According to CoinMeta, the long-term memory system of Instinct was reverse-engineered by Dhravya Shah, the founder of Supermemory. He discovered that Instinct did not simply reinstate all chat content back into its original context; instead, it organized long-term information into Markdown files managed by Git, while continuously maintaining the user's current state. Supermemory is specifically designed to provide long-term memory infrastructure for agent, capable of extracting facts from chats, files, and web pages, maintaining character relationships, tracking temporal changes, and creating user profiles. It also passes on relevant memories to agent through methods such as API. Shah believes that Instinct and Supermemory share very similar approaches to "how to remember" within the context of agent. With each conversation, Instinct obtains a summary of the user's state, which includes their background and communication habits, amounting to about 4250 token entries. More detailed personal information and tasks are stored in Markdown files, which can be accessed through keyword searches. Shah also notes that these memories are continuously organized in the backend, with the system incorporating temporary information into specific tasks and using new facts to update older ones.