Skip to sign in
UA DezquarePDF Accessibility Studio

PDF/UA · WCAG 2.1 AA · DezAI inside

Accessible PDFs, tagged right the first time.

Dez AI tags it, checks it, reads it aloud and learns from you.

Dezquare PDF Accessibility Studio tags, checks and repairs PDFs for PDF/UA — automatic where it can be, by hand where it must be. Meet Dez AI, the private AI that works inside your team and gets better with every file you correct.

  • ISO 14289-1 · PDF/UA-1
  • WCAG 2.1 AA checks
  • 61 machine checks on every build
  • Runs on your own server
  • Many people at once

Welcome back

Sign in — your files and the files given to you are waiting.

Forgot your password? Ask your admin to set a new one.

Your files stay on your server. DezAI works without the internet.

See it work

Upload a PDF. Dez AI tags it in one pass.

Headings, paragraphs, lists, tables, figures, captions, links and the reading order — proposed together, checked, and ready for you to review.

An illustration: a page of a research article is scanned from top to bottom. Dez AI marks the title as heading 1, the authors and the abstract text as paragraphs, a picture as a figure with a drafted description, a table with its header row, and a list, then shows that the PDF/UA checks end with no errors.

Why clients choose us

Accessible PDFs you can hand over with confidence.

Accessibility work is usually judged late — by an audit, a screen-reader user or a regulator. The studio moves the checking to the start, and keeps the proof.

  • Right the first time

    DezAI proposes the whole structure in one pass — headings, lists, tables, figures, links and the reading order — with a confidence on every tag. Your team starts from an almost finished file, not a blank one.

    The doubtful tags are listed first
  • Checked on every build, not hoped for

    61 machine checks of ISO 14289-1 and WCAG run on every build — plus veraPDF when you install it — and the report says exactly what is left for a person to judge.

    Contrast and table cells follow PAC 2024’s rules
  • Fast for one. Fair for many.

    300 pages in under a minute on a multi-core server. A file of 4,000 tags still answers a change in well under a second, and the queue serves each person in turn — nobody waits behind somebody else’s batch.

    Files, whole folders or the command line
  • Your files never leave

    DezAI and the Online AI that teaches it both run on your own network. Connections outside it are refused before a byte is sent — and logged. No cloud service, no tracker, nothing loaded from another server.

    Works without the internet
  • It learns each client’s house style

    Approved files teach the studio what a heading, a caption or a table looks like for that client. Every correction is learned live — the same fix is offered everywhere and used on the next file.

    Fewer fixes with every file
  • Proof for your client

    Every file ends with the accessible PDF, an Excel report — findings before and after, the tag tree, figures, tables, links, a manual checklist — and a history of who did what, when, and for how long.

    An audit trail for every file

Our own AI

Meet DezAI — the AI that lives on your server.

Dez AI is the studio’s own AI: a self-learning brain that works offline and is trained on your approved files — and taught by the Online AI, a language and vision model on your own network. It proposes every tag, drafts every alt text, explains its choice and learns from every correction — and nothing you give it ever leaves your network.

  • Tags a whole file in one pass

    Headings, paragraphs, lists, tables, figures, captions, links, footnotes, contents and reading order — proposed together.

  • Writes the alt text

    A draft description for every picture from the Online AI’s vision model; decorative ones become artifacts; a picture described once is remembered.

  • Learns from your corrections — live

    Change a tag or cut a zone once. A few seconds later Dez AI offers the same fix everywhere in the file and uses it on the next files of that client.

  • Learns from the Online AI, its teacher

    Every answer the Online AI gives is kept as a lesson. Next time Dez AI answers the same kind of question itself — in a blink, even offline — and checks itself against the Online AI now and then.

  • Knows each client’s style

    Learns the house rules from approved files — what is a heading, where a paragraph ends — and keeps one convention inside a file.

  • Shows its reasons

    Every proposal says how sure it is and why; the doubtful tags are listed first, so your time goes where it matters.

  • Private by design

    Dez AI refuses every connection outside your network. No cloud service, no tracker, no fonts from the internet.

Learns live

Correct it once. Dez AI remembers.

Two paragraphs ended up in one zone? Cut it by hand — one time. Dez AI sees what you changed, offers the same cut at every place in the file that looks alike, and from then on cuts the same way in the next files of that client and of clients that work like it.

  1. You fix one placeDraw, cut or join a zone — as you always do.
  2. Dez AI offers the rest“9 more places look the same — cut them all?” One click, one undo.
  3. The next file is rightIt must prove itself on files it has not seen before — only then does it act.

Learns from every answer

Online AI teaches. Dez AI remembers.

The Online AI can answer almost anything — slowly, and only once. So every answer it gives becomes a lesson. DezAI studies the lessons, proves that it can answer the same kind of question alone — and then does, in a blink, even when the Online AI is switched off.

  • Proven, not trusted. A student answers alone only after it agrees with its teacher on questions it has not seen — and it is checked against the teacher now and then.
  • People outrank both. What your team decides in a file counts more than any answer; an answer a person changed is never used again.
  • Your studio’s own words. Dez AI shows the teacher how your team worded similar things, so what it writes follows your house style.

An illustration: the Online AI, the teacher, answers a question; the answer travels to Dez AI as a lesson. As the lessons grow, the share of questions that Dez AI answers by itself rises, and so does the waiting time saved.

  • 61machine checks of PDF/UA and WCAG on every build
  • 300pages tagged in under a minute on a multi-core server
  • 4workflow steps: Production, QA, Final QA, Delivery
  • 0files sent outside your network

Why teams choose it

Everything an accessibility team needs, in one studio.

  • PDF/UA and WCAG, checked

    ISO 14289-1 machine checks on every build, colour-contrast checks with one-click repair, optional veraPDF, and an Excel report for every file.

  • An editor that feels like Acrobat — and is faster

    Draw a zone and press a key, drag the reading order, merge and split tags, edit tables and lists, undo and redo — with hotkeys for the everyday tags and edits.

  • Links, in one stage

    Cross-references, footnotes, contents, figures, tables, references, web and e-mail links found for you — one-way or two-way — checked and tagged.

  • Hear it the way a screen reader will

    Say-all with the words lit as they are spoken, quick keys, and every problem a listener would trip over — before your client does.

  • Batch, fast, multi-core

    Upload files or whole folders and let the queue run: several files at once, every page analysed on all your cores, and everybody served in turn.

  • House styles for every client

    Profiles per client, learned from their approved files, so a publisher’s books come out the way that publisher wants them — every time.

  • Scanned pages and many languages

    Optional OCR gives scanned pages a text layer; Unicode text — Tamil, Greek and more — is read and tagged, with the language set for each passage.

  • Made for teams

    Roles, assignments, due dates, PAC reports, review comments and a history of every step — and one person at a time in a file, so nobody’s work is overwritten.

…and the details that save hours

  • Draw a zone, press a key
  • Undo and redo — 60 steps
  • Hotkeys for the everyday tags and edits
  • One click: fix everything the tool can fix
  • Import the tags a file already has — never overwritten silently
  • Reading order checked by the Online AI’s vision model, page by page
  • Split two-page spreads
  • Zoom up to 800 %
  • Colour-contrast repair — only what you tick
  • Review comments pinned to the page and the tag
  • Rights for each person
  • Working time for every file
  • An offline rule book of PDF/UA and WCAG
  • 100+ options · 10 built-in profiles
  • Light and dark look
  • Windows and Linux

Built for real documents

Journals, books, worksheets, forms — tagged properly.

Every kind of content gets the structure a screen reader needs — not just a tag that happens to pass a check.

  • Headings

    Levels checked and never skipping, with a section around every top-level heading.

  • Columns and reading order

    Two-column articles are read in the right order — and you can drag it into shape.

  • Lists

    Nested, numbered or lettered; a list that runs over a page break stays one list.

  • Tables

    Header cells and scope, merged cells, and a table that runs over several pages becomes one table.

  • Figures and alt text

    A description drafted for you to approve; decorative pictures become artifacts.

  • Captions and footnotes

    Linked both ways — the note and its mark, the caption and its figure.

  • Links and cross-references

    Web and e-mail addresses, contents entries, sections, figures, tables and references.

  • Contents and index

    Contents entries and index page numbers become working links.

  • Reference lists

    One entry, one tag — however many lines it runs over.

  • Mathematics

    Formulas read aloud — “a squared plus b squared equals c squared” — with the LaTeX kept for the report.

  • Forms

    Fields are tagged as forms and get a label a screen reader announces.

  • Running heads and page numbers

    Marked as artifacts, so nobody has to hear them on every page.

  • Worksheets

    Write-in lines are announced as a blank line — not as noise.

  • Scanned pages

    Optional OCR gives a scan a text layer, ready to be tagged.

  • Many languages

    Tamil, Greek and more, with the language set for each passage.

  • Colour contrast

    Measured against the real background — photos and gradients included — with a colour suggested.

Screen-reader preview

Hear it before your client does.

The Preview tab reads the tags in reading order, in the phrasing a screen reader such as NVDA uses — heading levels, lists with their number of items, tables with their headers, links and descriptions. A missing label is heard, not just flagged.

  • Say-all with the words lit as they are spoken — by a voice on your own computer
  • Empty table cells are read as “Blank”, write-in lines as “Blank line”
  • Every problem a listener would trip over is listed with its page

An illustration: a page with a heading, a paragraph, a picture, a table, a list and a link is read block by block, and what a screen reader would announce appears beside it: heading level 1, the text, the picture’s description, a table with 3 columns, a column header, a list with 3 items and a link with its description.

Standards

61 machine checks. Nothing left to chance.

Every build is checked against ISO 14289-1 (PDF/UA-1) and WCAG 2.1 AA. You get a score before and after, a list of what is left — and a manual checklist for what only a person can judge.

  • Document is tagged and marked as tagged
  • Language and title are set, and the title is shown
  • The structure tree is whole and well formed
  • Reading order follows the page
  • Heading levels never skip
  • Every figure has alt text — or is an artifact
  • Tables have header cells and scope
  • Lists are lists, with their items and labels
  • Links have a destination and a description
  • Form fields have labels
  • Fonts are embedded and text maps to Unicode
  • Text contrast is measured on the real background
  • Bookmarks follow the headings
  • The metadata declares PDF/UA-1

Checker errors on our reference files

Before and after the studio’s automatic pass — nobody has touched a tag yet.

  • Journal article, two columns9 → 1
  • Book with contents and bookmarks6 → 0
  • Form with fields9 → 0
  • Attendance register with a table6 → 0
  • Tamil and Greek text6 → 0
  • Rotated pages9 → 1

The one left is a picture: its description is drafted for a person to approve.

An honest note. A machine check is not a conformance certificate. Confirm delivery files in PAC 2024 and veraPDF — the studio runs veraPDF for you when it is installed, and its workflow asks for the PAC report before a file moves on to QA.

Delivery

What lands on your client’s desk.

A file, its proof and its story — in one delivery.

  • The accessible PDF

    Tagged, with its language, title, bookmarks and metadata set. The pages look exactly as before — we check that pixel for pixel.

  • The Excel report

    Scores, findings before and after, the tag tree, figures, tables, links and fields — and a manual checklist with Pass, Fail or N/A.

  • The verdict

    veraPDF’s PDF/UA-1 verdict on every build when it is installed, and the PAC report attached to the file before it goes to QA.

  • The story of the file

    Every step with who and when, the review comments with their category and severity — and the time each person really worked on it.

How it works

From upload to delivery in four steps.

  1. 1Upload

    One file, many files or a whole folder. Pick the client’s profile — or let DezAI choose by the look of the file.

  2. 2Dez AI tags it

    The structure, the reading order and the alt texts are proposed in one pass, with a confidence for every tag.

  3. 3You review and fix

    Look at the doubtful tags first. Every correction teaches Dez AI — live, not at the end of the day.

  4. 4Build, check, deliver

    The accessible PDF is built, checked against PDF/UA and WCAG, and delivered with its report.

Built for teams

Many people. One file, one editor at a time.

Work moves from Production to QA, Final QA and Delivery with roles, due dates, PAC reports and review comments. While somebody works in a file, everybody else can open it and look — but nobody can change it, so two people never overwrite each other.

  • Fair queue — one file of each person in turn, urgent files first
  • Live “who is in this file” and “somebody changed it” notices
  • An admin can take a file over; a forgotten tab lets go by itself
  • Review comments with a category and a severity, pinned to the page and the tag
  • A PAC report is asked for before a file goes to QA
  • Working time for every file — only active time, never breaks
  • Rights for each person: who may upload, delete, download or use the AI

What changes

The same work, far less of the waiting.

Tagging by hand compared with working in the studio
WhatTagging by handIn the studio, with DezAI
The first structureBuilt tag by tagProposed for the whole file in one pass, with a confidence on every tag
Finding the doubtful placesReading everything againListed first — the answer to one settles the others that look the same
The same fix, again and againRepeated by hand, each timeOffered everywhere at once, and learned for the next file
Alt textWritten from scratchDrafted by the Online AI’s vision model; a picture described once is remembered
CheckingA separate tool, at the endLive issues while you edit, and 61 checks on every build
Quality controlE-mail and spreadsheetsReview comments, rework rounds, a PAC gate and a history
Many people on one jobOverwritten files and version namesOne person at a time in a file; everybody else watches it live
Where the files goDepends on the tool — some are online servicesNowhere — your own server

Private by design

Your files stay yours.

  • DezAI and the Online AI that teaches it run on your own network — no cloud service
  • Every connection outside your network is refused before a byte is sent, and logged
  • The pages load nothing from the internet: no CDN, no tracker, no web fonts
  • Users, settings and the learning database stay on your machine; the learning database has a one-click backup
  • Passwords are kept only as salted PBKDF2-SHA256 hashes; a sign-in ends when the app’s tabs are closed
  • Roles and rights for each person, and a log of who did what

Questions

What clients ask before they start.

Straight answers — including the limits.

Is the result certified PDF/UA?

No tool can certify that by itself. The studio checks every build against ISO 14289-1 and WCAG 2.1 AA with 61 machine checks, runs veraPDF when you install it, and lists what only a person can judge — the quality of an alt text, whether a reading order makes sense — in a manual checklist. Confirm delivery files in PAC 2024 and veraPDF; the workflow asks for the PAC report before a file goes to QA.

Do my files leave my server?

No. The studio, Dez AI, the Online AI that teaches it and the learning database all run on your own network. Every connection outside it is refused before a byte is sent, and logged — if the Online AI should live elsewhere, your admin has to name that host in the Privacy guard first. The pages load no font, script or picture from the internet.

What do we need to run it — and is a GPU required?

A Linux or Windows server with Python 3.10 or newer, and a browser for everybody else. Tesseract (for scanned pages), veraPDF and the Online AI (a language and vision model server on your network) are optional. Tagging, checking and editing need no GPU; the Online AI’s models run faster on one but work on a CPU too — and DezAI needs them less as it learns. Without the Online AI the studio still works.

Can the AI learn something wrong?

Learning has to prove itself first: a model or a rule acts only after it was right on files it was not trained on, a learned rule that people keep undoing is retired by itself, and what your team decides in a file counts more than any AI answer. An admin can switch each part off.

Can many people work at the same time?

Yes. The queue serves each person in turn and runs several files at once; in a file one person edits while the others watch, so nobody’s work is overwritten. An admin sees who works on what, and for how long.

Which PDFs does it handle?

Journal articles, books, reports exported from Word or InDesign, forms, worksheets, and files that already have tags — which can be imported as they are. Scanned pages get a text layer when Tesseract is installed.

Which languages?

Any text in Unicode — Tamil, Greek, accented Latin and more — is read and tagged, with the language set for each passage so assistive technology can switch voice. The languages OCR can read depend on the Tesseract language packs you install.

Can we follow a client’s house style?

Yes. A profile holds a client’s options, the client’s approved PDFs teach DezAI their conventions, and a client can have a tag structure of its own.

Can it run in batches or from scripts?

Yes: upload files in the browser (an admin can upload a whole folder), or use the command line — with a profile, options, a check-only mode that returns an error code, and a batch summary workbook.

Ready when you are.

Sign in and open your files — DezAI has been learning your way of working.

Sign in