PdfAI

Lists figures in a tagged document that have no alternate text and proposes a description, or decorative marking, for each (ALT). Descriptions come from ML Kit's on-device image description when PdfAIOptions.context is set and the device has it, else from the custom provider. Nothing is written until apply, which returns a new document.

Answers questions from the document, citing the sentences they came from (ANS).

Compares this document (the original) with a revised version: sentences aligned without a model, and each reworded pair labelled as a change in meaning or wording by the custom provider.

Only the core's public API is used (PKG-4).

Proposes values for the document's form fields from a profile the host passes in (FORM). The synonym list runs on every device; the custom provider matches fields it cannot. The profile is not stored. Nothing is written until apply.

Suggests likely personal data to redact (RED). Patterns run on every device; the custom provider, when configured, adds names, addresses, dates of birth and ID numbers. Nothing is written until apply.

Proposes tags and a reading order for an untagged document (REM), by rules on the page objects, identically on every platform. Nothing is written until apply, which returns a new document; describe its figures with altText afterwards.

Summarises the document or a page range (SUM).