Automatize
Products · 03 · Intelligent Document Capture
← All products

KognitCapture

Capture, extract and structure data from any document — on your own infrastructure, with no page meter running.

An API-first intelligent document capture platform. Documents arrive from thirteen channels, pass through a sixteen-step image pipeline, are read by local OCR, local models or twelve remote providers, are mapped onto versioned templates, reviewed by people where it matters, and delivered to thirteen kinds of downstream system. Built as a modern replacement for legacy capture suites.

Request a demo →Download brochure (PDF, EN) ↓
632REST endpoints
31Field types
109Agent tools
€0Marginal cost per page
The problem

Document capture is priced by the page and delivered by the quarter.

The incumbent platforms in this category share two properties. They meter you per page, so the more value you extract the more you pay. And they take eight to twelve weeks of specialist consulting to configure a single document type. Meanwhile the cloud alternatives require your invoices, claim forms and identity documents to leave your building — which for a growing number of buyers ends the conversation.

01
Page meters
Volume growth becomes a cost problem instead of an efficiency gain.
02
Forced migrations
Vendors retire the platform you built on and hand you a rebuild project.
03
Data egress
Cloud capture means sending regulated documents to somebody else's tenancy.
04
Consulting dependency
Every new document type is a change request, not a configuration.
The solution

One platform, on your hardware, with the economics inverted.

01
No runtime licence gate
There is no licence server, no entitlement check and no page counter. Run local OCR and local models, and the marginal cost of a page is the electricity to process it.
02
Blueprints, not projects
Six industry starter kits provision an entire working project — workflow, steps, routing, upload endpoint, export goal and a fully fielded template — in one transaction.
03
Nothing leaves the building
Tesseract and local models run in-process; air-gapped operation is a supported deployment, not a workaround. Remote model providers are an option, never a requirement.

Capabilities

01Capture & recognition
  • Thirteen import channels: upload, REST, SFTP, FTP, hot folder, e-mail, Kafka, RabbitMQ, ZeroMQ, webhook, script, plugin, dataset.
  • Driverless network scanning via server-side eSCL discovery, or direct-attach through the browser — nothing to install on the scanning desk.
  • Scan Hub sessions survive a closed laptop: reorder, rotate, delete, rescan, separate, then submit as one batch.
  • Eight recognition engine modes across Tesseract 5, native PDF text, hybrid and open data loading.
  • 129 bundled OCR language packs, warm pooled engines, no external OCR service.
  • Local models in-process plus twelve remote providers behind one interface with failover.
  • Barcode and QR reading on a dedicated decode path — never guessed from OCR text.
02Templates & classification
  • 31 field types and 10 extraction methods — zone, label search, regex, keyword proximity, model, manual, anchor text, companion file, classification, XML mapping.
  • Versioned templates that integrations pin to, so a design change never breaks a live interface.
  • Table and line-item extraction with column zones, row grouping, header detection and continuation across pages.
  • Template generation from a single sample page, and training from operator corrections.
  • Four machine-learning frameworks and 29 model types running in-process — no external inference service.
  • Document splitting on fixed page count, blank page, barcode match, model detection, header/footer or physical separator sheets.
  • Reference datasets for lookup, autofill and validation, plus template alternatives for per-supplier layout variants.
03Workflow & review
  • 16 workflow step types, from import and OCR through verification hubs to export and custom scripts.
  • A visual designer where the drawing is the workflow, with decision gateways on document properties and field values.
  • Three operator hubs — Verification, Processing and Scan — with the page image beside the fields and click-to-fill.
  • Per-field trail: raw recognised value, original value, who verified it and when.
  • 14 validation rule types, from not-empty and ranges through checksum, expression, script, REST call and database lookup.
  • Custom scripts in JavaScript, Groovy or Java, sandboxed, versioned and testable against sample documents.
  • Check-out locking so two operators never touch the same work; public single-document review links for outside parties.
04Delivery & integration
  • Thirteen export destinations and eight formats, including searchable PDF and PDF/A-2b built from persisted word geometry.
  • Field mapping per export goal — unmapped fields are dropped by design.
  • Eleven external connection types, each with a setup wizard and a test action.
  • Encrypted per-project credential vault.
  • Plugin SDK for compiled workflow steps, importers and exporters, with their own migrations and endpoints.
05Built for agents
  • A native Model Context Protocol server with around 109 curated tools across projects, documents, verification, recognition, templates, workflows and administration.
  • OAuth 2.1 with dynamic client registration and an explicit consent screen.
  • Scoped tools with risk tiers, and a separate audit channel that records agent reads, not only writes.
  • Embeddable Verification and Scan Hubs inside your own product, in an iframe, behind an API key, a short-lived session and a named origin.
  • Duty- and capability-aware cluster nodes that claim work without contention — drain a node, resume it, or dedicate it to OCR, export or import.
■Java 25 on Spring Boot 4, a single self-contained JAR with OCR libraries bundled; optional native image.
■PostgreSQL database, 90 versioned migrations, every table and column commented in-database.
■632 REST endpoints across 83 controllers; OpenAPI 3 with a built-in console.
■Files stored on a tenant-prefixed disk tree, never in the database; AES-256-GCM encryption at rest for keys and credentials.
■Model Context Protocol server over streamable HTTP with OAuth 2.1, for AI agent access.
■Self-hosted; no vendor SOC 2 or ISO 27001 attestation, and MFA/SSO are on the roadmap, not shipped.

What it does

  • 01No licence gate, no page counter — the marginal cost of a page is the electricity to process it.
  • 02Six industry blueprints provision a full working project in one transaction, against an industry norm of eight to twelve weeks.
  • 03Local OCR and local models keep regulated documents on your own infrastructure, with remote AI providers as an option, never a requirement.

At a glance

For whomAccounts payable · Insurance · KYC · AML · Logistics
LicenceLicensed per installation, never per page. One fixed annual fee covers unlimited users, projects and organisations inside one tenant, scaling with your own turnover — marginal like income tax, with a €900 minimum and yearly increases capped at 25%. Hosting, AI usage and professional services are billed separately.

KognitCapture in your organisation?

Start the conversation →