AI Memory Operating System Tackles Forgetting in Long-Running Tasks

Large language models are constrained by the capacity of their context windows and often suffer from memory lapses and hallucinations during long-running tasks. Early instructions and intermediate results can fall out of memory, making it difficult for AI agents to handle continuous work spanning several days. Giving AI a systematic way to manage memory, much as computers manage files, is widely seen as a key hurdle to making agent technology practical and has become a major focus for researchers and industry.

Researchers recently unveiled EverMemOS, an automatically organized memory operating system that they say is the first to elevate AI memory management to the lifecycle level of a computer operating system. Its three-stage process—episodic trajectory construction, semantic consolidation and reconstructive recall—allows memories to accumulate continuously, remain organized and be retrieved accurately when needed. According to the research team, AI agents equipped with the system can already perform ultra-long-horizon tasks spanning several days reliably, marking another step toward practical AI assistants capable of working continuously for days.