Knowledge base software for work you have already done once

Whoever fixed it knows the fix. They wrote it down nowhere, because at that minute it was obvious. Six months later the same fault turns up at another customer, and that person no longer remembers, or no longer works with you.

An article, not another folder of files

An article is a title, a body, a language and free tags. That is all. No folder tree, no taxonomy to invent in advance: order comes from search, not from structure.

An article born out of a task remembers which task it came from. From there the reverse direction opens: open the article and see the real case it was written from, not only the conclusion.

Deletion is soft. A deleted article leaves the list and the search but does not vanish from the database: «we were wrong, put it back» is a reasonable request, and a knowledge base that erases permanently teaches people not to write into it.

It enters search, not only a list

An article does not stay as text in a list. It is cut into fragments and enters meaning-based search, so it is found even when the question used entirely different words.

Only what changed is re-indexed: a fingerprint of the content is kept per article, and one nobody touched is not processed again. Without that every run would cost money for articles that had not moved.

An ordinary question, not a keyword search

A chat sits on top of the knowledge. You ask in your own words («how do we refund a payment that got stuck»), and the answer is assembled from the fragments found, with references to the source.

The answer comes back in the language of the question. Hebrew, Russian and English in one base: the person who writes and the person who asks are not always the same one.

Not only articles: the chat reads everything there is

Knowledge is not only what somebody wrote on purpose. Alongside articles, the material that appears in the course of work enters search too: tasks, email threads, attachments, messaging, time records and commits. A question finds what is relevant regardless of which screen it was originally written on.

And every answer shows its sources. An answer without where it came from is not knowledge but a convincing guess, and the difference between help and risk sits exactly there.

A completeness check: what exists and what actually got in

After every run, what exists in the system is compared with what sits in search, per kind. That makes «40 articles, 37 in search» visible immediately, rather than discovering six months later that part of the material never got in at all.

The customer too, but not all of it

The same chat has an entrance in the customer portal, and it sees less. Every knowledge point carries who it is meant for, and the customer's search is filtered by that.

The default is «internal». A new kind that was not explicitly recorded as shared does not open to the customer, because the cost of the mistake is not symmetric. Forget to open it and the customer missed something useful; open too much and they saw something internal, and that cannot be taken back.

Who writes, and who can reach it at all

Writing belongs to the provider's side: the screen where an article is written does not open to the customer at all. That is not a polite restriction but a screen with no door on that side.

And an article has a language of its own rather than «the system language». Whoever fixed the fault writes in the language they think in, and translating at question time is better than an article never written because it had to be written properly.

Files, not only text

Files can be attached to an article and are kept beside it. A screenshot of the correct setting is sometimes worth more than a paragraph describing it in words.

Here too the number today is zero, and there is nothing to dress up.

And what is actually in there today

Nothing. Zero articles and zero files on the live project: the mechanism is written, tested and standing empty.

From which a simple consequence follows: the chat today answers from the other material: tasks, email threads and messaging, and not from articles. That is useful, but it is not the same thing. A knowledge base starts working with the first article somebody bothers to write, and no mechanism does that for you.