ContextPilot: Teaching Agents for Proactive Context Management via Fine-grained RL
ContextPilot trains agents to manage their working context with finer RL credit for each edit.
The paper adds planning, long-term memory, and soft context offloading tools to proactive context management. Its RL method branches around critical context-editing choices, using context and entropy changes to decide where exploration matters. The authors report stronger results on long-context QA and deep search while keeping the working context more compact. Code is listed as available. ArXiv · AI/CL/LG's note
The paper adds planning, long-term memory, and soft context offloading tools to proactive context management. Its RL method branches around critical context-editing choices, using context and entropy changes to decide where exploration matters. The authors report stronger results on long-context QA and deep search while keeping the working context more compact. Code is listed as available. ArXiv · AI/CL/LG's note
score 4