Tiny Online Tools
What this tool actually reads
Most "PDF to JSON" converters give you a flat wall of text. A document is not flat: it has headings, paragraphs, lists, tables and figures, and losing that hierarchy is what makes the output useless for indexing, migration or feeding a retrieval pipeline. This tool reconstructs the hierarchy — and, more importantly, tells you where each answer came from.
Source 1 — the tagged structure tree
An accessible PDF carries a structure tree: the producer wrote down, explicitly, that this run of glyphs is an H2 and that one is a P, with figures carrying alternative text. When it exists, that tree is the document's own statement about itself and this tool uses it verbatim, including the heading levels and any /Alt text on figures. Pages read this way are labelled tagged.
Source 2 — layout inference
Most PDFs have no such tree. Then structure has to be inferred from geometry:
- The modal line height — the size most of the characters on the page are set in — is taken as the body text size.
- A line meaningfully taller than the body is a heading, and how much taller decides the level.
- A line that begins with a bullet or a
1.marker and is indented past the body's left edge is a list item; consecutive ones become one list. - Consecutive body lines with a normal line gap merge back into one paragraph, so a sentence split across four lines arrives as one sentence.
- Tables come from the table detector, and the lines inside a detected table are removed from the flow so the same content is not emitted twice.
Pages read this way are labelled inferred, and that label is not decoration. Inference gets a two-column layout, a running header or a pull-quote wrong from time to time, and you should check the outline against the page preview before trusting it.
Using the output
Every block carries its bounding box, so you can map a JSON node back to a place on the page. Click any row in the outline to highlight it in the preview, or click a highlighted region to jump to its row. The JSON can be copied or downloaded whole.
Privacy
The file never leaves your device. Parsing, structure recovery and export all happen in the browser, with no network requests during processing.