MIRIX Upgrades AI Memory: 410% Faster, 99.9% Lighter, App Now Live
MIRIX: AI memory breakthrough by UCSD & NYU—410% faster, 99.9% less storage. Multimodal, multi-agent, now on Mac. Redefining AI cognition.
"AI Disruption" Publication 7100 Subscriptions 20% Discount Offer Link.
MIRIX, a new system led by teams from UCSD and NYU, is redefining the landscape of AI memory.
Over the past decade, we have witnessed large language models sweep across the globe, capable of everything from writing assistance to code generation.
However, even the most powerful models still have a fundamental weakness: they don't remember you.
Addressing this current state, Yu Wang, a PhD student at the University of California, San Diego (UCSD), and Professor Xi Chen from New York University have jointly launched and open-sourced MIRIX — the world's first truly multimodal, multi-agent AI memory system.
MIRIX's performance is exceptionally impressive! On ScreenshotVQA, a challenging benchmark requiring deep multimodal understanding, MIRIX's accuracy is 35% higher than traditional RAG methods with 99.9% lower storage overhead, and 410% higher than long-text methods with 93.3% lower overhead.
In LOCOMO long conversation tasks, MIRIX significantly outperforms all existing baselines with a score of 85.4%, setting a new performance benchmark. Moreover, the team has also launched a Mac application product.
Through this ready-to-use application, anyone can let AI see what you see, understand what you do, and transform everything into persistent digital memory.