Tencent has released the ContextPilot-14B checkpoint, a Qwen3-14B model designed for proactive context management in long-horizon language-model agents. The framework teaches agents to plan, maintain long-term memory, and offload less useful context while reasoning and using tools.
ContextPilot combines an extended toolset with planning, structured memory, retrieval, and soft context offloading. It utilizes context-aware partial rollout to focus exploration on sensitive context-editing decisions and fine-grained credit assignment to train intermediate snapshots based on downstream outcomes. The resulting agents are evaluated on long-context question answering and deep-search tasks.
This checkpoint is intended for research on proactive context management, long-horizon agents, long-context QA, and deep search.