STM Research FAQs

Question 1

Can STM Research's memory behavior be customized?

Accepted Answer

Yes, STM Research is highly configurable. Users can tune parameters such as the memory half-life, decay model (e.g., Power-Law, Exponential), use count weighting, and thresholds for memory forgetting or promotion. This allows for tailoring memory behavior to specific use cases, from ephemeral notes to long-term archival.

Question 2

What is STM Research and what problem does it solve?

Accepted Answer

STM Research is an MCP server designed to provide AI assistants with human-like memory dynamics. It addresses the challenge of AI memory limitations by allowing memories to naturally fade over time unless reinforced, mimicking cognitive processes like the Ebbinghaus forgetting curve.

Question 3

How does STM Research integrate with Large Language Models (LLMs) and AI assistants?

Accepted Answer

It includes a smart prompting system with patterns for auto-save, auto-recall, and auto-reinforce, enabling AI assistants to naturally access and manage memories without explicit commands. It functions as a Model Context Protocol (MCP) server for seamless LLM integration.

Question 4

What is the two-layer memory architecture?

Accepted Answer

STM Research features a Short-Term Memory (STM) layer for active, temporally decaying data stored in Git-friendly JSONL, and a Long-Term Memory (LTM) layer for permanent storage, often integrated with Obsidian Markdown files. Memories are automatically promoted from STM to LTM based on retention thresholds.

Question 5

How does STM Research mimic human-like memory?

Accepted Answer

It employs a novel temporal decay algorithm, where memory scores continuously decrease based on time since last access. This decay is offset by reinforcement learning, where frequent usage strengthens memories, and an adjustable importance parameter, making memory retention highly dynamic and similar to human cognition.

STM Research

Acerca de

Características Principales

Casos de Uso