MemRouter: Memory-as-Embedding Routing for Long-Term Conversational Agents
Tianyu Hu Weikai Lin University of Central Florida University of Rochester [email protected] [email protected] Weizhi Zhang Jing Ma University of Illinois at Chicago Case Western Reserve University [email protected] [email protected] Song Wang University of Central Florida [email protected] · 2026
Read MemRouter: Memory-as-Embedding Routing for Long-Term Conversational Agents by Tianyu Hu Weikai Lin University of Central Florida University of Rochester [email protected] [email protected] Weizhi Zhang Jing Ma University of Illinois at Chicago Case Western Reserve University [email protected] [email protected] Song Wang University of Central Florida [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-00356.txt.
Description
Tianyu Hu and 4 co-authors speed up the memory decisions of long-term conversational agents. Such an agent must decide, at every turn, what is worth storing for later recall. Recent systems make that call by generating text with a large model at every turn, which is slow and costly. MemRouter replaces the call with a learned embedding-based router.
The router encodes each turn with its recent context, passes the result through a frozen language model backbone, and classifies it with light heads, training only a small set of parameters. On a standard long-conversation benchmark, with retrieval and answering held identical, MemRouter beat a language-model memory manager on every question type. It also cut the delay of each memory decision from about a second to a small fraction of one.
Catalog record
| CREATOR | Tianyu Hu Weikai Lin University of Central Florida University of Rochester [email protected] [email protected] Weizhi Zhang Jing Ma University of Illinois at Chicago Case Western Reserve University [email protected] [email protected] Song Wang University of Central Florida [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.00356