Overview & result
Breaking: Browser Use + Jev = Ultrafast ⚡ Findings flights took 7s and cost only $0.0039 🤯 > new action space every step > DOM state space > small LLM fallback to type (this video is at 1x speed btw) Built a tiny open source browser agent. try it below ↓
How Jev fits in the loop
- Codex hands the Skill a goal for navigation, clicks, toggles or scrolling.
- The Skill observes the page's controls as accessibility text, not screenshots.
- Jev chooses an action and the browser performs it inside the existing Computer Use connection; the loop repeats.
- Jev's completion signal brings Codex back to verify the result.
How to reproduce
- Inspect the original showcase and source material at https://github.com/wy-coliney/jev-browser-use
- Verify the bounded prompt and input candidate schema configured for Jev
- Benchmark decision latency and classification accuracy against baseline models
Why this build matters
It avoids a new host-model turn for every click while Codex keeps text input, visual judgment and the final check.
Limitations
- The 5–10× speed-up is the author's figure from their own EZCollegeApp workflows.
- Browser execution currently requires the Codex Computer Use runtime connected to Chrome; Claude Code support is listed as coming soon.
- Page text goes to the provider you select, and workflows were exercised on macOS.