Skip to main content
Manual home

Settings

Select Settings to open Engine & defaults. The settings below are saved as you change them; there is no separate Settings Save button. Resolve any error shown on the page before relying on a change.

Output folder

Under Defaults, find OUTPUT FOLDER and select Choose folder... (shown with an ellipsis in the app). Choose an accessible folder where you can create files.

The default is an Anonymize folder inside your system's Documents folder. Return to Workspace and check the OUTPUT path before running. If a previously chosen folder no longer exists, the app falls back to the default Documents location.

OCR mode

OCR means recognizing text in scanned or image-heavy pages. Under OCR MODE, select the mode to request for subsequent runs:

ControlRequested behavior
OFFSkip OCR. Text present only in scans may be missed.
AUTOUse OCR on pages with sparse or missing embedded text. This is the default.
ALWAYSRequest OCR on every page, which can take longer.

The desktop forwards this choice to the installed engine. OCR availability and results depend on that engine and the document. No mode guarantees detection of every sensitive item. OCR settings do not make standalone image files eligible for the Workspace queue.

Optional Local AI

LOCAL AI is disabled by default. Leave it disabled for the basic workflow. If you already have a suitable Ollama service and model available, enable it and select or type the model tag under OLLAMA MODEL. The desktop passes the configured model and host to the engine.

The model list includes suggestions even when Ollama cannot be reached; seeing a model in the list does not prove it is installed or ready. Do not treat selecting a tag as confirmation that a model has downloaded. The displayed ollama OK row is sample content, not a live health check. There is no host-editing control in the current screen.

The engine preview restricts optional AI detection to a literal loopback endpoint. The default host is http://localhost:11434, which points to this computer. A model can miss or misclassify content; deterministic rules and output review remain required. Only enable this option with a configuration you trust.