New:
- First-run setup: paste your Gemini key on first open, hit "Test key" for a live check — no more guessing if the key works
- Auto-scan: the panel finds questions the moment it opens; Scan button still there to re-run
- Answers survive reloads: solved answers are saved per page (last 10 pages) and restored when you come back
- Confidence ↑: reorder cards least-confident-first so you review the shaky answers before submitting
- ↻ Re-answer: regenerate a single wrong-looking answer without re-solving the whole page
- Export: copy all questions + answers to the clipboard as plain text

Changed:
- Default model is now gemini-3.1-flash-lite (previous default is no longer available to new API keys; also fixes image questions during Gemini capacity spikes)

Fixed:
- Questions rendered as images (NPTEL) are now detected and read via Gemini vision instead of failing with "no question text"

Version 2.1.0 brings chat functionality, automatic provider fallback, and improved extraction reliability.

New features:
- Chat sidebar for follow-up questions about any assignment
- Automatic fallback from Gemini to Groq if one provider fails
- Groq settings in the extension panel

Improvements:
- Better extraction of questions from complex page layouts
- Enhanced handling of image-only options
- Improved SPA support with automatic re-scan
- Build process now injects API keys from .env file

Bug fixes:
- Fixed extraction of questions inside shadow DOM and iframes
- Fixed navigation text filtering
- Fixed image alt text reading
- Firefox compatibility improvements