Every competitor sells a stateless parse.

Send a document, get structure back, pay again next time for the identical work. The five hundredth invoice from the same vendor costs exactly what the first one cost, and is reasoned about from scratch. That is the difference this page is measuring.

A stateless parse

Every page starts from nothing

parsed

docRefinery

Studied once, then remembered

learned

recalled

Illustrative · same sender, same layout

Three pillars, each shipped.

The claim is not that we parse better. It is that we do not parse twice.

PillarStruck is theirs, solid is ours

01Layout memoryRe-reason per requestRecognise, then remember

02Grounded schemaHunt for every field you asked forAnswer only what the page supports; never guess the rest

03Governed contractSchema is a per-call argumentSchema is account configuration — reviewable before the run, auditable after it

The full schedule.

Twelve capabilities across seven products, including the four rows where we are behind.

Capability comparison between docRefinery and seven document extraction products
Capability	docRefinery	Extend	Reducto	DocuPipe	Unstructured.io	Unstract	LlamaExtract	Textract
Memory and governance
Layout memory: Recognises a vendor’s layout and reuses the work already done on it	Shipped	Not offered	Not offered	Not offered	Not offered	Not offered	Not offered	Not offered
Document-grounded answers: Answers only what the page supports; absent fields come back null, never guessed	Shipped	Not offered	Not offered	Not offered	Not offered	Partial	Not offered	Not offered
Review before the run: A person approves what will be asked of a new layout before it runs, once per layout	Shipped	Partial	Partial	Partial	Not offered	Partial	Not offered	Not offered
Field-level governance: The organisation decides what may be extracted, not the caller	Shipped	Not offered	Not offered	Not offered	Not offered	Not offered	Not offered	Not offered
Trust
Confidence you can audit: A score that review can lower and nothing can talk up	Shipped	Partial	Partial	Partial	Not offered	Partial	Partial	Shipped
Human review surface: A threshold-driven queue for a person. Built, not yet shipped	Built, not yet shipped	Shipped	Not offered	Shipped	Not offered	Partial	Not offered	Not offered
Visual citations: Boxes drawn on the page. Built here, not yet exposed	Built, not yet shipped	Not verified	Shipped	Shipped	Partial	Shipped	Partial	Shipped
Coverage and deployment
Custom classifier, trained for you: Types outside the catalogue trained and hosted for the customer	Shipped	Not verified	Not offered	Partial	Not offered	Partial	Not offered	Partial
Self-hosting: VPC and air-gapped on premises, on models that run inside your network	Shipped	Partial	Partial	Shipped	Shipped	Shipped	Partial	Not offered
File-type breadth: Presentations and the long tail of office formats	Partial	Shipped	Shipped	Shipped	Shipped	Shipped	Shipped	Partial
Retrieval pipeline: Chunking, embeddings and connectors into a vector store	Not offered	Not verified	Partial	Not offered	Shipped	Partial	Shipped	Not offered
SOC 2, SSO/SAML, residency: The compliance paperwork an enterprise buyer asks for by name	Not offered	Not verified	Shipped	Shipped	Shipped	Partial	Shipped	Shipped

Compare with

CapabilitydocRefineryExtend

Memory and governance

Layout memory

Document-grounded answers

Review before the run

Field-level governance

Trust

Confidence you can audit

Human review surface

Visual citations

Coverage and deployment

Custom classifier, trained for you

Self-hosting

File-type breadth

Retrieval pipeline

SOC 2, SSO/SAML, residency

Shipped Partial Built, not yet shipped Not offered Not verified

Where we lose, who is closest, and how this was sourced

Where we lose. Five rows are ours to lose, and visual citations are the sharpest: Reducto, Unstract, DocuPipe and Textract draw the box on the page today, and we do not. DocuPipe is strong where trust is bought, with per-field confidence, clickable citations, a visual review interface, and SOC 2 Type 2, ISO 27001, GDPR and HIPAA with a BAA. Unstructured.io owns retrieval pipelines we do not have.

The closest in spirit. Extend markets self-improving document AI, ships evaluation, schema versioning and a review interface on every tier, and publishes a 99.2% accuracy figure. It improves a schema against test sets; it does not remember a layout. Where its capability is not separately verified, the cell says so.

Two different companies. Unstructured.io is an open-source ETL and retrieval ingestion platform; Unstract is Zipstack’s open-source LLM extraction platform with Prompt Studio. The names are one letter apart and routinely confused.

Competitor claims are sourced and dated in our landscape review, and checked against each vendor’s own published material. Re-check before quoting: published pricing and capability move.

Head to head.

Rival by rival, where docRefinery wins. Competitor claims are current as of August 2026 and move with published pricing — re-check before quoting.

REDUCTO · PRODUCT SHEET

Parsing and extraction APIs, priced per page and per field.

Full price for a layout they have already seen
A layout is learned once, then remembered
Charged for fields the document does not contain
Absent fields come back null, never hunted for
Anyone with a key can ask for any field
Your organisation decides what may be extracted

The five hundredth identical invoice costs them what the first did. Ours was learned on the first.

docRefinery

Run the comparison on your own documents.

Send one vendor's invoices through both. The difference shows up on the second document, not the first.

Book a demo How the pipeline works