Skip to content
Challenge

Build a knowledge base that doesn't go stale.

Building a knowledge base for your company is the easy part. Keeping it current is hard, because nobody has time for it in daily business. An AI agent takes over that upkeep on the side. It finds what is no longer true and suggests what is missing. You decide.

Definition

What is a knowledge base?

A knowledge base is where your company stores what it knows: processes, answers to common questions, decisions, responsibilities. What it looks like depends on who reads it. Customers need something different from your team, and an AI needs something different again.

Three types

What types of knowledge base are there?

For your customers

A help center and FAQ, so customers find answers without opening a ticket. Typical tools are Zendesk, HubSpot or Freshdesk. It fails when the answers no longer match the product.

For your team

An internal wiki with processes, onboarding and decisions. Typical tools are Confluence and Notion. It fails when nobody maintains it and everyone goes back to asking instead of searching.

For an AI

The source a chatbot or agent answers from. The method behind it is called RAG. It fails when outdated content sits in it, because the AI passes it on with full confidence.

The reality

Outdated before it's finished.

The start is full of energy: a project, a new tool, a kickoff. The team structures knowledge, creates pages and maybe even defines review rules, for example that someone confirms every six months that an entry is still correct.

Then daily business kicks in. The handbook waits, because there is always something more important. After a year, part of it is wrong, and nobody knows which part. So the team goes back to asking the colleague who knows. Until that colleague quits.

The facts

Why more reminders don't help

As of November 2026.

  • 19 percent

    That is how much of the work week goes into searching for and gathering information, according to a 2012 McKinsey estimate.

    mckinsey.com (opens in new tab)
  • Email notification

    When the verification of a Notion wiki page expires, the owner gets a message in their inbox and by email. The reminder arrives reliably. It doesn't bring the time for the review with it.

    notion.com (opens in new tab)
The difference

Maintained the classic way or by an agent?

ClassicWith an agent
A person reviews when they find the timeThe agent reviews in every task where it uses the knowledge
By calendar, every few monthsAs soon as something no longer fits
Gaps show up when someone searches in vainThe agent notices a missing answer and suggests an entry
Reminders nobody readsA ready-made change proposal
Pages inside a vendor's toolLinked Markdown files, readable without special tools
Whoever writes approvesThe agent proposes, a person approves
How it works

How does an agent maintain a knowledge base?

The knowledge lives in plain text files in Markdown format. Each file covers one topic, and the files link to each other, much like links in a wiki. You don't need special software for this. If you prefer an interface, open the files in Obsidian. At its core, it is a viewer for exactly this kind of linked Markdown files.

The agent works with this knowledge, for example when it answers a customer request, prepares a quote or writes code. Along the way it notices when an entry doesn't match what it is looking at, or when an answer is missing. Then it proposes the change. A person reads it and approves it.

So upkeep happens where the knowledge is used anyway. Nobody has to block calendar time they end up not having.

Step by step

How to build a knowledge base

  1. Decide who reads it

    Customers, team or AI. That decides language, depth and where it lives.

  2. Start with the most common questions

    What does your team get asked all the time? Those questions are a better start than a perfect structure.

  3. One file per topic

    Short entries that link to each other, instead of one long handbook.

  4. Name an owner

    One person per area who approves proposals. Without an owner, even the best knowledge base gathers dust.

  5. Bring in an agent

    It uses the knowledge at work and flags what is outdated or missing.

Data protection

What about data protection?

As soon as an AI reads your company knowledge, the question is where it gets processed. Copilot, Claude or a model on your own hardware each come with different rules. We sort that out before the first entry, not after. Which rules apply to which tool is covered in our overview of AI and data protection.

If you have no time for upkeep, you don't maintain it. Not even when the tool sends 120 reminders a week.
FAQ

What you usually ask us about knowledge bases.

For upkeep, the tool matters less than you think. Confluence fits if your team already works in Jira. Notion fits if you want wiki, tasks and databases in one place. Both fail the same way when nobody maintains them. Decide first who maintains it and whether an agent helps, then pick the tool.

The tool is rarely the big cost. Markdown files cost nothing, wiki tools usually charge per user per month. What gets expensive is the time for building and maintaining it, and that is exactly what an agent saves. We tell you what it costs for you after a short conversation and an effort estimate.

It takes over most of the work, the decision stays with a person. The agent finds outdated entries and gaps and drafts the change. An owner reads it and approves it. That keeps quality high, and nobody has to block maintenance time they never have in daily work.

No. To start, Markdown files that link to each other are enough. An agent reads them directly, and Obsidian gives you an interface on top. A vector database only pays off once you have a very large number of documents. How that works is explained in our guide to RAG.

Your knowledge base is gathering dust? Let's talk.

We look at where your knowledge lives today and who maintains it. Then we build a knowledge base with you that an agent keeps current, and you run it yourself afterwards. Not a traditional consultancy: we build what works. We give you a price after a short conversation.

Further reading