What it is
A local agent harness for people running tasks that are mostly typed judgements. Jev handles every judgement expressible as a type; an ordinary LLM is paid for only when prose is actually needed.
How it uses Jev
Jev answers typed questions against material: classifications, ratings, yes/no calls, each with a calibrated probability. It plans, assesses, routes, sizes each step's model, picks skills and roles, and decides app composition. Results drive rules that pick models, tools and actions.
Primitives:choicescore
Technique worth stealing
Split typed judgement from prose: one Jev call carries many judgements, and the plan is compiled once and cached for reuse.
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-centricyes0.92
- Shows a System One patternyes0.74
- Handles uncertaintyno0.10
- Measuredno0.23
- Runnableyes0.75
- Worth recommendingunclear0.48
- Model replicano0.13
- Problem scopescore on a 0–2 scale1.56
- About Jevyes0.97
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.