A knowledge base is the collection of articles, policies, and product information your support team uses to answer customer questions. The help center pages, the internal FAQ, the policy documents, the product manuals. Some teams publish all of it externally. Some keep half internal and surface the rest to customers as a help center.
For an AI agent, the knowledge base is the source of truth. Every reply is grounded in it. If something is not in the knowledge base, the AI either says "I do not know" or, if poorly built, makes something up.
What belongs in a knowledge base
The honest answer: everything a new hire would need to read to do the job. Returns and refunds. Shipping windows and carriers. Product specs. Account questions. Pricing rules. Common error messages. Onboarding steps. Sales objections and answers. Compliance disclosures. Holiday hours.
What does not belong: anything that gets rewritten weekly without versioning, anything sensitive that should not appear in a customer reply, anything specific to a single customer (those should be in the CRM or helpdesk profile, not the knowledge base).
Internal versus public articles
Most teams split the knowledge base. Public articles live on a help center website, indexed by search engines, written for the customer. Internal articles live behind a login, written for the team, with details about how to handle specific cases, when to escalate, what the policy is in edge cases.
An AI agent should be able to read both. The internal articles guide its decisions. The public articles get linked in citations when the customer might benefit from reading further. This split is one of the underrated reasons AI works well in support: the AI uses internal nuance, the customer sees the public summary.
Maintenance is the hard part
A knowledge base only helps if it stays current. Policy changes, product updates, new edge cases, all need to land in the documentation or the AI repeats yesterday's answer.
The good news: AI assistance shrinks this work. When the AI notices a new pattern of questions it cannot answer, it surfaces the gap. When a policy changes, you update one article and every future reply uses the new version. The maintenance is steady, not heroic.
In Keloa
In Keloa, your knowledge base is the foundation. The AI agent indexes it, retrieves with RAG, grounds replies with citations, and flags gaps when questions come in that no article covers. See the product and how the AI works.