MemGPT, short for Memory-GPT, is a project from UC Berkeley that addresses the memory limitations of Large Language Models (LLMs) like GPT-4. It introduces a tiered memory system inspired by operating systems' memory management, enabling LLMs to manage memory autonomously for more coherent interactions and document analysis. MemGPT is used for long-term conversational agents and document analysis, surpassing traditional context window limitations. Its core components include virtual context management and autonomous memory management. The innovative memory management approach of MemGPT opens the door to robust long-term interactions and document analysis capabilities, providing practical solutions for businesses and developers working with LLMs.