OctoPDF checks a PDF's accessibility the way PAC does, then repairs the tagging and structure automatically. Remediators can spend their time on the judgement calls, such as alt text, instead of hours of manual tag fixing in Adobe Acrobat.
OctoPDF never changes what is printed on the page. It only changes tags, structure, attributes, metadata and link and form annotations, so the repaired PDF looks exactly like the original.
1. Who it's for
OctoPDF is for PDF accessibility remediators and document teams who receive finished PDFs, often Word exports with no source file, and need them to meet PDF/UA and WCAG as checked in PAC and Acrobat's Accessibility Checker.
2. Beyond automatic remediation
Remediating a long PDF by hand can take days. OctoPDF turns days of work into hours, and hours into minutes, leaving remediators free to focus on the decisions that need a person.
Where automated remediation typically stops at adding tags, OctoPDF draws on years of hands-on manual remediation experience. It rebuilds lists, tables, reading order and table of contents links the way an experienced remediator would, then validates the result with veraPDF before saving.
Documents that have already been remediated by hand benefit too. OctoPDF finds and corrects errors that are easily introduced without noticing during manual work.
Finally, OctoPDF resolves issues that pose no real barrier to readers but still cause accessibility checks to fail, such as quirks in how checkers interpret a document. The aim is a document that is accessible in practice and passes in PAC and Acrobat's Accessibility Checker.
3. What it does
- Audit: A score from 0 to 100 from PAC-style checks, plus full PDF/UA-1 validation with veraPDF, which is included.
- Tag repair: Removes Span, Sect and Div clutter, empty tags and contradictory artifact and tag sequences, and rebuilds broken structure trees, and fixes problems PAC reports, such as figures nested inside headings and empty marked content in footers.
- Artifacts: Marks running headers and footers, including bare page numbers, as artifacts, along with backgrounds, borders, blank lines and dot leaders. A logo repeated in the header or footer on every page becomes an artifact, and the first one stays tagged.
- Reading order: Fixes reading order, including two-column pages tagged straight across, which are put back into left column then right column order.
- Headings: Fixes skipped levels, starts the document at H1, promotes paragraphs styled as headings and demotes body text tagged as a heading.
- Lists: Turns bulleted or numbered paragraphs into real lists with correct L, LI, Lbl and LBody tags. Joins lists split in two, nests sub-lists, splits numbered steps written inside one paragraph into a lead-in and a list, and gives tick and cross icon bullets meaningful alt text such as Included and Excluded.
- Tables: Sets header cells with Scope, header IDs and Headers associations, handles row and column spans, joins tables split by page breaks and adds table summaries.
- Links: Tags link annotations, joins links split across lines, turns plain-text web addresses into links, and links table of contents entries so each one lands exactly on its heading.
- Document settings: Builds bookmarks from headings and sets the title, language, tab order and PDF/UA identifier.
- Forms: Adds field descriptions from the visible labels and tags form fields.
- Fonts: Embeds missing fonts from the copy installed on your computer, only after checking it's exactly the same font so the text doesn't move. Only the characters needed are embedded, which keeps files small. The standard PDF fonts are embedded using their metric-identical equivalents, with every character width checked: Helvetica using Arial, Times using Times New Roman and Courier using Courier New.
- Unicode mappings: Rebuilds missing Unicode maps from the font's own data, so screen readers read the text correctly and it copies properly. Broken CIDSet font data that fails PDF/UA is also repaired.
- Untagged PDFs: Tags them from the page layout, keeping superscripts such as cm² in their line.
- Alt text: Writes alt text with a local LLM, a vision model that runs on your own computer, and marks logos and decorative images.
- Before and after: Shows the score before and after repair, with a list of every change made.
- Stable results: Running OctoPDF again on a repaired file changes nothing, and the pages always look identical.
4. How it works
OctoPDF is a portable Windows app: a single file with nothing to install. It opens in its own window. Drag a PDF or a folder of PDFs into the window, or onto the OctoPDF icon, and OctoPDF scores each file, repairs it, checks it again and shows the before and after score. For a folder, every PDF inside is repaired, followed by a summary of the whole batch.
Text description: OctoPDF window waiting for a file
Example of the OctoPDF app window. At the top is the OCTO PDF banner in orange block letters, the line PDF ACCESSIBILITY REMEDIATION and a copyright line. Below it a box says Drag a PDF or folder into this window, then press Enter. Under the box are the keys I for info and Q to quit, and an orange prompt waiting for input.
Text description: OctoPDF window while it repairs a PDF
Example of the OctoPDF window while it repairs Annual report.pdf, a 40-page, 736 KB file. Three steps are ticked as done with their times: Analysing document, 4.5 seconds; Repairing tag structure, 11.7 seconds; Artifacts and reading order, 18.3 seconds. Lists and tables is in progress. Still to come are Links and references, Headings and navigation, Alt text using the local LLM, Validating with OctoPDF and veraPDF, and Saving. The progress bar is at 40 percent after 51 seconds.
Text description: OctoPDF's results, with the score going from 78 before to 94 after
Example of the results at the end of a repair. The Result box shows a score of 78 before and 94 after, PDF/UA-1 passed in veraPDF's machine checks, and 5 issues fixed, 1 left, in 51 seconds. The What OctoPDF fixed box lists counts by category: 48 tables, 20 artifacts and content, 2 lists and 1 document settings. The Worth a look box says the local LLM wrote 6 alt texts and marked 2 images decorative, please spot-check, and notes a figure without alt text on page 3. Last comes Saved, [OCTOPDF] - Annual report.pdf, and the keys O to open the PDF, F to show it in its folder, C to list changes, I for info and Q to quit.
The repaired copy is saved next to the original as [OCTOPDF] - your file name.pdf, so the original is never overwritten. Press C to see every change, O to open the repaired PDF or F to show it in its folder. A command line is also included for batch work.
Text description: OctoPDF's list of changes
Example of the change list for [OCTOPDF] - Annual report.pdf, which made 42 changes. The first six are: Fixed invalid language code "en_GB" to "en-GB"; Merged 2 list(s) that had been split into separate <L> tags; Associated 48 table cell(s) with their header IDs; Added PDF/UA identifier to XMP metadata; Marked running headers, footers and page numbers as artifacts; Linked 12 table of contents entries to their headings.
5. Private by design
Your documents stay on your computer. Alt text is written by a local LLM, a vision model running through Ollama on your own computer, so no document is ever uploaded. The first time OctoPDF starts, it offers to download and set up Ollama and the model for you, showing the download size, the memory your computer has and a warning if it's low.
Text description: OctoPDF offering to set up the local LLM
Example of the Local LLM box. It says the local LLM is not set up, and that OctoPDF uses it for steps such as writing alt text for images. Without it, accuracy is degraded and those steps are left for you to do by hand. Download: about 7 GB, Ollama and the vision model qwen3-vl:8b. Memory: 16 GB RAM recommended, 8 GB minimum, and this computer has 32 GB, shown in green. Privacy: runs on this computer only, documents are never uploaded. Licences: Ollama MIT and the model Apache 2.0, listed in the notices file. A warning says it is not recommended for low-specification computers, because with less memory the LLM is slow and can make the computer unresponsive while it works. It asks Would you like to install it now?, with the keys A to install, S to skip for now and N for don't ask again.
The local LLM is optional. Without it, everything else still works and alt text is left for you to write.
6. What it can't fix
- Scanned or image-only PDFs: These need text recognition first. OctoPDF tells you when a file needs it.
- Some fonts: OctoPDF can't embed a font that isn't installed on your computer, isn't exactly the same as the font in the PDF, or whose licence forbids embedding. It can't rebuild Unicode mappings for the rare fonts with no usable character data, which need the source document.
- Visible content: OctoPDF never edits page content, so problems such as colour contrast stay as they are.
- Human judgement: Alt text written by the local LLM should always be checked by a person. Very complex layouts may still need a manual pass, and no automatic tool can guarantee a fully accessible document.
7. Requirements
- Windows 10 or 11, 64-bit.
- Microsoft Edge WebView2 for the app window. It's built into Windows 11 and included in updated Windows 10.
- Everything needed for validation is included.
- The local LLM is about a 7 GB download, set up once. 16 GB of memory is recommended and 8 GB is the minimum.
8. Availability
OctoPDF is a private tool and is not available to download. It's being considered for public release in the future. If you'd like to hear if that happens, email support@octopusscanner.com.
Was this page helpful? Email your feedback about OctoPDF.
