Semantic memory stores facts, concepts, and stable knowledge that remain consistent across contexts—the “things I know” of an AI agent. It can support personalization and adaptation through a user profile or curated documents. Long-term memory lives outside the model; retrieval is explicit, and only matching entries enter the context window. section 5.1.2
Examples of semantic knowledge include:
- User facts, such as a preference for Python or premium subscriber status. section 5.1.2
- Domain knowledge, such as the significance of a yield curve inversion in finance. section 5.1.2



