ExpGraph: Model-Agnostic Experience Learning with Graph-Structured Memory for LLM Agents
Tao Feng 1 , Chongrui Ye 1 , Tianyang Luo 1 , Jingjun Xu 1 , Xueqiang Xu 1 , Haozhen Zhang 2 Zhigang Hua 3 , Yan Xie 3 , Shuang Yang 3 , Ge Liu 1 , Jiaxuan You 1 1 University of Illinois Urbana-Champaign 2 Nanyang Technological University 3 Meta Monetization AI · 2026
Read ExpGraph: Model-Agnostic Experience Learning with Graph-Structured Memory for LLM Agents by Tao Feng 1 , Chongrui Ye 1 , Tianyang Luo 1 , Jingjun Xu 1 , Xueqiang Xu 1 , Haozhen Zhang 2 Zhigang Hua 3 , Yan Xie 3 , Shuang Yang 3 , Ge Liu 1 , Jiaxuan You 1 1 University of Illinois Urbana-Champaign 2 Nanyang Technological University 3 Meta Monetization AI (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-2605-30712.txt.
Description
Feng and colleagues present ExpHarness to improve frozen language-model executors, leaving their parameters untouched. Agents gather experience across tasks that rarely flows back into the harness. It distills trajectories into skills and failure lessons within an experience graph. It trains a copilot deciding how broadly to explore the graph and how strongly to favor useful experiences over similar ones. The copilot learns from a reward based on score difference with and without experience. Tests show ExpHarness beats the baseline while cutting interaction steps by up to a fifth.
Catalog record
| CREATOR | Tao Feng 1 , Chongrui Ye 1 , Tianyang Luo 1 , Jingjun Xu 1 , Xueqiang Xu 1 , Haozhen Zhang 2 Zhigang Hua 3 , Yan Xie 3 , Shuang Yang 3 , Ge Liu 1 , Jiaxuan You 1 1 University of Illinois Urbana-Champaign 2 Nanyang Technological University 3 Meta Monetization AI |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 160936 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.30712