Milla Jovovich creates MemPalace AI memory tool with “perfect score” on benchmark, but devs aren’t buying it
Milla Jovovich’s MemPalace AI Tool Sparks Viral Success—and Developer Skepticism
DUBAI – Hollywood star Milla Jovovich is making headlines far from the film set, but this time it isn't for a new Resident Evil installment. The actress has co-created MemPalace, an open-source AI memory tool that claims to solve the industry’s "AI amnesia" problem, a major talking point in mena ai news circles currently driving discussions on advanced AI capabilities across the region.
Launched on April 6, 2026, the project exploded onto GitHub, amassing over 23,000 stars in just 48 hours. However, the viral celebration was quickly met with intense scrutiny from the developer community over "perfect" benchmark scores and the true nature of the actress’s technical contributions.
The Concept: Human Mnemonic Meets Modern AI
The project was born out of Jovovich's personal frustration with AI models "forgetting" long-term conversations. Alongside engineer Ben Sigman, CEO of Libre, she developed a system based on the "Method of Loci" (the Memory Palace), an ancient Greek technique adapted to the digital age.
Technical Highlights:
- Structured Retrieval: Unlike flat databases, MemPalace organizes data into virtual "Wings," "Rooms," and "Halls", effectively building an AI Memory Palace.
- AAAK Compression: A proprietary dialect developed by the duo that claims 30x lossless compression, allowing AI to read massive histories with minimal token usage.
- Local-First Privacy: Using ChromaDB and SQLite, the tool stores all data locally on the user's machine—meaning zero API costs and total privacy.
The Benchmark Controversy: From 100% to 96.6%
The primary point of contention lies in the LongMemoryEval benchmark. Jovovich and Sigman initially claimed a "perfect score" of 100%—a world first.
The developer community on X (formerly Twitter) and Reddit quickly dismantled the claim. Critics argued the score was achieved through "teaching to the test"—writing specific code patches for the three questions the model originally missed. Following the backlash, Sigman revised the "raw mode" score to 96.6%.
"Vibe Coding" or Celebrity Front?
A viral thread on X has accused the project of being a "celebrity-faced grift." Skeptics pointed to Jovovich’s sparse GitHub history—showing only 7 commits—and alleged that a "mystery developer" named Lu was the true architect behind the code.
Despite the drama, the tool is already seeing real-world adoption. Brian Roemmele, a prominent tech commentator, reportedly implemented MemPalace into his "zero-human" company. For professionals tracking the AI intelligence gulf, these open-source tools represent an uneven but fascinating frontier in AI development.
Whether MemPalace is a technical breakthrough or a high-profile marketing experiment, it has succeeded in one thing: forcing a conversation on how AI retains—and discards—the information that matters most.