Intrinsic Memory Agents: Heterogeneous Multi-Agent LLM Systems through Structured Contextual Memory
Sizhe Yuen, Francisco Gomez Medina, Ting Su, Yali Du, Adam J. Sobey · 2025
Read Intrinsic Memory Agents: Heterogeneous Multi-Agent LLM Systems through Structured Contextual Memory by Sizhe Yuen, Francisco Gomez Medina, Ting Su, Yali Du, Adam J. Sobey (2025) 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-2508-08997.txt.
Description
Sizhe Yuen and co-authors present a framework that gives each agent in a multi-agent language model system its own structured memory. Such teams promise strong collaborative problem solving, but fixed context windows erode memory consistency, role adherence, and procedure following. Their method keeps role-aligned memories that grow out of each agent's own outputs. The memory stays focused on task-relevant content and keeps each specialist's viewpoint.
A generic memory template applies to new problems without hand-crafted prompts. Benchmarks on three datasets, covering planning, fact checking, and a household task world, show performance at or above leading rivals, with the highest consistency. On a complex data pipeline design task, the approach scores better on five quality measures, from scalability to documentation. The authors argue intrinsic memory is a practical fix for a core weakness of agent teams.
Catalog record
| CREATOR | Sizhe Yuen, Francisco Gomez Medina, Ting Su, Yali Du, Adam J. Sobey |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-08997