TY  - RPRT
TI  - Learning What to Remember: Observability-Safe Memory Retention via Constrained Optimization for Long-Horizon Language Agents
AU  - Qingcan Kang
AU  - Liu Mingyang
AU  - Shixiong Kai
AU  - Kaichao Liang
AU  - Tao Zhong
AU  - Mingxuan Yuan
PY  - 2026
UR  - https://arxiv.org/abs/2606.10616
ID  - 2606.10616
ER  -