What it is
A browser automation agent from browser-use. Given one natural-language goal, it drives a real Chrome page, choosing operations and target elements itself and calling a small LLM only when text must be typed.
How it uses Jev
Each observation builds an indexed element table. One TypeSafe request returns operation probabilities plus speculative target heads (click_target, type_text_target, select_target). The matching target executes; if the operation is TYPE_TEXT, a small LLM writes the text. Two decisions share one network round trip.
Primitives:choice
Technique worth stealing
Speculative fan-out: operation and target heads share one observed state, so two decisions cost one round trip.
Try it
git clone, uv sync, add keys to .env, uv run jev, open http://127.0.0.1:8766.
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-centricyes0.82
- Shows a System One patternyes0.85
- Handles uncertaintyno0.04
- Measuredyes0.90
- Runnableunclear0.57
- Worth recommendingno0.34
- Model replicano0.23
- Problem scopescore on a 0–2 scale0.83
- About Jevyes0.97
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.