Docs / source reviewed
jevnav
Record browser target choices as replayable action traces
How this project uses Jev
- Input
- The page title and URL, visible interactive-element candidates and the intent of the current step
- Jev decides
- Use Choice to select an element or no suitable target, retaining probabilities and confidence
- Code executes
- Apply confidence, domain and sensitive-action gates before executing, reviewing or refusing an action; save traces for replay and change checks
Evidence and limitations
Decisions use text candidates, not pixel understanding; candidate truncation can omit targets. Replay avoids inference but still needs the target page, and execution replay repeats actions. A done claim needs a success condition. Author speed and accuracy claims were not independently tested.
This project has not been run independently here. Author-reported results are not independently verified results.
Original sources
- https://github.com/dtduc-git/jevnav/blob/9a46efd2519e46da86e0ec17f282e65dd3fb0ee1/README.md ↗
- https://github.com/dtduc-git/jevnav/blob/9a46efd2519e46da86e0ec17f282e65dd3fb0ee1/src/jevnav/decide.py ↗
- https://github.com/dtduc-git/jevnav/blob/9a46efd2519e46da86e0ec17f282e65dd3fb0ee1/src/jevnav/gates.py ↗
- https://github.com/dtduc-git/jevnav/blob/9a46efd2519e46da86e0ec17f282e65dd3fb0ee1/src/jevnav/replay.py ↗