Tag
Introduces AdaCoM, an external LLM-based context manager for frozen agents, using reinforcement learning to improve long-horizon task performance by preserving task constraints and pruning stale content, with experiments on web search and deep research benchmarks.