What is RAG, and why does it need your documents converted?
RAG - retrieval-augmented generation - makes your personal and work files AI readable by splitting them into passages, turning each passage into a vector, and looking up the relevant ones before the model answers.
The problem RAG solves
A language model knows what was in its training data and nothing else. Ask it about your company's refund policy and it will produce something plausible and wrong, because a fluent guess and a correct answer look identical coming out of a model. RAG fixes this by giving the model the actual passage to read before it answers.
Why documents have to be converted first
A model cannot search a folder. Retrieval works over vectors - lists of numbers that place similar meanings near each other - so every document has to be split into passages and each passage turned into one of those vectors. That conversion is the whole of what a RAG converter does, and doing it badly is why many RAG systems return confident nonsense.
Chunking decides more than the model does
The single largest quality lever is how a document is split. A chunk cut mid-sentence produces a vector describing a fragment whose subject sits in the previous chunk, so retrieval returns a passage that matched but cannot answer. Good chunking cuts on paragraph and sentence boundaries and overlaps consecutive chunks, so a fact that straddles a boundary still appears whole somewhere.
What you get out
The practical output is JSONL: one JSON object per line, each with an id, the passage text, its embedding and some metadata. There is no formal standard for RAG corpora - every vector database has its own binary format - but nearly all of them import newline-delimited JSON, which makes it the working standard.