MEM1: A Constant-Memory RL Framework for Long-Horizon Language Agents

ABSTRACT

Modern language agents must operate over long-horizon, multi-turn interactions, but most rely on full-context prompting which leads to unbounded memory growth. We introduce MEM1, an end-to-end reinforcement learning framework that enables agents to operate with constant memory. At each turn, MEM1 updates a compact shared internal state that jointly supports memory consolidation and reasoning. This state integrates prior memory with new observations while strategically discarding irrelevant information. Experiments across internal retrieval QA and multi-turn web shopping show that MEM1-7B improves performance by 3.5x while reducing memory usage by 3.7x compared to larger-scale instructors.