MemPalace: Milla Jovovich's Open-Source AI Memory System

Milla Jovovich -- yes, the actress from The Fifth Element -- just open-sourced an AI memory system called MemPalace, and it is topping the benchmarks. The project has gained over 23,000 GitHub stars within days of its April 2026 release.

How It Works

MemPalace organizes AI memory into spatial geometries inspired by the ancient Greek method of loci -- the memory palace technique. Instead of dumping everything into a flat vector database, MemPalace structures memory into wings, halls, and rooms. This architectural metaphor mirrors how human spatial memory works, giving the AI a navigable structure for retrieval rather than a bag of floating embeddings.

A custom compression algorithm called AAAK condenses months of conversations into a tiny token footprint without losing semantic meaning. Where other memory systems either summarize (losing detail) or store verbatim (burning tokens), AAAK finds a middle ground: a compression dialect that only the AI can parse, preserving the full semantic content at a fraction of the token cost.

Why It Matters

The core problem MemPalace solves is that AI models cannot remember conversations. Every session starts from zero. Existing solutions like Mem0 and Zep use AI to decide what is worth remembering, which introduces a lossy filter. MemPalace takes a different approach: store everything, compress it intelligently, and organize it spatially for fast retrieval.

Feature	MemPalace	Mem0	Zep
Cost	Free (local)	$19-249/mo	$25/mo+
Storage	Verbatim + compression	AI-selected summaries	AI-selected summaries
Architecture	Spatial (wings/halls/rooms)	Flat	Flat
Runs locally	Yes	No	No

Key Takeaways

• MemPalace uses the method of loci to organize AI memory into spatial structures
• The AAAK compression algorithm preserves semantic meaning while drastically reducing token usage
• It runs entirely locally with zero API costs
• It is MIT-licensed and supports Claude Code, ChatGPT, and Cursor via MCP
• It scored 96.6% on the LongMemEval benchmark

Resources

• MemPalace on GitHub -- Open-source repository
• MemPalace Official Site -- Project homepage
• Cybernews Coverage -- Coverage of the launch