How documents are read and searched
Reading levels, scanned pages, what reading costs, and how agents search a knowledge base.
Every document goes through the same path: it is read into text, cut into passagesPassageĐoạn (passage)A short section of a document, cut along its headings and stored with an embedding. Search returns passages, not whole files.Open in glossary, optionally embedded for search by meaning, and then searched by agents.
Upload a file, or add a SharePoint folder that syncs on a schedule. Each file is stored once, keyed by its content, so the same bytes attached again match the copy already there.
Word, Excel and other typed formats are always read as Text. Going back to a shallower level keeps what was already read and costs nothing.
Reading levels
| ReadingReading levelMức đọcHow deeply the portal reads a PDF or image: Text (text layer and OCR, no model), Text + figures (pictures described), or Full rewrite (a model writes out every page). Deeper costs more.Open in glossary | What it does | Cost |
|---|---|---|
| Text | Takes the text layer and reads scanned pages with OCR. | Free |
| Text + figures | Adds a description of each picture and chart. | One model call per page with a picture |
| Full rewrite | A model writes every page out as Markdown. Makes a file easier to check against its reading; does not improve search. | One model call per page |
Uploads follow the knowledge base’s Uploads default unless you choose otherwise in the upload dialog; SharePoint sources set their own; any file can take its own setting and be re-processed.
Scanned pages and pages of figures
Every page is checked before anything is sent. A page with no text worth the name, and a page whose text layer flattened a table of figures, is rendered as an image and read by OCR, which costs nothing. A model reads a page only under Text + figures or Full rewrite.
- A file needing more model pages than the cap is refused whole.
- One page failing does not fail the document; it is counted unread.
- A stopped run is marked stopped, not failed, and adds nothing to the index.
Semantic search
With semantic searchSemantic searchTìm kiếm ngữ nghĩaFinding passages by meaning rather than exact words, using embeddings. It runs alongside keyword and document-code search, which are always on and free.Open in glossary on, agents find files by meaning as well as by word, and every passage is embedded on your department’s budget. Off, they find files only by the words they contain, and nothing is embedded.
How agents search
Retrieval combines a meaning search with a keyword search, and the keyword search finds document codes, part numbers and revision references.
Who pays
Everything an upload spends is billed to the uploader’s department and counts against its budget. Every run of a SharePoint source is billed to the department of whoever created it. Pages already read are not paid for twice, and a file renamed upstream costs nothing.