0.6.0

What's Changed

• Guided setup page with explicit model choice by @OJPOJ in #27
• add report a problem link which opens a github issue by @steven-ngle in #26
• Simplified settings: basic and advanced (#9) by @OJPOJ in #28
• Jump to flagged paragraphs from the popup tiles by @OJPOJ in #29
• Check selected text in the browser's PDF viewer (#31) by @OJPOJ in #37
• PDF viewer based on pdf.js components, opt-in (#32) by @OJPOJ in #38
• Scan PDFs in the viewer: paragraph extraction and local scoring (#33) by @OJPOJ in #39
• PDF viewer: marking, scroll to place, shorter paragraphs, balanced zoom (#34) by @OJPOJ in #40
• PDF support docs: privacy, store listing, checklist (#36) by @OJPOJ in #43
• Save an annotated copy of a scanned PDF (#42) by @OJPOJ in #44
• Release 0.6.0 by @OJPOJ in #45

New Contributors

• @steven-ngle made their first contribution in #26

Full Changelog: 0.5.1...0.6.0

0.5.1

What's Changed

• fix: draggable text areas by @OJPOJ in #2
• 3 switch repo language to english only by @OJPOJ in #6
• Release 0.5.1: automatic GitHub release pipeline, shorter manifest de… by @OJPOJ in #7

Full Changelog: 0.5.0...0.5.1

0.5.0

AI Content Flag 0.5.0 – first public release

Browser extension that checks text paragraphs on web pages for AI characteristics and marks them with a traffic light (🟢 unremarkable · 🟡 unclear · 🔴 conspicuous · ⚪ uncertain), with an AI score of 0–100. By default everything runs locally in the browser – texts read do not leave your computer.

▎ A hint, not proof. The score is not a probability. Human-written texts are sometimes also marked red. Please do not assume AI use for anyone on that basis alone.

✨ What's included

• Three models to choose from, loaded once from Hugging Face (no account needed):
  • Accurate (desklib) – new default, fewest false alarms (~1 in 100 human paragraphs in tests)
  • Balanced (fakespot) – new, 125 MB, good compromise
  • Fast (TMR) – very fast, but more false alarms
• Own backend optional: own server, cloud service or Hugging Face Inference API – the settings clearly show before anything is sent.
• Right-click check: check selected text or a single paragraph (Alt+Shift+C), also on pages that are not enabled.
• Scan modes: only on button press, on selected pages (default), or on all pages.
• Group short paragraphs: consecutive short paragraphs can be evaluated together.
• Language detection: non-English paragraphs are recognized and deliberately not scored.
• "Check model": built-in reference set to measure the accuracy of the selected model yourself.
• Feedback in the check result: indicate where a text comes from – only with consent, only locally.
• Score cache: already scored paragraphs are remembered as a hash (without text, without address), 30 days by default, deletable at any time.
• Welcome page, about/license page and privacy policy in the extension.

🔒 Privacy

• No tracking, no analytics services, no account.
• Built-in blocklist with around 14,000 sensitive domains (banking, webmail, authorities) – never reloaded from the network.
• Automatically pauses on pages with password, credit card or one-time code fields.
• Alt+Shift+A stops all processing immediately.

📦 Installation

1. Download ai-content-flag-0.5.0.zip below and unpack it into a permanent folder.
2. Open chrome://extensions (or edge://extensions), turn on developer mode.
3. "Load unpacked" → select the folder with the manifest.json.
4. In the settings choose a model and click "Download".

Runs in Chrome, Edge, Brave and other Chromium browsers. Detailed instructions in the README (https://github.com/OJPOJ/ai-check-extension#-installation).

Update from an earlier developer version: replace the contents of the previous folder and click ↻ at chrome://extensions. Settings and model are kept; whoever used TMR so far keeps TMR.

⚠️ Known limits

• Only English is evaluated.
• Texts under about 120 words are mostly marked as "uncertain".
• Content in iframes and in the Shadow DOM is not reached.
• Newer or deliberately reworded AI texts may go undetected.