The extraction method to use for processing documents.
legacy extracts text and embedded images in-process without any external model call.
legacy_with_llm extracts the same text, and additionally renders each page to a screenshot that is
described by a vision model — covering content the text layer cannot carry (figures, layout, scanned
regions) at the cost of one model call per page. Both run with no external OCR provider.
The extraction method to use for processing documents.
legacyextracts text and embedded images in-process without any external model call.legacy_with_llmextracts the same text, and additionally renders each page to a screenshot that is described by a vision model — covering content the text layer cannot carry (figures, layout, scanned regions) at the cost of one model call per page. Both run with no external OCR provider.