Knowledge Base

What is a knowledge base?

A knowledge base is an organized collection of information that people or software can search and use. It may contain articles, policies, product documentation, frequently asked questions, structured facts, procedures, or relationships between entities.

AI agents often retrieve from a knowledge base to answer questions or support decisions with domain-specific evidence.

How a knowledge base is organized

A knowledge base can be stored as documents, database tables, a content-management system, a knowledge graph, or a combination of formats. Useful records normally include source, ownership, version, effective date, and access metadata.

For example, a support knowledge base may contain troubleshooting guides and return policies. An agent can retrieve the relevant guide, place selected passages into context, and produce a grounded response.

Knowledge base versus vector database

A knowledge base is the information collection and its organization. A [vector database](/glossary/vector-database/) is a storage and search technology for vectors.

A knowledge base may use a vector database for semantic retrieval, but it can also use keyword search, relational queries, graph traversal, or manual navigation. The vector index is not the knowledge itself; it points back to source content and metadata.

Knowledge base versus model knowledge

Model knowledge is learned during training and encoded in model parameters. A knowledge base exists outside the model and can be updated, permissioned, cited, and deleted without retraining.

This makes knowledge bases useful for private or changing information. However, the agent must retrieve relevant content into the current context before the model can use it.

Quality and governance

A large collection is not automatically a good knowledge base. Duplicate, outdated, contradictory, or ownerless content produces unreliable retrieval. Missing metadata can make it difficult to choose the current policy or enforce permissions.

Builders should define publishing ownership, review cycles, source authority, versioning, and deletion rules. Retrieval evaluation should identify both missing content and content that exists but cannot be found.

Why it matters

A knowledge base provides a controlled reference layer between AI systems and organizational information. It can improve grounding, reduce repeated manual lookup, and make answers traceable.

Its value depends on content governance as much as search technology. See [Memory in AI Agents](/memory-in-ai-agents/) for how external knowledge and retrieval differ from model knowledge and agent memory.