What is a knowledge base?
The organized collection behind every grounded answer, and why the word describes a property of your documents rather than a product you buy.
A knowledge base is an organized collection of information that a tool can search and answer from. In AI work it is the material a retrieval system indexes, so the model answers from documents that stay put, and every claim can point back to the passage it came from. The collection itself is what matters: a folder of sources a tool can search behaves as one, no platform required.
The collection behind the answers.
A knowledge base sounds like software, and the word gets used for products. At its core it is a property of a set of documents: they are organized enough to be searched, and identifiable enough that an answer drawn from them can point back to its source. That is the substance behind retrieval-augmented generation, where the model answers from found passages instead of from memory. Notes are fine, screenshots are fine, PDFs are fine; what makes the collection a knowledge base is that passages in it can be found again and cited when they are.
Why not just ask the model.
A model without sources answers from its training, which stops at a date, never included your files, and offers no way to check anything. Facts inside weights are frozen and unattributable; what training data is explains why. A knowledge base inverts that: the content stays outside the model, current and replaceable, and the question of what an answer rests on has a real answer. It is also what separates this from a chatbot's "memory", which is a summary of past chats that no one can open or verify. When a claim goes wrong in a sourced system, you can find the passage and correct the document, which is the whole advantage.
What retrieval needs from it.
For a knowledge base to serve AI answers, three things have to hold. Sources must stay intact, so the tool knows which document and which passage a claim came from. Search must work by meaning as well as by wording, which is the job of semantic search over an index; what a vector database is covers the machinery, and for a personal library it needs no server at all. And the collection should keep growing, since knowledge bases decay when nothing new goes in. None of that requires a special product, which is why a local tool over a folder of documents is the honest shape for one.