Sample tenant app · Document Intelligence

How this works

Choose documents. PDF, Excel (.xlsx) or images. Use test or public documents only.
Submit. This app sends them to the platform on behalf of one company, the tenant shown in the top bar. Behind the scenes it signs in with the product's credential; the browser never sees it.
Wait. The platform reads each document, extracts facts and checks every value against the text it came from. That takes seconds to a few minutes. The first job after a quiet period is slower, because workers start on demand.
Read the result. Every file returns a JSON result: a list of facts (what the value is, where it was found, whether it is accepted, flagged or rejected). See "How to read the result" below.

This page plays the part of a customer system calling the platform's API. A real tenant system does the same calls from its own backend.

1 Choose documents

Drop files here or

PDF, Excel (.xlsx), PNG, JPEG or TIFF

Use case What the documents are, and which data to extract. How to send

Priority

Your reference (optional) A label of your own, such as an order number. The platform stores it on the job and returns it; it does not change processing.

2 Processing

…

Send
Process
Fetch results
Done

File	Size	Status	Pages	Facts

Timeline

3 Results

How to read the result

File and job status

completed: Processed; every fact was accepted.
completed with review: Processed, but at least one fact needs a person: flagged, rejected or unmapped. This is a normal outcome, not an error.
failed: The file could not be processed. The error code says why; there is no result for it.
refused: The platform turned the file away at upload (for example an unsupported type). The other files go on.

Facts

accepted: Delivered; the value was found in the text it cites.
flagged: Delivered but marked for a check (for example a weak match to its source).
rejected: Not delivered as a fact: the value was not found in the text it cites, or it does not fit the field's type. Listed under "Rejected facts" with the reason.
unmapped: A field your data model has no item for yet. Not delivered as an accepted fact.

Text view

Markdown: The document's text as the platform read it, in reading order, one line per element. For scans it is OCR text and can contain errors. It is a derived view; the facts are the contract.
Anchors: Labels like [p3-e4] in front of each piece of text (page 3, element 4; table cells [p3-t1-r2-c1]). A fact's source names the same id, so "in text" on a fact jumps to its line.
JSON with positions: The same text with each element's position on the page: left, top, right, bottom as fractions (0 to 1) of the page width and height, from the top-left corner.

Columns

Code: The use case's standard name for the field (its global code).
Item: Your own name for the field, when your company has a data model. Empty ("—") if it has none.
Value: The normalised value, with its unit. The text exactly as printed is in the JSON (raw_value).
Confidence: Not available yet: the platform does not calibrate it, so it shows "n/a". Use the status instead.
Source: Page and text the value came from. The JSON also has its position on the page.