MemGPT

Key findings used in wiki

• MemGPT introduces a virtual context management system that allows LLMs to operate with effectively unbounded memory, informing GiveCare's approach to maintaining long-term caregiver context across SMS sessions.
• The information extraction and memory management architecture provides patterns for how the Mira agent decides what to store, retrieve, and forget from conversation history.
• The paper's distinction between "main context" (working memory) and "external context" (archival memory) maps to GiveCare's session-level vs. persistent caregiver profile storage.
• MemGPT's self-directed memory operations (push/pop/search) inform the Mira agent's autonomous memory management, where the agent decides when to consult or update stored facts.
• The operating-system metaphor frames memory as a managed resource rather than a passive store, shaping GiveCare's approach to memory lifecycle management including expiry and relevance scoring.