Access Internal Knowledge Base - Agentic RAG

Knowledge Base is intended to be used by agents to gain access to additional information that were not part of language model training. This is particularly useful for domain-specific knowledge or proprietary information.

Different collections are meant to separate bodies of data into logical groups for example based on topic, source of information or where this information is intended to be used.

To give an agent access to the Knowledge Base the agent must be equipped with a KnowledgeBaseSearch tool. This tool is used to query the Knowledge Base and retrieve relevant information. The agent needs to have an information on which collection it should query. The agent Persona is a good place to include the name of the collection it may use and when to use it.

Agent persona may for example include:

You are a research assistant. You have access to the "Research" collection of the Knowledge Base. Use KnowledgeBase tool with this collection to find information about current research trends, internal notes and reports. Use it to answer questions about research papers, methodologies, and findings.

Agent of course does not need to be limited to a single collection and the description in Persona may include more than one collection name.

When the agent is equipped with the KnowledgeBaseSearch tool, it can use it to query the Knowledge Base and retrieve relevant information. The agent can use the KnowledgeBaseSearch tool to search for specific topics or keywords within the specified collection. Agent may take multiple iterations to refine the search and retrieve the most relevant information. The process of querying the Knowledge Base is similar to using a search engine, where the agent can input specific queries and receive relevant results. Used method is known as Agentic Retrieval-Augmented Generation (RAG).