Measured results
- Text PDFs
- 0.999 mean ink F1 over 68 pages, in Chrome. Firefox 0.999, Safari 0.998.
- Scanned pages
- 0.993 mean ink F1 over 12 scans from 1902 to 1983, rebuilt as live text where the words can be matched closely.
- Unusual PDFs
- 678 files from Mozilla's pdf.js test suite, mean ink F1 0.985 to 0.998 per batch.
How we measure. Each score compares the PDF as pdfium draws it (the engine inside Chrome's PDF viewer) with our HTML as a browser draws it, both at 110 dpi. Ink F1 is the F1 score of dark pixels matched with one pixel of tolerance: 1.000 means every dark pixel lines up. Measured in October 2026 on our test sets; your documents may score differently.
What works today
Working in our builds now, not yet open to the public.
-
Text PDFs, faithfully
Every glyph is placed to a hundredth of a point over a background that keeps the page's own vector art and images. The same HTML renders alike in Chrome, Firefox and Safari.
-
Scans become live text
Scanned pages are read, their typefaces identified, and lines re-set in matching typefaces over the cleaned paper. Whatever can't be matched closely, like handwriting, signatures and stamps, keeps its original pixels. Or keep the scan as it is, with an invisible text layer on top.
-
Structure, not just text
Headings, paragraphs, lists, links, the outline and form fields come through as HTML. Tagged PDFs keep their structure.
-
Exports
Download a self-contained HTML page, or export Markdown, plain text or JSON.
-
Built for hard PDFs
Tested on hundreds of damaged and unusual files from Mozilla's pdf.js test suite: missing fonts, odd encodings, broken structure.
Who it's for
For your own documents, on the web.
- Publishers and archives putting scans online with text people can search and select.
- Developers who want PDFs in their app as real web pages instead of images on a canvas.
- Accessibility: real text for screen readers and translation tools, with the look intact.
- Anyone who wants their own PDFs readable on the web.
Planned
Ideas we're working towards.
Planned Not available yet, not part of any plan, no date promised.
- A developer API for converting documents at scale.
- Accounts that keep your documents and their history.
- An AI reviewer that compares a conversion with the original and fixes what it finds.
- Photographed pages: straighten and flatten a photo of paper before converting it.
- Digital PDFs whose form artwork is a scanned image, converted to live text too.
- Fine-tuning individual pages in the editor.
Join the waitlist.
Email us and we'll write when Sheerpage opens. Tell us what you'd use it for: it shapes what we build first.
We keep your address only to tell you about Sheerpage. See the waitlist privacy notice.