What it means
An AI agent's knowledge base is the curated source of business-specific information it can draw from: services, pricing, hours, policies, FAQs, terms, product specs. The agent uses retrieval-augmented generation (RAG): for each customer question, it searches the knowledge base for relevant documents and uses them as context for its response.
Knowledge bases can be plain text files, structured documents, or live data sources (a CRM table, a Notion workspace, a Google Sheet). Most production agents use a mix.
Why it matters
This is the single biggest determinant of agent accuracy. A model on its own will hallucinate or guess; a model grounded in a clean, current knowledge base will cite real, business-correct facts.
The maintenance discipline matters. A knowledge base full of outdated prices, decommissioned services, or stale policies produces an agent that confidently misleads. Most agent-quality drift traces back to knowledge-base drift.
Example
A car detailing studio's AI agent is told prices have changed. The team updates one Google Sheet that is the connected knowledge base. The agent reflects the new prices in customer chats within minutes. Compare to the old workflow: edit prompt, retest, redeploy, hope nothing broke.