Niches · Document parsing & extraction
Extract with page and box references a person can check
Two or three companies' docs show a full fit.
What counts as done
Every extracted field or block carries a page number and a bounding box on that page, table cells carry their own boxes, and a confidence per field, so a reviewer can open the page at the box.
How agents ask for it: show me where in the document this figure came from
, extract the fields with citations
, highlight the source of each value
Does the job 3
The docs show it meets the test above, and nothing we read says otherwise.
| Listing | Why |
|---|---|
| Amazon Textract BB Amazon Web Services, Inc. | Every block, cell and expense field has Page, a BoundingBox and polygon as ratios of page width and height (documented), and a Confidence, so a reviewer can open the page at the box. source |
| Azure Document Intelligence B Microsoft Corporation | Fields, paragraphs and table cells each carry boundingRegions with pageNumber and polygon, and fields carry confidence. source |
| Google Cloud Document AI BB Google LLC | Every entity has a pageAnchor with page and bounding polygon plus confidence, and table cells carry their own layout bounding polygon. source |
Does the job, with a limit 12
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Adobe PDF Services / PDF Extract API C Adobe Inc. | Every element and table cell carries Page and Bounds in documented 72 DPI PDF coordinates, but there is no field extraction and no confidence, so citations are per layout element only. source |
| Datalab B Endless Labs, Inc. | Each field cites block ids such as /page/0/Text/3, and table cells can carry data-bbox, but field citations point to blocks rather than boxes, and per-field confidence is a verification status, not a score. source |
| Docling B Docling Project (LF AI & Data Foundation) | Converted items and table cells carry provenance with page and bounding box (a visual grounding example exists), but template extraction fields carry no box and confidence is per page. source |
| Extend API + MCP B CrowdView Inc. dba Extend | Each extracted field carries citations with page number, page size and polygon, plus confidence; whether table cells get their own boxes was not documented in the pages read. source |
| Invofox D Invofox, Inc. (Delaware), with a registered branch in Spain | geometry.boxes mirrors the data fields with a page index and an eight-coordinate box per value, but the coordinate units are not defined, and confidence is per document. source |
| LandingAI Agentic Document Extraction BB LandingAI, Inc. | Ground maps each extracted value to page numbers and boxes, table cells carry their own normalised 0 to 1 boxes, but extraction returns no confidence per field. source |
| LlamaParse API + MCP B LlamaIndex, Inc. | Extract citations and confidence are optional, and Parse can return word, line and cell boxes, but in Max effort citations are text-level only and granular boxes are rejected. source |
| Mindee API C Mindee SAS (Mindee Inc. outside the Eurozone) | The polygons option adds a polygon per extracted field in normalised page coordinates, with confidence available, but there are no table cells to box. source |
| Mistral OCR API B Mistral AI (RCS Paris 952 418 325) | include_blocks returns paragraph blocks with pixel boxes against page dimensions and block confidence, but annotation fields carry no box and tables have no cell boxes. source |
| Nanonets API + MCP E Nano Net Technologies Inc. | Per-field confidence is returned, but element bounding boxes are documented for parse only, so extracted fields have no box of their own. source |
| Reducto API + MCP B Reducto, Inc. | Extract wraps each value with citations, and parse citations give boxes per word and per table cell in page fractions, but extract-level confidence per field was not confirmed in what was read. source |
| Unstructured API + MCP E Unstructured Technologies, Inc. | Extract field_metadata carries page_number and coordinates, but coordinates can be null, metadata can be absent, and there is no confidence per field. source |
Couldn't check 1
The docs were closed to us, or silent. These carry no mark either way.
| Listing | Why |
|---|---|
| Veryfi API + MCP C Veryfi, Inc. | Whether fields carry page numbers and boxes is not in the SDK README, and the API docs were not read. |
Doesn't do this job 4
Listings in the category that do something else, with the reason.
| Listing | Why |
|---|---|
| Clairic BB Cloudflare, Inc. | Text answers carry character spans into the state, but there are no page numbers or boxes on images. source |
| Nemotron 3.5 Parse B NVIDIA Corporation | Blocks carry normalised boxes and bbox_pixels, but there is no field extraction and no confidence. source |
| OpenDocRouter B LlamaIndex, Inc. | Layout elements carry page-fraction boxes and markdown line ranges, but there is no field extraction to cite. source |
| Xiaomi-OCR-0 C Xiaomi Corporation | Outputs carry no page boxes or confidence. source |
6 listings in the category were added after this reading and aren't marked yet.
What stops agents finishing
- boxes given per block but not per field or cell
- boxes in a coordinate space the docs do not define
Around it
Before: document-extraction.rag-ingest. After: human-in-the-loop.review-output, search.verify-citation.
Where this comes from
Weight 6 of 100 in Document parsing & extraction: 4 of 46 accounts ask for boxes or a check step; the category's buyer checks and Reducto and LandingAI docs give it more weight than the accounts
The same as data: /api/v1/use-cases/document-extraction.cited-extraction.json, or the MCP tool get_use_case with id document-extraction.cited-extraction. Other jobs in this category: /api/v1/niches?category=document-extraction.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.