What it is
jev-browser drives a real headless browser from a task and a start URL, via an MCP server, CLI, or library. It is aimed at developers who want browser automation that is fast and very cheap compared to LLM-driven browsing.
How it uses Jev
Each step, Jev picks one action from the page's clickable, typeable, and selectable elements, and scores how likely the goal is met or the run is stuck. Code owns the loop: budgets, recovery, stop gates. Results come back as a step trace with confidences, the final page, console errors, and a screenshot.
Primitives:choicescore
Technique worth stealing
Keep the control loop in code and let the model only choose one action and score goal/stuck per step.
Try it
npx -y @jkudish/jev-browser run "<task>" <url>
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricyes0.94
- Shows a System One patternyes0.92
- Handles uncertaintyno0.13
- Measuredyes0.91
- Runnableyes0.92
- Worth recommendingunclear0.56
- Model replicano0.05
- Problem scopescore on a 0–2 scale1.15
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 21 Sept 2026.