Progress & roadmap
What is shipped, and what is next.
An honest status board. Nothing here is a promise of a date or a price; an app is listed as released only when it can be bought today.
Released · 12 October 2026
TagSmith-PDF 0.7.0
- Built-in AI for alt text. A small vision model ships inside the installer and describes every photo, chart, diagram and logo during the run, in the document’s language. Each description stays an AI draft until a person accepts it. Ollama can still be used instead, on the same computer only.
- Tables rebuilt from the page grid and charts drawn as shapes tagged as figures that need a description.
- Standard fonts (Helvetica, Times, Courier) embedded with metric-compatible fonts, so PDFs that used them without embedding now pass.
- Reports: before/after HTML report, Accessibility Conformance Report in the VPAT® 2.5 layout, audit.csv and a draft EU accessibility statement in English, Dutch or German.
- Measured on the release build: 97 of 97 real public-sector PDFs fully PDF/UA-1 conformant (EU 31/31, Dutch 40/40, US tax forms 26/26); 344 automated tests, 0 failures, 2 skipped.
Buy it on Gumroad (€149 once, one PC) · full details.
In development · not for sale yet
Next apps
EAA Alt-Text Batcher
Pick a folder of images and a vision model installed on your PC; get one line of alt text per image in a CSV next to the images. Resumes interrupted runs, never uploads anything, flags descriptions that are longer than the advisory length. Windows only; not code-signed yet.
DossierDesk
Local dossier search (full-text, page-correct citations), Dutch business-document entity extraction, OCR for scans, Peppol e-invoice generation (UBL 2.1, Peppol BIS Billing 3.0, NLCIUS) and Kadaster enrichment. Kadaster runs in a demo mode until a paid subscription is configured; the invoice validator is structural and must be followed by an official Peppol check before the first live transmission.
RedactSmith
Reviews a PDF, finds personal data (local OCR for scans), lets you draw or confirm regions, physically removes them, cleans metadata and reports what it checked. A prototype: detection accuracy is not yet measured and it is not validated for production use.