Products // 2 of 4 · Extract

Kleis

In development

The data you need, from any document.

Kleis recognises what each document is, sends it to the right place, and pulls out the fields, names and amounts your systems need, each one checked and linked to the spot on the page it came from.

What goes in and what comes out

You send

  • Pages read by Grammata
  • Your list of fields, or none

You get

  • Document type and routing
  • Fields, names and amounts
  • Checks that fail, with the reason

Inside Kleis // 4 features

What it does.

Classification & routing

Know what it is before you read it.

Kleis recognises the document type from its text, layout and image together, and routes it to the right workflow: invoice to accounts payable, amendment to legal, register page to the right series. New types are learned from a handful of examples.

Field extraction

The fields you need, with the box they came from.

Kleis pulls the fields a workflow needs (supplier, totals, dates, account numbers, dosages, parties) into typed, normalised values. Every value carries a confidence and the exact region it was read from, so a reviewer checks the source in one click.

  • Typed and normalised. Dates, amounts, currencies and identifiers come out in one format, whatever the layout they came from. Cross-modal document understanding · PhD thesis 2022
  • Every value cites its box. Provenance from field to region and version, so extraction can be audited.
  • Templates learned, not configured. Recurring supplier and form layouts become schemas in memory, so the next one is read against them.

Names & entities

Who, what, where and when, across documents.

Kleis finds people, organisations, places, dates, amounts and domain entities such as drugs or contract parties, and links each mention to the same entity across every document it appears in, building a graph your other systems can query.

  • Entities in context. Entity extraction tuned per domain; our MTRACE study ablates its effect on downstream retrieval. MTRACE · arXiv 2025
  • A graph, not a list. Mentions are linked into a relation graph that persists across sessions and can be rendered back into text. KG-to-text with LLMs · arXiv 2026

Consistency & ID checks

Is this document what it claims to be?

Kleis checks capture quality, the presence of the elements a document type requires (photo, signature, stamp), and internal consistency: do the line items add up to the total, does the closing balance follow from the transactions. Failures go to review with the reason attached.

  • Identity document quality. Deep quality detection for identity document images, with bandpass filtering. IDTrust · WACV Workshops 2025
  • Faces on camera-captured IDs. Face detection evaluated on camera-captured identity documents under complex backgrounds, lighting and defocus. Faces on ID documents · ICDAR Workshops 2019
  • Numbers that add up. Totals, balances and cross-field rules checked on the extracted values, each check pointing to its boxes.

See it work // illustrative

Kleis, on fictional documents.

The demo runs in your browser on fictional documents. It shows what Kleis returns; it is not production software and it measures nothing.

Pipeline · one document through the chain Illustrative
Supplier invoice Fictional through the chain
StageProductWhat it outputs
01 IngestGrammata · Capture & quality checks Scanned PDF, 1 page, capture quality 0.96: accepted
02 ReadGrammata · OCR & handwriting Printed French, Latin script; 3 words flagged below 0.90
03 LayoutGrammata · Layout & tables 13 regions in reading order; line-item table rebuilt as 5 rows × 4 columns
04 ClassifyKleis · Classification & routing Supplier invoice (0.97), routed to accounts payable
05 ExtractKleis · Field extraction 13 fields normalised; 1 below 0.90 held for review
06 EntitiesKleis · Names & entities 9 entities, 5 linked to known records
07 VerifyKleis · Consistency & ID checks 5 checks: 4 pass, 1 warning
08 AskSibyl · Questions & answers 3 answers with citations, 1 explicit abstention
09 SummariseSibyl · Summaries 3 sentences, each citing its regions
10 TranslateSibyl · Translation 2 regions, French → English, terminology kept
11 RememberMnemos · Document memory 4 claims written with provenance; matches a known supplier template
Illustrative: scripted outputs on fictional documents. Confidences and scores are examples, not measurements.

Try Kleis on your documents.

We start with a pilot on a sample of your own documents, measured before anything goes to production.