Retrieved knowledge

As discussed previously, retrieved knowledge refers to information that is retrieved in real time from a knowledge store based on the user’s request. Unlike training knowledge, which is static and fixed at the time the model was trained, retrieved knowledge is dynamic to the context of the conversation. The pattern is as follows:

1. User asks a question that requires external knowledge.
2. The agent system retrieves data relevant to that question from a knowledge source. This is achieved through a tool call, which could search a database, a piece of text, a vector store of embedded documents, a search engine API, etc.
3. The retrieved information is then fed into the LLM.
4. The LLM generates an answer that incorporates or is grounded in the retrieved information.

This incorporates external knowledge into an agent. In this context, “external” knowledge refers to any information that is not inherently stored in the...