A knowledge base is the curated store of documents a team relies on to do its work — the internal wikis, runbooks, onboarding guides, support macros, and policies that hold the answers people keep asking for. It is the shared memory of an organisation, written down so it doesn't live only in someone's head.
Why searching it is hard
The trouble is that having the answer written down somewhere is not the same as being able to find it. A knowledge base grows in layers: pages get added, half-updated, and quietly abandoned. The same topic ends up described three different ways across three different teams. Keyword search rewards exact wording, so a question phrased differently from the document never surfaces — and even when it does, you get a list of pages to open, not an answer to read.
Semantic search helps with the phrasing problem by matching on meaning rather than literal words, but it still hands you documents to sift through. What people actually want is to ask a question and get a reply, grounded in their own runbooks and policies, that they can trust.
How RAG and citations turn it into answers
This is what retrieval-augmented generation is for. Instead of asking a model to answer from memory, RAG first retrieves the most relevant passages from the knowledge base — often after chunking long pages into searchable pieces and reranking the candidates — and only then writes an answer from that retrieved text.
The piece that makes it trustworthy is the citation. When every claim points back to the exact passage it came from, an answer drawn from an internal wiki stops being a confident guess and becomes something you can verify. You see which runbook, which line, and whether it still holds.
Sidenote treats your documents this way: it reads them, grounds each answer in the passages it retrieved, and attaches a citation that scrolls you straight to the source — so your knowledge base finally answers, instead of just storing.