MemGPT is a fixed-context LLM processor enhanced with a hierarchical memory system and a set of functions, enabling it to manage its own memory. The primary context is the (fixed-length) LLM input. MemGPT parses the LLM's text output in each processing cycle and controls or returns control when executing function calls, used to move data between the main and external contexts. When the LLM generates a function call, it can request immediate execution to link functions. In yield scenarios, the LLM will not execute again until the next external event triggers (e.g., a user message or scheduled interruption).