HomeResearch LibraryMemGPT: Towards LLMs as Operating Systems
system architectureChapter 2arXiv · 2023

MemGPT: Towards LLMs as Operating Systems

Charles Packer (UC Berkeley), Sarah Wooders (UC Berkeley), Kevin Lin (UC Berkeley)

Abstract

We present MemGPT, a system that intelligently manages different memory tiers in order to effectively provide extended context within the LLM's limited context window. MemGPT knows when to push critical information to a vector database and when to retrieve it.

Key Contributions

  • →Hierarchical memory management for LLMs
  • →Self-directed memory paging
  • →Persistent agent state across sessions
Eigenvector Commentary

MemGPT is the most directly applicable paper for enterprise long-horizon agents. The OS memory management metaphor is exactly right: agents need virtual memory, not just a context window. Every enterprise deployment of a long-running agent should implement some variant of this architecture.

Topics

memory managementcontext windowlong-horizon agentspersistent memory