What it does
Jev reads page text and controls, not screenshots. The maker’s side-by-side form demo plays 2.5 times faster than the recorded runs.
Browser & computer use
A browser agent uses Jev for routine steps and asks a separate vision model for help when it gets stuck.
Only you can see your notes.
Recording unavailable. Open the experiment ↗
Jev reads page text and controls, not screenshots. The maker’s side-by-side form demo plays 2.5 times faster than the recorded runs.
For browser chores on your own test pages, study how this agent gives simple choices to Jev and sends uncertain steps to another model that can read screenshots. Follow the source's setup instructions. Running it needs an OpenRouter access key and separate access for the model that handles difficult steps.
Begin with harmless test forms and leave changes disabled until you want the agent to submit something. Check the submitted values yourself. The reported perfect result covers repeated runs of five local tasks. One completion check loses the new page text before it reaches Jev, so check that path separately.