Ait is a technical exploration of an idea relating to incorporating long-term memory in applications using large language model (LLM).

Ait generates a new context for every user query. That context is built by finding prior query-response pairs (experiences) which are related to the current query. The experiences are selected by traversing a tree of related experiences to build a set of candidates experiences, and then keeping those experiences most similar to the current query.