Memory for agents

Memory is an essential aspect of developing effective AI agents. While LLMs do not inherently remember information, developers can intentionally add memory to improve user experience. Memory is application-specific, and different types of memory, such as procedural, semantic, and episodic, can be used depending on the agent's purpose. Developers are also considering how to update agent memory, with options including "in the hot path" and "in the background." LangChain has built various functionalities for leveraging memory in applications, including low-level abstractions for a memory store and templates for running memory both ways.