What it is
tiershift reads each LLM request, picks the cheapest model tier that can handle it, and escalates on evidence. It is for developers who want to cut API costs without losing quality.
How it uses Jev
Jev answers 11 typed questions about the request (difficulty, stakes, needs_reasoning, safety, etc.) in one ~180 ms call. Your YAML policy turns those probabilities into a tier; code does the arithmetic. Jev never sees prices or token counts.
Primitives:choicescorenoul
Technique worth stealing
Use a calibrated decision model to route LLM calls by policy, not by training data.
Try it
npm install tiershift or pip install tiershift; set TYPESAFE_API_KEY; run tiershift route "..." or use createRouter().complete().
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-centricyes0.90
- Shows a System One patternyes0.72
- Handles uncertaintyunclear0.42
- Measuredyes0.75
- Runnableyes0.85
- Worth recommendingno0.35
- Model replicano0.19
- Problem scopescore on a 0–2 scale1.88
- About Jevyes0.93
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.