Tagged original + clean source. One PDF in, two compliant artifacts out.
The tagged original preserves every pixel of the source — same fonts, same layout, same signatures of authenticity — with the PDF/A-2A + PDF/UA-1 tag tree layered on top. The clean source is a fresh DSL (and markdown, when expressible) you can re-render any time the document needs to change. Established remediators give you the first; we hand back both on one job, billed once.
Both is the default
Both — tagged original + regenerated DSL on one job, billed once. Recommended.
In-place — pixel parity with the source. Pick this for signed letters, government forms, statements.
Regenerate — clean editable DSL. Pick this when you want to retire the legacy authoring toolchain.
A clean source you own
Apryse, PDFix, CommonLook tag legacy PDFs in place. The next version of the document needs the same treatment again. We hand you back DSL — every future version is just another /api/v1/render call.
PDF/A-2A + PDF/UA-1
Embedded sRGB ICC profile, structure tree, language metadata, alt text on figures (via vision), tab + reading order. Output is dual-compliant by construction — not by best-effort post-processing.
Edit the regenerated source
The DSL we hand back is editable in our embedded chat at /ai/chat — describe the changes you want, see the PDF rebuild live, save it as a template you can re-render with new data. You don't need an AI subscription of your own; the chat runs on our models.
What it works on
Born-digital PDFs with extractable text, single-column layouts, headings, paragraphs, lists, tables, and images. Up to ~50 pages per submission.
Not a fit for v1: scanned / image-only PDFs (need OCR — separate higher tier later), interactive form fields, multi-column magazine / newspaper layouts, or DRM-protected PDFs. PDFs with digital signatures should run regenerate-only — tagging invalidates existing signatures.
Pricing
10 credits per page on success. Each page drives a Sonnet vision call, so /enhance is priced 100× a /render page (which is 1 credit per 10 pages). Buy credits up front on any plan, or settle per-request with an x402 USDC wallet. Partial outcomes (when at least one page can't be reconstructed) are not billed. The first page of any upload is free — sign in and try it on your own document.