Changelog

Fixes and improvements for the Local AI (Beta) mode:

• Formulas — LaTeX is now shown as real formulas in the result editor, and Word (.docx) downloads contain native Word equations
• Tables — AI mode now returns real tables with merged cells; Excel/CSV export keeps the structure
• Fixed: .docx download no longer swallowed the text around an inline formula
• Fixed: "Download all pages" Excel/CSV buttons could stay hidden after an early preview of a PDF that was still being processed
• The AI mode checkbox is more noticeable on mobile
• Files recognized with AI are now marked with an AI badge in "My queries"

Local AI (Beta) era — our biggest release yet:

• New "Local AI (Beta)" recognition mode — tables, formulas and handwriting. Text is read by a general-purpose multimodal vision model (not an OCR-only engine) running on our own GPU server, not a third-party AI service. Rolled out gradually by country
• Compare with original — new fullscreen view: scan and text side by side, words highlighted on the scan, desktop hover zoom lens and mobile tabs
• Table export to Excel/CSV — per page or all pages of a PDF at once
• Low-confidence words are highlighted in the editor, with a friendly tooltip explaining what to check
• Feedback invite under the editor — tell us what to improve right from the result page
• Editor toolbar tooltips in all 10 languages
• "Save changes?" dialog when leaving an edited result via Download
• Print button for results
• Local AI mode accepts PDFs up to 100 pages

Visible improvements:

• Laravel upgraded to version 10 — faster, more stable
• Frontend build system upgraded from Mix to Vite — assets load faster
• PDF-to-Image: per-page progress bar with raised limits (up to 400 pages)
• Specific error type shown on redirect (unablerec / captcha) instead of silent failure
• Upload form buttons disabled until both file and language are selected
• Button text changes to "Recognize" when a file is chosen
• Clarified data retention notice — now mentions auto-deletion after several hours

Visible improvements:

• hOCR support: rich-text OCR editor with word-level highlighting
• PDF-to-image conversion with a real-time progress bar
• Character and word counter in the OCR editor
• PDF page count shown in My OCR list
• Status badges and format indicators in My OCR list
• Automatic rotation correction for skewed images
• Improved recognition quality for noisy images
• Ctrl+V paste support in the upload area
• Localization improvements across all 10 supported languages

Under the hood:

• Refactored OCR processing pipeline
• Migrated styles from Sass to Less
• Added ESLint and code quality tooling
• Switched to Podman for containerized development

• Updated OS Fedora (fc35) and all software on the server to the latest versions
• Updated Laravel framework to version 8

Visible fixes on the site:

• Improved processing of PDF files
• Fixed a bug that did not allow to recognize many pages from PDF
• The selected active menu item is highlighted in the recognition section

Edits which are not visible on the site:

• Updated OS Fedora and all software on the server to the latest versions
• Updated Laravel framework to version 7
• Code refactoring

Minor edits and fixes for PDF

• Added new functionality for saving edited text (when AdBlock is disabled and if you are the owner of the post)
• Removed the Facebook widget (the third time the number of likes went wrong)
• Changes to the backend code

This update affects 26 files. What has been done:

• Improved micro-layout for info pages
• Added information "About the company"
• Updated FAQs
• Service edits

We rewrote this changelog and entered the information into the database.
Added FAQs.