Hebrew renders right-to-left, out of the box
Hebrew text now renders in correct right-to-left order with no font to upload — including mixed Hebrew/Latin lines, numbers, mirrored brackets, and right-aligned paragraphs. Copy-paste returns your source text, not the reversed glyphs.
Hebrew PDFs now come out right. Noto Sans Hebrew (regular and bold) joins the built-in font fallback, so a Hebrew document renders correctly on /md, /preview, /render and canvas without uploading a font — and a document with no Hebrew loads nothing extra.
- Correct visual order. Lines are reordered per the Unicode Bidirectional Algorithm: in a Hebrew paragraph the first word sits on the right, mixed Latin and number runs keep their own left-to-right order inside the line, brackets are mirrored, and vowel points stay attached to their letters.
- Right alignment by default. A paragraph whose first strong character is Hebrew is right-aligned; an explicit
alignstill wins. - Logical copy-paste. Selecting text out of the PDF, or extracting it with
pdftotext, gives the source order rather than the visual glyph order — the line carries its logical text as/ActualText, so tagged PDF output stays accessible. - Mixed content works. English words, numbers like
123, and inline markup inside a Hebrew paragraph render in the right places.
Arabic, Persian and Urdu are not supported yet: they need contextual joining forms (OpenType shaping), which the layout engine still lacks, so Arabic text keeps rendering in logical order rather than being reversed unshaped. Hebrew needs no shaping, which is why it can ship first.
Limits worth knowing: one embedding level only (no explicit dir attribute or Unicode isolates yet), and the reorder is line-based across the whole paragraph. Details in the fonts skill and docs/rtl-bidi.md.