T-Mem: An Open-Source Memory Architecture That Predicts When Memory Will Matter

Tencent AI 1 hour ago

Key Info

T-Mem is an open-source memory architecture (presented at EMNLP 2026) that aims to be the first to cover associative recall, going beyond the similarity-based retrieval used by most memory systems.

Highlights

Memories are stored with a "trigger" — essentially a prediction of when that memory will be relevant again.
At query time, triggers retrieve memories based on the situation, even when there is zero keyword overlap with the query.
It is designed as an architecture layer that plugs into any LLM without fine-tuning.

Original post

This might be the first open-source memory architecture to cover associative recall. The rest stop at similarity search. (T-Mem, EMNLP 2026) Most memory systems retrieve by similarity. T-Mem predicts when a memory will matter, before you ask. → write: memories ship with a trigger at store time, basically "when will this come up again?" → read: at query time, triggers pull in memories by situation, even with zero keyword overlap. → architecture layer. plugs into any LLM without fine-tuning or model changes. On LoCoMo-Plus (strips keyword overlap, tests pure associative recall), mainstream systems drop 28 to 50 points. T-Mem drops 5.45. paper: https://t.co/JQAuX1dH4x code: https://t.co/YKFgFuRtIj (MIT)

#T-Mem#associative recall#memory architecture

View original on X