🎉 Initial Release
Core Features
VisionEnhancedAgent— drop-in replacement for browser-use Agent with vision capabilities- Florence-2 vision backend — OCR + Dense Region Caption via remote GPU API
- Set-of-Mark (SoM) screenshot annotation — numbered labels on interactive elements
- Adaptive vision strategy — DOM confidence scoring (~50% GPU savings)
Benchmark
- Real-world benchmark: 6 tasks across icon-heavy / mixed / DOM-rich websites
- Vision-enhanced: 100% success rate (6/6) vs baseline 67% (4/6)
Testing
- 72 unit tests + 3 E2E integration tests
- GitHub Actions CI (test + lint), all green
- Core module coverage: 67-95%
See CHANGELOG.md for details.