Empowers AI assistants with human-like memory dynamics by integrating temporal decay, reinforcement learning, and a two-layer memory architecture.
STM Research provides a Model Context Protocol (MCP) server that equips AI assistants with sophisticated, human-like memory dynamics. It incorporates a novel temporal decay algorithm, inspired by the Ebbinghaus forgetting curve, ensuring memories naturally fade over time unless actively reinforced through use. The system features a two-layer architecture, distinguishing between volatile Short-Term Memory (STM) and persistent Long-Term Memory (LTM) which leverages Obsidian and Git for storage. It also includes smart prompting patterns for seamless LLM integration, supports Git-friendly JSONL storage, and builds a knowledge graph for richer memory organization.