Intelligent Document Platform

From any document to a verified decision.

Cogneris is a document AI extraction API. Classify the document, parse every page and extract the fields you define, with each value checked against your rules and linked to the page it came from.

Median response 150 ms · API p95 321 ms Per-tenant encryption Audit-ready

Get 400 free credits Read the quickstart

No credit card. 400 credits for 14 days. Or talk to our team.

Compliance, with the evidence attached · every claim links to the document behind it · see the evidence →

Document API catalogue

Start with the document. Follow the workflow.

Capture common business records, extract complex packets, and verify identity or claims evidence. Every link opens the API page for that document type.

Capture

Turn high-volume operational documents into structured records.

InvoicesPopular ReceiptsPopular Purchase orders Checks Bank statementsHot Bills of lading Proof of delivery

Extract

Read long, variable, and multi-document files with source evidence.

Payroll and W-2 Tax returns Contracts ACORD forms EOBs Mortgage packets Financial spreading

Verify

Cross-check identity, eligibility, address, and claim evidence.

KYC documentsHot ID cards and passports Proof of address Health insurance cards Insurance claims Claims processing

Modular solutions

Combine products for your flow — a custom document stack.

Smart extraction

Recognizes IDs, dates, totals and tables — with context, not regex. Tenant-configurable classification and zero-shot extraction.

Ask your documents

Natural conversation with PDFs, contracts and reports. Document Q&A is limited release; citation coverage varies by workflow and response.

Automated workflows

Fire webhooks, queue jobs and push extracted fields straight into your ERP, CRM or data warehouse.

Document hub

A single repository with versioning, semantic search and granular permissions per team or client.

Try it

Read the output before you talk to anyone.

Four documents run through the extraction API, with the response we return — every field carrying its own confidence on a 0–100 scale and, where the value was located, the page and bounding box it came from.

vendor_name Acme Industries Ltd. 98
invoice_number INV-2026-0417 99
invoice_date 2026-04-17 97
total_amount 12,480.50 USD 99

{
  "data": {
    "id": "9d8a7c6e-1b2f-4a3d-9e8f-7c6d5b4a3210",
    "metadata": {
      "vendor_name": {
        "value": "Acme Industries Ltd.",
        "confidence": 98,
        "page": 1,
        "bbox": [
          0.13,
          0.14,
          0.67,
          0.17
        ],
        "bbox_confidence": 92
      },
      "invoice_number": {
        "value": "INV-2026-0417",
        "confidence": 99,
        "page": 1,
        "bbox": [
          0.7,
          0.15,
          0.92,
          0.17
        ],
        "bbox_confidence": 95
      },
      "invoice_date": {
        "value": "2026-04-17",
        "confidence": 97,
        "page": 1,
        "bbox": [
          0.7,
          0.18,
          0.92,
          0.21
        ],
        "bbox_confidence": 90
      },
      "total_amount": {
        "value": 12480.5,
        "currency": "USD",
        "confidence": 99,
        "page": 3,
        "bbox": [
          0.7,
          0.82,
          0.92,
          0.85
        ],
        "bbox_confidence": 96
      },
      "line_items": [
        {
          "description": {
            "value": "Annual platform subscription",
            "confidence": 97,
            "page": 2,
            "bbox": [
              0.1,
              0.39,
              1.16,
              0.42
            ],
            "bbox_confidence": 94
          },
          "quantity": {
            "value": 12,
            "confidence": 94
          },
          "unit_price": {
            "value": 249.0,
            "currency": "USD",
            "confidence": 96
          }
        }
      ]
    },
    "createdDate": "2026-05-19T13:42:11Z"
  },
  "meta": {
    "httpStatusCode": 200,
    "messages": []
  },
  "hasErrors": false
}

account_holder Northwind Trading LLC 98
account_number_masked ••••4417 99
period_end 2026-03-31 99
closing_balance 48,210.75 USD 96

{
  "data": {
    "id": "6f1d4b20-7a8e-4c93-b6f2-1d5a8e9c3047",
    "metadata": {
      "account_holder": {
        "value": "Northwind Trading LLC",
        "confidence": 98,
        "page": 1,
        "bbox": [
          0.1,
          0.12,
          0.63,
          0.15
        ],
        "bbox_confidence": 91
      },
      "account_number_masked": {
        "value": "****4417",
        "confidence": 99,
        "page": 1,
        "bbox": [
          0.1,
          0.16,
          0.38,
          0.19
        ],
        "bbox_confidence": 92
      },
      "period_end": {
        "value": "2026-03-31",
        "confidence": 99,
        "page": 1,
        "bbox": [
          0.1,
          0.2,
          0.38,
          0.22
        ],
        "bbox_confidence": 91
      },
      "closing_balance": {
        "value": 48210.75,
        "currency": "USD",
        "confidence": 96,
        "page": 6,
        "bbox": [
          0.62,
          0.78,
          0.92,
          0.81
        ],
        "bbox_confidence": 92
      }
    },
    "createdDate": "2026-05-19T13:44:08Z"
  },
  "meta": {
    "httpStatusCode": 200,
    "messages": []
  },
  "hasErrors": false
}

employee_name J. Okafor 96
employer_name Cascade Legal Services 98
pay_period_end 2026-04-30 99
gross_pay 5,980.00 USD 94
net_pay 4,312.08 USD 97

{
  "data": {
    "id": "1e7c9a34-5b62-4f80-a1d9-3c6e7b2a4058",
    "metadata": {
      "employee_name": {
        "value": "J. Okafor",
        "confidence": 96,
        "page": 1,
        "bbox": [
          0.1,
          0.18,
          0.44,
          0.2
        ],
        "bbox_confidence": 90
      },
      "employer_name": {
        "value": "Cascade Legal Services",
        "confidence": 98,
        "page": 1,
        "bbox": [
          0.1,
          0.11,
          0.56,
          0.14
        ],
        "bbox_confidence": 93
      },
      "pay_period_end": {
        "value": "2026-04-30",
        "confidence": 99,
        "page": 1,
        "bbox": [
          0.1,
          0.26,
          0.42,
          0.28
        ],
        "bbox_confidence": 93
      },
      "gross_pay": {
        "value": 5980.0,
        "currency": "USD",
        "confidence": 94,
        "page": 1,
        "bbox": [
          0.7,
          0.51,
          0.92,
          0.54
        ],
        "bbox_confidence": 89
      },
      "net_pay": {
        "value": 4312.08,
        "currency": "USD",
        "confidence": 97,
        "page": 1,
        "bbox": [
          0.7,
          0.56,
          0.92,
          0.59
        ],
        "bbox_confidence": 92
      }
    },
    "createdDate": "2026-05-19T13:45:31Z"
  },
  "meta": {
    "httpStatusCode": 200,
    "messages": []
  },
  "hasErrors": false
}

full_name MARIA T. ALVES 97
document_number X1234567 95
date_of_birth 1991-08-14low confidence 71
expiry_date 2031-02-28 93

{
  "data": {
    "id": "4a2f8c16-9d73-4b50-a6e1-2c7d5f8b3094",
    "metadata": {
      "full_name": {
        "value": "MARIA T. ALVES",
        "confidence": 97,
        "page": 1,
        "bbox": [
          0.16,
          0.16,
          0.56,
          0.19
        ],
        "bbox_confidence": 93
      },
      "document_number": {
        "value": "X1234567",
        "confidence": 95,
        "page": 1,
        "bbox": [
          0.16,
          0.21,
          0.41,
          0.24
        ],
        "bbox_confidence": 89
      },
      "date_of_birth": {
        "value": "1991-08-14",
        "confidence": 71,
        "page": 1,
        "bbox": [
          0.16,
          0.26,
          0.38,
          0.29
        ],
        "bbox_confidence": 66
      },
      "expiry_date": {
        "value": "2031-02-28",
        "confidence": 93,
        "page": 1,
        "bbox": [
          0.16,
          0.31,
          0.38,
          0.34
        ],
        "bbox_confidence": 85
      }
    },
    "createdDate": "2026-05-19T13:47:02Z"
  },
  "meta": {
    "httpStatusCode": 200,
    "messages": []
  },
  "hasErrors": false
}

These are pre-processed sample documents, not a live upload — the values and confidences above are fixed. To run your own documents, use the quickstart or talk to our team. The response envelope is the one documented under extraction.

Walkthrough

The same run, recorded end to end.

The samples above are fixed. The walkthrough is a screen recording of the product doing the work on a real document — including the field it was not sure about, which is the part a scripted demo leaves out.

The recording is not published yet, so there is nothing to play here. We would rather leave the space empty than fill it with a reel that is not the product.

Upload — a real invoice or bank statement, running at the speed the platform actually processes it.
JSON output — the response envelope, with each value's bounding box drawn back onto the source page.
Confidence — the per-field scores, including the one field the model hesitated on.
Flagged field — the low-confidence value marked in place, with the region it was read from highlighted on the page.
Human review — the same document waiting in a workflow’s review queue, a reviewer correcting the flagged value, and the change recorded against them.
Webhook — the signed delivery landing, carrying the document id from the first step.

For developers

One POST, one file, structured JSON back.

The extraction endpoint takes multipart/form-data and a bearer key. No SDK is required and no schema to declare — the platform picks the template from the document itself. The per-language samples below are an optional convenience, not a dependency.

Endpoints answer in a median of 150 ms, p95 300 ms, measured 2026-10-29T14:35:00Z to 2026-10-30T14:30:00Z across 19 components in the EU production region only. That is endpoint response time, not how long an extraction takes — document processing moves with size and profile. How it is measured →

              
              POST /Document/extraction
              
            

export COGNERIS_API_KEY="xtkt_live_…"
export COGNERIS_BASE_URL="https://api-eu.cogneris.ai"

curl -X POST "$COGNERIS_BASE_URL/Document/extraction" \
  -H "Authorization: Bearer $COGNERIS_API_KEY" \
  -F "file=@./invoice.pdf"
            

import os
import requests

with open("invoice.pdf", "rb") as document:
    response = requests.post(
        f'{os.environ["COGNERIS_BASE_URL"]}/Document/extraction',
        headers={"Authorization": f'Bearer {os.environ["COGNERIS_API_KEY"]}'},
        files={"file": document},
        timeout=60,
    )

response.raise_for_status()
print(response.json())
            

import fs from "node:fs/promises";

const form = new FormData();
form.append("file", new Blob([await fs.readFile("./invoice.pdf")]), "invoice.pdf");

const response = await fetch(`${process.env.COGNERIS_BASE_URL}/Document/extraction`, {
  method: "POST",
  headers: { Authorization: `Bearer ${process.env.COGNERIS_API_KEY}` },
  body: form
});

console.log(await response.json());
            

The response envelope — data, meta, and a per-field confidence — is the one shown in the sample output above. Full walkthrough in the Quickstart; every endpoint and status code in the OpenAPI spec.

API & webhook samples

Standalone examples you can paste into a project, next to the published SDKs for TypeScript, Python, .NET and Java — we ship the ones we actually maintain.

Language support, measured

Any language works today — plain HTTP with multipart uploads and a bearer key, nothing stack-specific.
Published packages — SDKs on npm and PyPI (0.2.0) and on NuGet.org and Maven Central (0.1.0); nothing on pkg.go.dev or pub.dev.
Generate your own client — four generate commands run against our spec, and the output compiled.
No mobile client library — keep uploads server-side and call the REST endpoint from your backend.

Read the full language support matrix →

Prefer a ready-to-import workspace? Download the Postman collection or browse the public examples repository.

Need another stack? The API is plain HTTP with multipart uploads — the OpenAPI spec generates a client in any language you already use.

Model Context Protocol

Hand your documents to an agent.

Cogneris runs an MCP server. Point Claude Desktop, Cursor or your own client at it and an agent can list extraction templates, upload a PDF, run a fill job and read usage — through the same tenant, the same quota and the same audit trail as the REST API.

53 tools, 5 scopes — 21 read, 10 write, 22 with a scope of their own, enforced per call
OAuth 2.1 with PKCE — no long-lived key in a config file
Regional hosts — mcp-eu and mcp-us, matching your tenant

Read the MCP guide

.cursor/mcp.json

{
  "mcpServers": {
    "cogneris": {
      "url": "https://mcp-eu.cogneris.ai/"
    }
  }
}

ReAct architecture

Agents that reason before they act.

Instead of one monolithic model, Cogneris orchestrates specialist agents in a ReAct loop (Reason + Act). Each agent has a clear scope, dedicated tools and step-by-step audit trails — critical for regulated industries.

Full traceability per document
Granular reprocessing per step
Auditable cost per run

cogneris.agents.pipeline

Orchestrator

receives the document

Classifier Agent

identifies the type

Extractor Agent

structures the data

Validator Agent

checks the rules

Fraud Agent

scores the risk

Output

JSON + score + webhook

Who uses Cogneris

Built for teams that can't afford a wrong extraction.

Cogneris runs in production at banks, accounting firms, real-estate platforms, and embedded-finance products — wherever the cost of getting a document wrong is paid in audits, regulator fines, or lost revenue, not just in engineering time.

Banking and lending

Loan applications, KYC packages, bank statements, and tax forms — extracted and cross-validated for credit decisioning. SOC 2 and GLBA-aligned controls. Banking →

Accounting and tax

1040 source documents, invoices, receipts, and bank statements parsed for tax prep, AP automation, and monthly reconciliation. Accounting →

Real estate

Closing packages, title commitments, deeds, and appraisals — every field captured with full provenance for underwriting and closing. Real estate →

Embedded finance

Merchant KYC, statement underwriting, and counterparty risk reviews — wired into platform decision engines at scale. Embedded finance →

Verified output

Can a model invent a number?

It can. Model output is not inherently deterministic, so Cogneris does not treat a generated value as trusted just because it looks plausible. Each field carries confidence and can move through an evidence-and-control path before your workflow accepts it.

Field-level confidence — one uncertain total can be reviewed without blocking every reliable field in the document.
The document stays attached — source page and region citations let a person or system check where a value came from.
Acceptance is configurable — deterministic rules, model agreement, and a manual review step can each be required before a critical value is accepted.

Read the anti-hallucination architecture →

Candidate value

8,420.00 USD

0.94 confidence

01Source citation
Page 2 · total due region
02Deterministic validation
Currency, subtotal, tax, and total agree
03Multi-model consensus
Configured models return the same value
04Human review
A step you place in the workflow; the run waits in a review queue
05Audit trail
Record model version, prompt and response hashes, checks, and reviewer in tamper-evident audit metadata with contract-specific, tenant-configurable audit retention

Compliance

Every claim, with the evidence attached.

Cogneris is not certified against any standard yet, and we will not imply otherwise. What we can show you today is published, contractual, or in progress — and every badge below opens the document that backs it, not another page about how much we care.

Certified — nothing yet
Published — the document is on this site
Contractual — a term of the agreement
Unverified — pending compliance review

Published

Sub-processor disclosure

Every third party that touches customer data — named, with purpose, legal entity, and processing location. Customers get 30 days' notice before any addition, with a right to object.

Data Protection §2 → Contractual

GDPR — DPA and EU SCCs

Cogneris processes as an Art. 28 processor — the DPA incorporates the EU Standard Contractual Clauses for transfers out of the EEA and binds every sub-processor to equivalent terms.

Data Protection §1 → Contractual

CCPA / CPRA — service provider

Service provider under Cal. Civ. Code §1798.140(ag) — the eight commitments §1798.140(ag)(1) requires are published in full on this site, not summarized into a logo.

Data Protection §1.4 → Published

Data subject requests

A route you can actually use — access, deletion, correction, and portability, with a 30-day SLA. Where the data belongs to a customer, we route the request back to that controller.

Privacy request form → Published

Measured availability

Measured, not asserted — the figure is scripted from the public status feed, committed to the repository, and a build check fails if the page and the measurement disagree. It is history, not the contractual SLA.

How the number is produced → Published

Control inventory

The control titles our program tracks — published by category in the Trust Center. Per-control status and evidence go to enterprise customers under NDA through the report request flow.

Trust Center controls → In progress · unverified

SOC 2 readiness

Readiness work is in progress — we do not claim an issued report here. The status wording is pending confirmation by the owner of our compliance program before it is treated as final.

Trust Center roadmap →

Nothing on this page is a certification. For encryption, tenant isolation, and incident response, read the security overview; for the full sub-processor table, control inventory, and evidence requests, use the Trust Center.

Evidence in context

Illustrative scenarios.

These illustrative scenarios describe possible workflow designs. They are not benchmarks or production-wide performance claims.

Cascade Legal

A scenario for reviewing legal-service records with a traceable extraction path.

Meridian Credit

A scenario for structuring lending documents before downstream review.

Northwind Health

A scenario for organizing prior-authorization documents for a care workflow.

Capacity at a glance: the application interface ships in 15 languages, and public synchronous extraction endpoints accept files up to 10 MB. Document types, extraction fields, currencies, and page ceilings depend on the tenant schema and processing route. See the source-backed coverage matrix.

From the Cogneris library

Research before you shortlist.

Start with the category, compare the alternatives, then model the economics. These are the guides buyers use to move from research to a defensible decision.

Buyer's guide

What is intelligent document processing?

A practical guide to IDP, OCR, extraction accuracy, integrations, pricing, and vendor selection.

Read the guide → Market report

State of Document AI 2026

Market context, adoption patterns, accuracy benchmarks, and the shifts shaping document AI in 2026.

Read the report → Comparison

Cogneris vs AWS Textract

Compare OCR, forms, tables, validation, workflow automation, pricing, and audit trail requirements.

Compare options → Comparison

Cogneris vs Veryfi

Evaluate receipt and invoice extraction, bank statements, fraud checks, workflow routing, and pricing.

Compare options → Pricing

Document AI pricing comparison

See how page rates, minimums, add-ons, and review costs affect the real price of an extraction workflow.

Compare pricing → Business case

Document automation ROI calculator

Estimate monthly savings, payback period, and 3-year economics using your document volume and review effort.

Model your ROI →

FAQ

Common questions.

How is Cogneris different from generic document AI APIs?

Generic document AI APIs return text or fields. Cogneris returns structured records with per-field confidence, cross-document validation, an audit trail (model version, prompt hash, reviewer ID), and a tenant-scoped permission model. Built for workflows where the cost of a wrong extraction is paid in audit findings or revenue loss, not in engineering time.

What document types are supported out of the box?

Cogneris supports tenant-configurable classification and zero-shot extraction. Teams define the document classes and fields their workflows need, then validate the output against their own representative documents before production use.

How does pricing work?

Credits, at published rates. Every operation spends credits from one pool: a 1-page document classified and extracted uses 4.2 credits. Free gives 400 credits for 14 days with no card, and Pro is €99/month for 5,000 credits. Enterprise contracts include negotiated rates that reflect committed volume. See the full rate card or use the ROI calculator to estimate spend at your volume.

How long does integration take?

Cogneris provides REST APIs and signed webhooks suitable for custom integrations with systems of record. Production timelines depend on the document mix, validation rules, review workflow, and downstream integration.

Where can I learn more about the document AI category?

Start with the IDP buyer's guide — a 14-minute read covering accuracy, pricing, vendor selection, and integration patterns. The 2026 State of Document AI report covers market sizing, adoption by industry, and accuracy benchmarks. The document AI glossary covers terminology — IDP, OCR, HITL, ReAct, audit trail, sub-processor, and the rest.

Get started

Fewer spreadsheets. More data.

Book a 30-minute demo. We'll show you the platform processing your own documents.

Book a demo See pricing