MemGPT

MemGPT is an LLM operating system (OS) that uses virtual context management to provide unbounded context, enabling perpetual conversations and large document analysis.

MemGPT (Memory-GPT) is a system that solves the fundamental context window limitation of large language models (LLMs). It employs virtual context management, a technique inspired by hierarchical memory systems in traditional operating systems, to create the illusion of extended memory. The architecture features a two-tier memory system: an in-context 'core memory' for immediate use and an out-of-context 'archival memory' (typically a vector database like Chroma or pgvector) for long-term storage. The LLM itself acts as the OS, using designated tool calls (like `core_memory_replace` or `conversation_search`) to intelligently move data between these tiers. This design allows MemGPT agents to excel in tasks like comprehensive document analysis (far exceeding the model's context limit) and multi-session chat, where agents dynamically remember, reflect, and evolve over long-term interactions.

https://memgpt.ai