Can you import our FlexiCapture document definitions?

Not today. Field sets, zones and validations are rebuilt here — the same work the Vantage path asks of you. What shortens it is blueprint provisioning and model-assisted template authoring: drop in a sample page, get a draft template with zones and data types, correct it, train it.

Is Tesseract going to lose against the ABBYY engine?

On degraded scans and dense forms, yes, measurably. That is why the hybrid path exists: Tesseract first, low-confidence words or pages re-read by a vision model. Settle it with a bake-off on your own documents — we publish the fields we lose.

We run ICR-heavy forms. Should we switch?

Probably not on that use case alone. Handwriting runs through the model path here, which carries a cost per page and a latency the traditional engines do not. We price it honestly rather than pretending the gap is closed.

Vantage has pre-trained skills. What do we get on day one?

A blueprint — finance, accounts payable, insurance, KYC, anti-money-laundering or supply chain — that provisions the project, workflow, routing, upload endpoint, export goal and a full field set in one action. That is a pre-authored field set, not a pre-trained model: you still author the template, with a model drafting it from a sample page. We would rather state that plainly than lose the pilot.

Can we run entirely offline?

Yes. Tesseract for text and a local GGUF vision model through a managed llama-server cover recognition with no outbound network. Vantage cannot run offline.

How does the cost curve actually compare?

Below roughly 100–150k pages a year, a metered platform is often cheaper than running infrastructure — we say so. Above it the gap widens every year, and for a service provider with twenty clients on one install it is not close.

Can you match its pages per hour?

Not as a blanket claim. Throughput here scales by adding nodes that declare the workflow duty, and the honest answer for any given box is a benchmark on your hardware and your paper. We run it before quoting volume.

Is your PDF/A really PDF/A?

The renditions assert PDF/A-2b with an sRGB output intent and PDF/A identification in XMP, built with PDFBox from stored page images and persisted word geometry. Archive buyers audit that claim, and we encourage validating a sample set with veraPDF as part of acceptance.

We already run TotalAgility for process automation. Why change capture?

You may not need to. The strongest shape is often capture here, feeding your existing orchestration through REST, JDBC, message bus or file drop — you keep the process estate and stop paying capture volume fees.

Does anything replace VRS?

The pre-processing chain covers deskew, despeckle, binarize, crop, line and punch-hole removal and colour dropout, ordered per project. VRS remains the benchmark on hostile paper; on ordinary departmental scanning the difference rarely survives a bake-off.

Do we have to draw a template for every supplier?

No. A template carries alternatives for layouts that differ, label search finds fields that move, and the model paths read documents you have not templated at all. Where a template does earn its keep is auditability: the same input produces the same output, and the field says which method produced it.

Can you learn from corrections the way Rossum does?

Templates train from operator corrections, and the verification hub's word-group mode applies one correction everywhere the same misread text appears. It is a different mechanism from a continuously retrained global model, and it stays inside your tenant.

We are happy with ChronoScan. When should we move?

When a second operator, a second client, an audit requirement or a compliance question arrives. Those are the four triggers we see — not features.

Is this overkill for one mailroom?

Possibly, and we will say so on the call. A single-machine install runs fine here, but if you will never need users, tenants, workflow or audit, the cheaper tool is the right tool.

Why would we leave a platform that works?

Usually one of three reasons: the renewal, a tenancy requirement Transact cannot meet, or a mandate to run models locally. If none of those apply, staying is a defensible answer and we will say so.

Our developers say they can wire this up in a sprint.

The recognition call, yes. The forty-odd capabilities around it — batch assembly, separator handling, template versioning, validation rules, an operator review UI, retries, retention, audit, tenant isolation, export mapping — are the sprint that never ends. Ask what happens when the person who wrote it leaves.

Can we use Textract as an engine here?

Recognition is pluggable and the plugin SDK exists for exactly this. Most teams end up local for volume and remote for the hard pages.

So you will not quote against ReadSoft for AP?

Not on SAP-native accounts payable with live master-data matching. We will quote for the other 70% of the documents in the building, and we will say the same thing on the first call as on the last.

Interactive demo · fixture data

The actual interface, running on made-up documents.

This is the product's own markup and stylesheets — the same header, sidebar, hubs and field rows an operator sees. The documents are synthetic and nothing is submitted anywhere. Click a field or a zone in the verification hub; open the project configuration modal; switch the app to dark mode from its own toggle.

Or drive it the way an operator does: the top menu switches section, the sidebar walks all thirty-one screens — batches, documents, queues, templates, classifiers, scripts, connections, datasets, import profiles, users, tenants, the cluster and the audit log — and the row buttons drill into the classifier, the alternative designer, a dataset and an import profile's mapping.

KognitCapture

›

Dashboard

Batch throughput and the documents moving through this project right now.

Pages processed

—

last 30 days

Straight-through

86 %

No operator touched the document

Avg verification time

44 s

From queued to approved

Model spend

€44.70

last 30 days · 13 % of pages

Pages per day

22 Jul – 20 Aug 2026

Recent Batches

View All

Name	Status	Documents	Source	Created
MAILROOM-2026-08-20-07	PROCESSING	3	Scanning hub · Reception MFP	20 Aug 2026 08:41
ORDERS-MAILBOX-4412	EXPORTED	58	Email · orders@vandeputte.be	20 Aug 2026 08:05
EDI-ORDERS-DROP-118	EXPORTED	212	Import profile · e-invoice (UBL + CII)	20 Aug 2026 07:30
SFTP-LOGISTICS-0819	VERIFICATION	44	SFTP · logistics-drop/inbound	19 Aug 2026 22:15
API-ERP-PUSH-9902	FAILED	2	REST · api-key “erp-bridge”	19 Aug 2026 18:02

Projects

Working units — each owns its own templates, workflows, scripts, and settings.

Name	Slug	Description	OCR Engine	Language	Active	Created	Actions

MAILROOM-2026-08-20-07

Mailroom · duplex 300 DPI · draft · operator l.peeters

DRAFT

Drop files here

or click to pick. PDF, TIFF, PNG, JPEG, BMP and GIF. Multi-page files are split into pages.

scan-0820-0741.pdf · 12 pages imported

Network scanners registered for this organisation. The server talks to them directly over eSCL, so this machine does not need to be on the same network.

Reception MFP

http://192.168.1.50/eSCL · idle

Mailroom Fi-8170

http://192.168.1.61/eSCL · idle

Resolution

Sides

Colour

Source

Pull documents from an SFTP or FTP server, or attachments from a mailbox. The connection and its credentials are the ones configured under Connections.

Connection

Browse with

Session name

Profile

Mailroom · duplex 300 DPI

Target workflow

Quality

10 pages clean · 2 flagged: skew 3.1°, dark edge

Assembly cuts the page stream into documents. Your own separator marks always win; the profile's split rules only get a say when you have marked nothing.

3 documents · 12 pages · split on barcode separator

PO-2026-04417.pdf NEEDS VERIFICATION

Supplier invoice · template “Customer purchase order v4” · batch MAILROOM-2026-08-20-07

6/11 verified 1 error

Queue 8

Page:

Region preview

Fields 11

Severity	Field	Message

Distributed Processing

Run this installation across several servers, each with a declared set of duties.

Distributed processing

Six servers are claiming work from the queue. Each takes only the duties it was assigned.

SCALED OUT

Web interface

Two servers answer the browser behind the load balancer; sessions and login throttling are shared.

Public address

https://capture.vandeputte.be

Cluster-wide

Forwarded-for header

Trusted from 10.4.0.0/16

Queued

Oldest waiting 42 s

Running

across 6 node(s)

Completed / hour

4 180

last 60 minutes

Failed

gave up after all attempts

Nodes

6 registered

Name	Role	Duties	Capabilities	Status	Jobs	Heap	Version	Last heartbeat	Actions

Scheduler leases

Exactly one node performs each of these, whichever holds the lease.

Lease	Held by	Acquired	Renewed	Expires

Jobs

Type	Duty	Capability	State	Step	Attempts	Node	Duration	Error	Actions

Memory Management

Heap, engine pools, and loaded model memory for this instance.

Used Memory

5 204 MB

64% of max

Free Memory

1 884 MB

Total Allocated

7 088 MB

Max Heap

8 192 MB

GC Settings

Saved on this instance — no restart needed

Automatic GC

Collect when heap use passes the threshold

Log heap statistics

Write a usage line to the log on every check

Threshold

% of max heap

Automatic GC runs when heap use reaches this level.

Schedule

Every minutes

GC Collectors

Name	Collections	Time (ms)
G1 Young Generation	18 402	212 884
G1 Old Generation	44	9 118
G1 Concurrent GC	1 260	31 042

Memory Pools

Name	Type	Used (MB)	Committed (MB)	Max (MB)	Peak (MB)
G1 Eden Space	Heap	2 048	3 072	—	3 584
G1 Old Gen	Heap	2 904	3 584	8 192	4 102
G1 Survivor Space	Heap	252	432	—	512
Metaspace	Non-heap	318	352	—	318
CodeHeap 'non-nmethods'	Non-heap	18	22	128	19
Compressed Class Space	Non-heap	41	48	1 024	41

GC History

Time	Trigger	Before (MB)	After (MB)	Freed (MB)	Duration	Status
20 Aug 08:40:11	Manual	6 408	5 204	1 204	84 ms	OK
20 Aug 08:12:02	Threshold 80 %	6 611	5 002	1 609	102 ms	OK
20 Aug 07:44:38	Scheduled	6 120	4 880	1 240	77 ms	OK

System Information

Java Version	21.0.4
JVM Name	OpenJDK 64-Bit Server VM
OS	Linux 6.8.0-45-generic (amd64)

Processors	16
Uptime	9 d 04 h 12 m
Start Time	11 Aug 2026 04:29

System Logs

Application log output, streamed live from the running instance.

Log File

Lines

Levels

ERROR WARN INFO DEBUG TRACE

Logging Configuration

Enabled

Logger	Effective Level	Set Level

Size: 184.2 MB Modified: 20 Aug 2026 08:42:11 Showing: 0 / 0 lines

Auto-scroll

DEBUG INFO WARN ERROR

0 entries LIVE

Timestamp	Level	Document	Event	Message	Duration

Built from the application's Thymeleaf templates and its own CSS bundle. Live behaviour — recognition, validation, workflow routing, export — runs on the server and is not part of this page. Ask for a hosted trial on your own documents.