When Knowledge Lives Only in People's Heads: Company Knowledge by Chat, with Sources
A knowledge chat is useful only when the sources are right. Why the real work happens before the chat and how operating knowledge becomes a usable system instead of a bottleneck.

Most smaller businesses have one information desk for everything: the owner. Which screws for the facade type X, how was it solved at customer Y, where is the installation guide for the old model. The team asks, the boss answers, and both lose: one his concentration, the other time in which he cannot continue working. When the owner is on holiday, the knowledge stands still.
The problem is old. What is new is that it can be solved without anyone writing a handbook that nobody reads anyway.
What came out of the Lab
Kollege·KI is a public demo of exactly this solution: the team asks questions in plain language and gets answers from the business’s own stored documents. Data sheets, installation guides, internal notes, project documentation. The answer arrives in seconds, and it arrives with a source citation: which document, which passage. The approach now lives on as the “Wissen” module in Werklinie, and the demo remains public.
Two properties decide whether such a system is accepted in daily use. Both are decisions, not features.
The source citation. An AI that phrases freely always sounds convincing, even when it is wrong. That is why the system answers exclusively from the stored documents and shows for every answer where it comes from. Anyone who does not trust the answer clicks the source and reads for themselves. That fundamentally changes the relationship to the tool: it is not an oracle, it is a very fast colleague who always says where he read it.
The own infrastructure. Company knowledge is the most confidential thing a business owns digitally. It does not belong in a third-party chat service. The system runs with European AI on infrastructure the business controls, with its own access for the team and no per-seat subscription. Why this question comes before every AI project, I have written up separately.
The real work happens before the chat
The part vendors like to keep quiet: a knowledge chat is only as good as the knowledge base beneath it. Anyone who dumps folders in indiscriminately, outdated price lists next to current ones, three versions of the same guide, gets confidently phrased answers from outdated documents. The source citation makes that visible, but it does not heal it.
The real work is therefore curating: deciding which documents are the binding truth of the business, sorting out the outdated, spotting the gaps. That is not a technical task but an entrepreneurial one, and it is manageable if you start small: one topic area, the twenty most important documents, then let it grow. The side effect is remarkable: many businesses know for the first time after this exercise where their knowledge actually lives.
The core
Company knowledge that exists only in heads is a business risk with a friendly face: it works until the head is missing. A curated knowledge base with source citations turns it into a business asset that stays, reachable for the whole team, without the boss becoming the information desk. The technology for it is running, publicly viewable. The smart first question is not which tool to buy, but which twenty documents contain the truth of your business.