Adding Long Term Memory to OpenGPTs

OpenGPTs is an open source implementation of OpenAI GPTs and Assistant API, allowing for the creation of conversational agents with a flexible cognitive architecture. Memory plays a significant role in these agents, but current implementations only support basic conversational memory. Long-term memory remains underexplored. This blog post discusses different types of memory, such as conversation memory, semantic memory, and generative agents, along with their challenges and potential solutions. The authors propose an application-specific approach to long-term memory, focusing on the state that is tracked, how it is updated, and how it is used in a given context. They demonstrate this concept by building a "Dungeons and Dragons" chatbot with character and quest memories within OpenGPTs.