MemGPT (MemoryGPT) is an innovative system designed to overcome the inherent limitation of large language models (LLMs) regarding their constrained context windows. Drawing inspiration from hierarchical memory systems in traditional operating systems, MemGPT employs virtual context management to create the illusion of extended virtual memory. This allows LLMs to effectively manage and utilize context far beyond their native limits. The system has demonstrated significant utility in two key domains: analyzing large documents that typically exceed an LLM's context window, and facilitating multi-session chats where conversational agents can remember, reflect, and dynamically evolve through long-term user interactions. MemGPT empowers LLMs to manage their own memory, making them more versatile for perpetual chat and complex analysis tasks.