TypeSafe Computer Use
A macOS computer-use loop uses OCR and Jev to choose bounded actions.
Added to Jevfast
How it uses Jev
It builds browser state from the goal, page, and action history, then asks for the next action and whether the goal is satisfied. When page items exist, Jev also selects an element; code gates and executes the decision.
What Jev decides
Illustrative Jev decisionExample answers · not a recorded Jev response · Source ↗
Question 1 · satisfied
YOUR APP
INSTRUCTION
Is the goal visibly achieved on the current page?
STATE
Browser goal, current page snapshot, available items, and action history.
JEV · NOUL
YesNo
App workflow
- Read browser state
Capture page text and actionable items alongside the goal and recent actions.
- Ask Jev
Request action Choice and satisfaction Noul, plus element Choice when items are present.
- Apply policy
Interpret the typed answers and confidence before acting.
- Execute next step
Use the selected browser action and element, then observe the page again.
Why it is interesting
The request combines a bounded action classification with a separate completion judgment, while deterministic page parsing supplies the candidate elements.