User guide¶
The app has two screens: an input screen and a result screen. This section covers both in detail.
| Page | Covers |
|---|---|
| Anonymizing documents | Input formats, batches, scanned PDFs, what the progress stages mean |
| Reviewing the result | The panels, entity highlights, per-entity corrections, manual redaction |
| Advanced settings | Per-type policy, always-redact and always-keep terms, extra LLM instructions, forced OCR |
| Warnings & validation | What PASS / Prüfbedarf / fehlgeschlagen mean and what to do about each |
| Exporting | Text, redacted PDF, ZIP, and the filename trade-off |
Before you start¶
The result is not proof of legal anonymization
The tool supports a human review; it does not replace one. Read the anonymized output before it leaves your hands, and validate the pipeline on your own documents (Evaluation) before trusting it for a new document type.
Two habits worth forming:
- Check the header chip. It warns when document content will leave this machine, or when an enabled detector is not configured.
- Take
Prüfbedarfseriously. It means the validator found something in the output that looks like it should not be there.
Interface language¶
The interface speaks German, English, French and Spanish. It opens in the language your browser asks for and falls back to German; the globe button in the header switches it at any time, and that choice is remembered in this browser (a UI preference — no document content is ever stored).
That covers everything the app says — including the notices and warnings the
backend produces. What a run writes into the document is a separate setting:
the placeholders ([GEBURTSDATUM] vs [DATE_OF_BIRTH]) follow the output
language of that run, which defaults to the interface language and is fixed
when you press Anonymisieren. See
Advanced settings.
This documentation uses the German labels for anything you click, so the docs and the default interface stay aligned.