MAGE: Active Execution-State Management for Long-Horizon Agents
In brief
- MAGE stores interactions in a hierarchical tree to track agent execution state.
- The system isolates errors and limits context growth for better performance.
- MAGE cuts token usage by 55.1% while raising task success rates by 7.8–20.4 percentage points.
Researchers introduce MAGE, a memory system for LLM-based agents that tracks execution state via a hierarchical tree. Unlike semantic similarity approaches, MAGE isolates errors and caps context growth, boosting performance and cutting costs.
This summary gives you the essentials. The full article, its figures and quotes are on cs.AI updates on arXiv.org.
Read the original articleAI-generated summary of the source, checked by rules. Report an error