If you're interested in where memory is heading at Weaviate, sign up for a preview today.
Memory is a term you've probably been hearing like a steadily increasing drumbeat over the last year. This is for good reason – many of 2024 and 2025's PoC's have graduated and become full-fledged, mission-critical production applications. And with that an interesting problem has emerged, and it's not one that can be solved by today, or tomorrow's LLMs. Not because models aren't capable, but because this is fundamentally a systems problem and not a model limitation.
What we're running into is the infinite loop the user and the system (read: the LLM embedded in an application) find themselves stuck in. A loop created by the absence of one critical trait: continuity.
I Am A Limited Loop
Today's AI applications operate in what might be called a limited loop, where each interaction is treated largely as disposable, bound to a single session and with incidental carryover between sessions. For those of you who have used chat-based LLMs without memory, some of these frustrating loops may already feel familiar. You might have a particular preference for how you want information to be presented, or simple facts about yourself like your name, where you live, what your favorite foods are. Personally, I like it when complex or unfamiliar topics are explained simply at first, and only then do I want the deep dive.
