Every company wiki has the same biography. It is born in a burst of good intentions, filled for a quarter, trusted for a while, then quietly demoted the first time it hands someone last year's price list. Within eighteen months the real knowledge base is once again the person who has been there longest, and the wiki survives as a museum with an honest founding date. Whether it lives in Confluence, Notion or a SharePoint site makes little difference; the decay is structural, not a software choice. This post is about why that happens to every wiki, what a knowledge base that stays current actually looks like, and how Naxis Assistant, our private knowledge engine, is built to be exactly that.
Why every company wiki goes out of date
A wiki is a second copy. The knowledge it holds already lives somewhere: the terms live in the signed contract, the prices live in the ERP, the decision lives in the mail thread where it was made, the process lives in the heads of the people who run it. Writing a wiki page duplicates that knowledge, and from that moment the company maintains two versions of the truth, one of which changes on its own and one of which waits for a volunteer.
Three forces then work on the copy, all in the same direction.
- The incentives point the wrong way. Writing a page costs the person who already knows the answer, and pays off later, for someone else. Nobody's job depends on the wiki being right, so coverage follows enthusiasm rather than need.
- Staleness is invisible. A page written yesterday and a page that has been wrong since March look identical. The reader finds out the hard way, once, and then quietly stops trusting all of it.
- Reality outruns volunteers. Every new client, hire, price change and version bump adds maintenance debt to pages someone wrote in a different quarter, and the debt compounds with headcount.
Companies fight this with documentation days, wiki gardener rotas and policies that end with "and update the wiki". These fail slowly, and not because people are lazy. They fail because the design demands permanent, unowned, invisible maintenance, and no process survives demanding that forever.
A wiki is a cache with no invalidation
In engineering terms the diagnosis fits in one sentence: a wiki is a cache of what the company knows, with no invalidation. A cache is useful because it answers faster than the sources it copies, but it is only trustworthy while something updates or expires it when a source changes. In a wiki, that something is a human noticing. Most never do.
Here is the liberating part: the primary sources are not stale. The drive holds the executed contract, the database holds this morning's numbers, the mailbox holds the decision as it was actually worded, the tracker knows who owns what. The company's knowledge is current; only the copy is old. Teams chase "a single source of truth" by building one more place to copy truth into, which is how the problem started. Staleness is not a knowledge problem. It is a copy problem, and the way to fix a copy problem is to stop needing the copy.
What a self-updating knowledge base actually is
A self-updating knowledge base is not a wiki that writes itself. It is a system that answers questions from the primary sources directly, so there is no second copy to go stale.
That sentence hides real requirements, and they are worth being concrete about, because they separate the idea from a toy:
- It has to be connected to everything. Drives, mailboxes, the CRM, the SQL databases, the trackers: knowledge that lives outside the system does not exist to it.
- It has to stay current by itself. Naxis re-sweeps every connected source every fifteen minutes, unattended, so "the latest version" means the latest version, not the latest anyone copied over.
- It has to show its evidence. Every answer arrives with the source attached, so the reader checks the contract, not the software. A knowledge base you cannot audit is a rumour with a search box.
- It has to be allowed to say no. When nothing in the record supports an answer, the honest reply is that it is not in the record, not a fluent guess.
- It has to respect who may see what. A wiki was all-or-nothing, which is why half of what mattered never went in. Answers here are scoped to each person's permissions, so the payroll folder can finally live in the same brain as the holiday policy.
This is what Naxis Assistant is. Each company runs its own private deployment, in the cloud or on its own servers, connected to its own sources, and its people ask questions where they already work, in Teams, Slack or WhatsApp, or in the browser. Nobody writes a page first.
What still deserves to be written
Not everything lives in a system already. A decision and its why, a policy, the runbook for the thing that breaks at 2am: until someone writes those down, they exist nowhere. Keep writing them. Write them as small, dated documents in the drive, where they are indexed within minutes like everything else, instead of as pages in a parallel system that must also mirror prices, owners and versions to stay worth opening.
That is the honest division of labour. People write down the things only people know: decisions, reasons, judgement. The engine holds those next to everything the company's systems already know, and keeps the whole current because it never copied anything in the first place.
The difference is easiest to see with a new hire. The old answer to "where do I find anything" was a wiki tour and a list of people to ask. The new one is a question, asked in plain words, answered in seconds with the source attached, scoped to what they are allowed to see. Nobody updated anything last night to make that answer true.
See every source it connects to
The demo runs on a deliberately messy fictional company, full of contradicting versions and half-finished folders, because that is what real corpora look like. Ask it the kind of question that used to mean twenty minutes of digging, and notice that nobody has ever maintained it.
