Options: PDF import, OCR and licence

PDF import and OCR

Option	Default	Effect
PdfImportMaxOcrPages	10	pages per PDF that go through OCR; the rest is skipped (latency protection)
OcrResourceFolder	(empty)	the folder holding the Tesseract language data. Empty = OCR off; scans then yield an empty extraction with a notice, while PDFs with a text layer carry on. Environment: OCR_RESOURCE_FOLDER
PdfImportLexiconOverrideEnabled	(not set)	not set = on when OnPremise; off otherwise
PdfImportLexiconOverrideFile	data/pdf-import-lexicon.custom.json	your own field labels as a difference (added/removed per concept and language) from the bundled list

Your own field labels

The vocabulary file is deliberately a difference and not a copy: a product update brings new terms and keeps your additions.

A typo in a concept name is not passed over silently

It lands in the log, and the application carries on with the bundled list.

The concepts can be looked up in the read-only copy Resources\pdf-import-lexicon.json in the target directory; changes to that file have no effect.

The vocabulary is built once at startup

Restart the service after every change.

Optional refinement by a language model

Option	Default	Effect
PdfImportLlmEnabled	false	fills only empty or uncertain fields
PdfImportLlmEndpoint	(empty)	the base URL of the self-hosted model service (Ollama-compatible)
PdfImportLlmModel	(empty)	the model name in that service
PdfImportLlmTimeoutMs	10000	a hard time limit per call; after it, the heuristic alone counts
PdfImportLlmMaxChars	6000	the upper limit on page-text characters passed in

Only against a self-hosted model on your own network

The refinement is only registered if the switch, the endpoint and the model name are all set. Invoice content does not go to third parties.

The PDF engine licence

Option	Default	Effect
GdPictureLicenseKey	(empty)	required for the DOCX-to-PDF step of template creation and for the OCR of scanned PDFs. Empty = both functions off, everything else runs. Environment: GDPICTURE_LICENSE_KEY

The provider is resolved lazily; without a key the licence-free routes remain fully usable: the auto-layout PDF, ZUGFeRD/Factur-X embedding, all format conversions and validation.

Without a key the gap is silent

OCR and template PDFs deliver no result, but also no error that aborts processing — the reason is in the log. Anyone expecting OCR checks there. See Security boundaries.

Values verified against the code on 22 October 2026.