Product · Extraction
Read off the page, not out of a text layer.
On the way in, every document becomes page images — a born-digital PDF, a spreadsheet, a slide deck and a phone photo of a receipt alike — and those images are what the model reads when a column asks its question. There is no separate text-extraction step standing between the page and the answer.
Free credits on sign-up. No card, no subscription.
| Row | Documents | Registered Proprietor | Encumbrances | Title Defects | Title Plan |
|---|---|---|---|---|---|
| 12 Mill Lane title.pdf | Northgate Property Ltd p.1 | Charge, Right of way p.2 | No p.3 | Title plan, cropped p.4 | |
| Unit 7 Harbour Road title.pdf | Eastdock Holdings Ltd p.1 | Lease, Restrictive covenant p.2 | Yes p.5 | Title plan, cropped p.6 | |
| Wharfside title.jpg | Wharfside Lettings Ltd p.1 | Charge p.1 | No p.1 | Whole page p.1 |
Three steps, and only the first is charged by the page.
| Step | What happens |
|---|---|
| Read once | Each document is turned into page images and indexed as it is uploaded — a PDF, a spreadsheet, a slide deck and a photo alike. That is what ingest is charged for: one credit a page, once. |
| Find the pages | When a column runs, its question is used to find the pages of each row’s documents most likely to answer it. |
| Answer and cite | The model reads those page images, answers in the column’s type, and quotes what it read. A second look locates that quote on the page, so the citation can highlight it. |
Eight kinds of answer.
A column’s type decides the shape its answers take, and the shape decides what you can do with the column afterwards.
| Type | Returns |
|---|---|
| Text | A written answer, with the passage behind it. |
| Number | A bare figure, so the column sorts and filters as numbers. A currency is a display setting, not part of the answer. |
| Yes/No | Yes or no. |
| Date | A calendar date, shown in the format you choose. |
| Category | One of the options you write, and nothing else. |
| List | Several values from the same document. |
| Image | The page, or the part of it, that shows what you asked for. |
| Image list | Several pictures from the document, shown as thumbnails. |
A picture is an answer too.
Ask an image column for a floor plan, a title plan or a photograph of the damage, and the cell holds the page that shows it — normally cropped to the part that matters, or the whole page where that part cannot be located. It still cites the page it came from.
It finds pictures; it does not make or alter them. No new file is created — the crop is drawn over the page image the app already holds. And a picture column costs what any typed column costs.
A picture cannot be corrected with another picture: typing over an image cell replaces it with text, for good.
| What it costs | Credits |
|---|---|
| Reading a page, once | 1 |
| A cell in a typed column | 4 |
| A cell in an image column | 4 |
| A cell in a list column | More |
| Each document past the third in a bundle | +4 a cell |
What it will not do.
- It does not promise to be right. Nothing measures its accuracy, so this page states none. That is why every answer is cited, and why the uncertain ones are listed for checking.
- It does not measure drawings. It reads what a page says. It does not scale a plan or count quantities off one.
- It reads only what you give it. No web search and no registry lookups: if a fact is not in the documents, no cell will claim it.
Extraction is one part of it. Tables are where the answers land, citations are how you check one, and integrations are how documents arrive without anyone uploading them.
Start with the pile you already have.
Signing up comes with free credits. No card, no subscription.