API

All callable SQL lives in schema pageindex. Source of truth: sql/pg_pageindex--1.0.sql.

Ingest

Function	Returns	Notes
`build_pdf(path text, options jsonb default '{}')`	`jsonb`	C, strict
`build_markdown(path text, options jsonb default '{}')`	`jsonb`	C, strict
`build_from_pdf(...)`	`jsonb`	SQL wrapper → `build_pdf`
`build_from_markdown(...)`	`jsonb`	SQL wrapper → `build_markdown`

options is passed through to the PageIndex bridge; shape depends on upstream PageIndex (see PageIndex docs).

Function	Returns
`tree_search(doc jsonb, query text, options jsonb default '{}')`	`jsonb`
`tree_search_with_preference(doc jsonb, query text, preference text, options jsonb default '{}')`	`jsonb`

These operate on JSON shaped like PageIndex Document.structure (array of root nodes).

Function	Returns
`structure_to_list(structure jsonb)`	`jsonb`
`create_clean_structure_for_description(structure jsonb)`	`jsonb`
`remove_structure_text(structure jsonb)`	`jsonb`
`post_process_tree(items jsonb, end_physical_index integer)`	`jsonb`
`add_node_text(structure jsonb, page_texts jsonb)`	`jsonb`
`add_node_text_with_labels(structure jsonb, page_texts jsonb)`	`jsonb`
`generate_summaries_for_structure(structure jsonb, model text)`	`jsonb`
`generate_doc_description(structure jsonb, model text)`	`text`

page_texts is a JSON array of strings aligned to 1-based page order (index 0 → page 1), per comments in the SQL script.