms, specific topics within a wing, like “authentication” or “billing”
• Halls, connections between related rooms
• Closets and Drawers, where the actual memories live, compressed for speed but with the originals always preserved

The team claims this structure alone improves retrieval accuracy by 34% compared to a flat, unstructured approach. It’s not just a metaphor, it’s an architectural decision that makes search results significantly better.

The Benchmarks

Here’s where things get interesting. MemPalace claims the following scores on established AI memory benchmarks:

• 100% on LongMemEval, 500 out of 500 questions answered correctly, which they say is the first perfect score ever recorded on this benchmark (achieved with an optional lightweight AI reranking step)
• 96.6% on LongMemEval without any AI assistance, just the local search engine, no cloud calls
• 92.9% on ConvoMem, reportedly more than double the score of Mem0, one of the leading paid alternatives

These are self-reported numbers from a brand-new repo with only a handful of commits, so independent verification is still pending. But the benchmark runner scripts are included in the repository for anyone to reproduce.

Local-First, Zero Cloud

Perhaps the most interesting angle for privacy-conscious users: MemPalace runs entirely on your machine. There’s no subscription, no API key required, no cloud service, and no data ever leaves your computer. The only dependency is ChromaDB, an open-source vector database. The whole system is MIT-licensed and free.

It also includes a custom compression dialect called AAAK, essentially a shorthand that AI models can read natively. The team claims it achieves 30x compression, letting months of conversation context load in roughly 120 tokens. Whether “lossless” is the right word for aggressive shorthand is debatable, but the practical result is that your AI can wake up knowing who you are before you type a single word.

So Why Is Milla Jovovich Building AI Tools?

According to Sigman’s post on X, the two spent months building the system together using Claude. Jovovich confirmed the project on her Instagram. The most likely read here is that Sigman handled the engineering while Jovovich was a deeply involved power-user and co-creator, not unlike how a producer shapes a film without necessarily operating the camera.

Either way, the project stands or falls on its own technical merit. Celebrity attachment gets eyes on the repo, but perfect benchmark scores and a genuinely novel architecture are what will keep developers there.

MemPalace is available now on GitHub and installable via pip install mempalace.