New: Intake — from document to verified data, with evidence
BiVelio Intake

From document to verified data, with the evidence attached.

Intake receives PDFs, images, email and business files; it extracts the relevant information, links every value to its source, and routes only the exceptions to human review.

Try a sample document View the API
Invoice_2026.pdf
SupplierACME, S.L.99%
Total amount€12,450.0098%
VAT21%97%
Due date30 Sep 2026needs review
3 validated1 to reviewData ready

Every value arrives with provenance, a confidence score and a review path.

Evidence and provenance, in the API too

The differential is not reading the document — it is that every extracted value links to the exact page and region it was read from.

Source document
ACME, S.L.
API response
{
  "supplier": {
    "value": "ACME, S.L.",
    "confidence": 0.99,
    "evidence": { "page": 1, "bbox": [120, 80, 310, 112] }
  },
  "total": {
    "value": "12450.00",
    "currency": "EUR",
    "confidence": 0.98,
    "evidence": { "page": 1, "bbox": [412, 640, 528, 668] }
  },
  "due_date": {
    "value": "2026-09-30",
    "confidence": 0.71,
    "status": "human_review_pending"
  }
}

Illustrative example with fictional data — it does not come from any customer.

What Intake does

One entry layer for everything that arrives unstructured.

Business formats

PDFs, scanned images, email with attachments and office files come in through the same door.

Document classification

Every document is identified by type before anything is extracted: invoice, contract, deed, application.

OCR and extraction

Relevant fields are extracted with a confidence score per field, not as one block of text.

Deterministic validations

Formats, sums, dates and identifiers are checked with rules, not with another model call.

Human review

Only exceptions reach a person: validated data moves on without waiting for anyone.

Field-level evidence

Every value keeps its source page and region, so any figure can be audited later.

For developers

Built to integrate, not to replace your systems.

Async API and webhooks

Send the document, receive the result when it is ready, and subscribe to pipeline events.

Multi-tenant isolation

Each organization's data lives and is processed in isolation from the rest.

Privacy and Savings built in

Sensitive information is protected and every model call is optimized before it runs.

Known limits and benchmarks

We publish no speed, accuracy or capacity figures until there is a reproducible benchmark describing hardware, dataset, document type, resolution, batch size, latency percentiles and error rate.

Until then, the best proof is your own case: ask for a run on a real document and review the result field by field.

Start with one document, not a full transformation.

Process a sample document with our team and decide with the result in front of you.

Talk to an engineer Read the documentation