Uploading documents
Uploading is the one step that reads your documents, and the one step charged by the page. Everything after it — building tables, adding columns, rerunning — works against documents that have already been read.
What ingestion does
Drop files into a workspace and each one is read, split into pages, and indexed so that a question can find the passages that answer it. That is charged at one credit per page, once per document — not once per table, and not again when you add a column later.
The consequence is worth planning around: upload everything relevant up front. A workspace holding all forty leases costs the same to read whether you ask one question of them or thirty, and a table you build next quarter will find them already indexed.
Formats
| Format | Notes |
|---|---|
| The main path, including scanned documents. Pages are read as page images, which is what makes a citation a page you can look at. | |
| DOCX and DOC | Word documents, including long contracts with schedules. Legacy .doc is accepted; a page is a page. |
| PPTX | Slide decks. One slide reads as one page. |
| XLSX | Spreadsheets, laid out before they are read — see below. Legacy .xls is not accepted. |
| JPEG, PNG and WebP | Single images — a photographed page, a screenshot, a scanned certificate. One image is one page. |
Audio and video are not supported, and nor is anything else — a format that is not in the list above is rejected at upload rather than accepted and then failed.
Images are a page, not a document
An image is already what every other format is converted into: a page picture. So it skips the conversion entirely and is stored as a one-page document, normalised to the same size and quality a PDF page is rendered at. One image costs one credit to read.
The one difference you will notice is in the citations panel: a page of a PDF can be downloaded as a PDF, and an image cannot, because there was never a PDF to cut it out of. The page picture and the highlighted quote work exactly as they do everywhere else.
A picture can also be an answer rather than only a source. An image column fills its cells with the page — or the part of a page — that shows what you asked for, whether that page came from a PDF, a slide deck or an uploaded image. Nothing about uploading changes for it; the reading is the same reading.
Spreadsheets are laid out before they are read
A slide is a page and a Word page is a page, but a sheet is an unbounded grid with no intrinsic page — so a spreadsheet is rewritten before conversion rather than left to a converter’s default print settings. Each sheet is fitted to one page wide in landscape, its header row is repeated on every page, its name is printed on every page, and a sheet too wide to stay legible is split into column groups that each repeat the label column.
This matters for three reasons that are easy to hit and hard to diagnose: a row torn across two pages loses the thing a spreadsheet means, a page with no sheet name on it cannot answer “which sheet did this come from?”, and an invented pagination inflates the page count — which is both what you are charged and what the per-document limit is measured against.
What a file is doing while you wait
An upload moves through a short set of states, and the panel names each one rather than showing an undifferentiated spinner:
| State | What it means |
|---|---|
| Pending | Uploaded, waiting to start. |
| Queued | Waiting for a free processing slot. Your organisation processes a set number of files at once, by tier, and the rest wait their turn — see tiers and limits. |
| Processing | Being read, split and indexed. |
| Ready | Indexed. A table can ask questions of it. |
| Failed | Accepted but could not be read. See below. |
Queued and Failed are the two worth telling apart, because only one of them needs you. A queued file is waiting and will run — dropping fifty documents at once is a perfectly reasonable thing to do, and the queue is what makes it safe rather than something to avoid. Nothing was refused, and nothing needs uploading again.
Two limits to know about
Pages per document. One ingest job may contain up to a fixed number of pages, which starts at 100 and rises with lifetime spend. A document over the limit does not ingest. This is a cap on how large a single job can get, not a billing gate: every page is charged at the same rate at every tier.
Upload size. The largest single file starts at 10 MB and rises the same way. A scanned PDF hits this well before a text one does. Both numbers are in tiers and limits.
Where your documents live
A workspace has three views — Tables, Bundles and Files — and you move between them with the control beside the workspace name. Files is the whole list; the other two are what you do with it.
The Files view is a table. It carries each document’s name, status, type, size, how many tables use it, and when it arrived, and you can search it, sort any column, filter on any of those six properties, and select rows to act on together. Twenty-five to a page. The two questions it exists to answer are the ones a list could not: where has this file got to, and is anything still using it — the second being the one worth checking before you delete anything, since deleting a document removes it from the workspace and from every row that pointed at it.
The filter offers the values your workspace actually has rather than every value that exists, so a condition you can build is a condition that can match something.
There is also a files panel beside a table, which is the same documents in a narrower column with bundles alongside. It is for working without leaving the table: preview a file, open one full-screen, drag files onto rows or into bundles.
Uploading does not require a table. A workspace can hold documents that no table asks anything of, which is the normal state of a workspace early on.
Dropping a folder
Dragging a folder in uploads the files inside it. Every drop target takes one — the files panel, a bundle, the files view, a table’s rows.
The folder decides which files, not how they are grouped. Ragextract has no notion of a folder: the tree is flattened and the files land exactly as if you had selected them by hand. Subfolders are walked, and a folder dropped onto a bundle joins that bundle.
Two things are done for you, and both are about not paying to read something you did not mean to send. The contents are filtered to the formats above, so a stray .zip or a .DS_Store never reaches ingest — a file you drop by hand is a deliberate choice and still goes through untouched. And a folder drop is capped at 200 files, which is announced rather than applied quietly: if the folder held more, you are told, so nothing goes missing without your knowing.
Documents can also arrive by email
A table can be given a private email address — a mailhook — and attachments forwarded to it become documents in that workspace and a row on that table. Everything on this page applies to them unchanged: same formats, same rate, same states, same page limit. The two differences are that only addresses you authorise may send, and that one email carries about 18 MB in total, which is less than most tiers allow for a single upload.
When a document fails to ingest
Failed documents surface on your home page rather than going quiet in a list — the same place failed and stale cells are collected. The usual causes are a page count or file size over the tier’s limit, a file that is corrupt, or a PDF that is a container for something else. See troubleshooting.
A document that fails has not been indexed, so no table can ask anything of it. Fix the file and upload it again rather than expecting a retry to behave differently.