Tencent has released the ContextPilot-E4B checkpoint, a Gemma4-E4B 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.
- Combines an extended toolset with planning, structured memory, retrieval, and soft context offloading.
- Uses context-aware partial rollout to focus exploration on sensitive context-editing decisions.
- Employs fine-grained credit assignment by training intermediate snapshots using downstream branch outcomes.
- Evaluated on long-context question answering and deep-search tasks, showing stronger performance with compact working contexts.
The checkpoint is intended for research on proactive context management, long-horizon agents, long-context QA, and deep search.