A wiki and a knowledge base can hold the same documents and still do different jobs. A wiki waits for someone to search it. A knowledge base is read by the agent on every question, including inside unattended jobs — so the policy is applied rather than merely available.
Every company that has written its processes down has had the same experience. The document is good. It is accurate, somebody spent a week on it, and it answers the question.
And people still ask the question in Slack.
Not because they are lazy, but because searching costs more than asking, and because they do not know the document exists, and because the answer they need is one paragraph inside a page about something else. The document was available. It was not applied.
That gap is the entire argument for the different shape.
What is a knowledge base?
A knowledge base is a searchable collection of your own documents that agents retrieve for grounded, trusted context. Documents are split into chunks, embedded with a selected model, and indexed for hybrid vector and keyword search.
Practically that means the retrieval is happening on the agent’s side of the conversation. Nobody navigates to anything. The question gets asked in the normal way, and the relevant passage arrives with it.
The difference in one line
A wiki waits for someone to search it. A knowledge base is read by the agent on every question, including inside unattended jobs, so the policy is applied rather than merely available.
That phrase — inside unattended jobs — is the part that is easy to skim past and is doing the most work.
A wiki can only influence a decision when a human decides to consult it. A knowledge base also influences decisions nobody is present for. When a scheduled job runs at 06:00 and drafts a summary, the policy is in the room. There is no version of a wiki that does that.
| Wiki | Knowledge base | |
|---|---|---|
| Read when | Someone searches | Every question, including unattended runs |
| Requires | Knowing it exists | Nothing |
| Returns | A page | The relevant passage, with its source |
| Failure mode | Nobody looks | Stale content is applied confidently |
| Best at | Narrative, onboarding, why | Answering a specific question |
What this changes about writing
Once documents are read by an agent rather than a person, some editorial habits stop paying and others start.
Structure beats prose. Content is chunked, so a page where each section stands alone retrieves well and a page whose meaning accumulates across ten paragraphs does not. Write in units that survive being lifted out.
Say the thing in the section about the thing. A rule mentioned in passing under a different heading is hard to retrieve, because retrieval matches the question against the passage, not against your table of contents.
Delete rather than deprecate. This is the one that changes most. A wiki tolerates stale pages because a human reading them notices the 2023 date. An agent has no such instinct — it retrieves the passage and answers from it. Superseded content in a knowledge base is worse than absent content, because it is applied confidently. Marking a page outdated is not enough; take it out.
Keep the sources you can maintain. Files, crawled websites and connectors to GitHub, ClickUp and Google Drive are all supported, and each one you add is something that can go stale. What documents a knowledge base can read covers the formats and limits.
Keep the wiki
This is not an argument to delete anything.
Wikis are good at things a knowledge base is bad at: narrative, onboarding, the reasoning behind a decision, anything somebody reads start to finish. “Why we price this way” is a document a person should read. “What is the discount threshold” is a question an agent should answer.
The healthy arrangement is both, with the knowledge base holding the material that answers specific questions and the wiki holding the material that builds understanding. Overlap is fine. Contradiction is not — and if the two disagree, the knowledge base is the copy being applied, so fix that one first.
What you get that a wiki cannot give you
Citations. Retrieved passages come back with their source, so an answer can be checked rather than trusted. Why AI answers need citations covers what that changes about adoption.
Scoping. You control which agents read which documents, so a customer-facing agent and an internal one need not see the same material. Controlling which agents see which documents covers the mechanics.
Hybrid retrieval. Both meaning and exact terms, which matters when your organisation has internal names for things — see hybrid vs vector search.
Frequently asked questions
How is a knowledge base different from a wiki? A wiki waits for someone to search it. A knowledge base is read by the agent on every question, including inside unattended jobs, so the policy is applied rather than merely available.
Should we move our wiki into a knowledge base? Move the material that answers specific questions. Keep narrative, onboarding and rationale in the wiki, where a person reads it start to finish.
Does document structure matter? Yes. Content is chunked for retrieval, so sections that stand alone retrieve well and meaning that accumulates across many paragraphs does not.
What happens to outdated pages? They get applied. An agent has no instinct that a page looks old, so superseded content should be removed rather than marked deprecated.
What sources can a knowledge base hold? Uploaded files, crawled public websites, and synced connectors including GitHub, ClickUp and Google Drive.
Can an agent show where an answer came from? Yes. Retrieved passages return with their source document, so an answer can be verified rather than taken on trust.
Takeaways
- A wiki is available; a knowledge base is applied — including inside jobs nobody is watching.
- Write in units that survive being lifted out, because retrieval works on chunks rather than pages.
- Delete superseded content instead of deprecating it. An agent cannot tell that a page looks old.
- Keep the wiki for narrative, onboarding and rationale. Those are read by people, start to finish.
- If the two disagree, fix the knowledge base first — that is the copy being applied.
- Citations, per-agent scoping and hybrid retrieval are things a wiki structurally cannot offer.
Written policy is only worth what gets applied. See how knowledge bases put yours inside every answer.
Sources
Primary sources for the platform rules cited above. Last verified August 16, 2026. Cloud providers change fees, eligibility, and program terms without notice — check the source before relying on a figure.
- Suger Insulin docs: Knowledge Base — That knowledge bases are searchable collections of your own documents agents retrieve for grounded context; chunking, embedding and hybrid vector plus keyword indexing; and the file, website-crawl and connector source types
Stay Updated
Get the latest Cloud GTM insights, product updates, and marketplace strategies delivered to your inbox.