Extraction settings

Extraction unit Output one row per

Document Processes documents page by page and concatenates results in a single row per document. Page Processes documents page by page. Sliding window of pages Processes multiple consecutive pages for better context.

Window size

Number of consecutive pages processed together as a unit.

Window overlap

Number of pages shared between units to preserve context.

The window overlap should be lower than window size. The overlap is automatically set to window size - 1

Customize extraction prompt

Use a custom prompt rather than the default one (can be useful to extract specific information)

Extraction unit Output one row per

Document Section

Max section depth

Maximum depth of sections to extract - deeper sections are considered as plain text.

Image handling mode

Skip non-informative images

Skip images like logos, stamps, or signatures.

VLM

Customize description prompt

Use a custom prompt rather than the default one (can be useful to extract specific information)

Use OCR on PDF and image files

Outputs structure

Embed

Prompt output (chunked if applicable) Extracted text (chunked if applicable) Extracted text and image description for images (chunked if applicable)

Store for retrieval

Extracted text (chunked if applicable) and images Extracted text (chunked if applicable) This output can later be used to augment LLMs in the generated Knowledge Bank

Store images in

{{computable.box.name}}{{computable.localProject ? '' : (' (from project ' + computable.projectKey + ')')}} No selected folder.

An output folder is required to store images extracted from documents with this strategy. Select it from the Input / Output tab.

Advanced update settings

Force reprocessing

Rule edits don't trigger reprocessing on unmodified documents by default. Check to force reprocessing.