Model router that asks Jev eleven questions in one call and lets a YAML policy turn the answers into a tier. Every request starts on the fast tier and only escalates on evidence such as multi-step reasoning, hard difficulty, high stakes or safety, and every decision is logged without the message text.
User repository on GitHub
Shift every LLM call to the cheapest model that can handle it. Routing decided by TypeSafe Jev in ~180 ms. No training data. Policy in plain YAML. TypeScript and Python.
Repo stats from the GitHub API, cached Oct 2026.
Stop paying flagship prices for prompts that don't need one. tiershift reads each request, picks the cheapest model tier that can handle it, and escalates on evidence. The judgment comes from TypeSafe Jev , a calibrated decision model. About 180 ms. Four cents per thousand routes.
Quoted from the project's README. Read the full README.
Watch out
The benchmark is 120 single-turn prompts with clear answers, and the README says plainly that it cannot speak to long agentic tasks with tools, large context or recovery from mistakes. The large saving quoted by the report command is a token-count estimate against the most expensive model in the default config, while the 40 percent in the table is measured against one flagship with judged answers, so your number depends on which flagship you pay for today. The flagship tier sits in the config, off by default. Provider keys are optional and a tier silently skips models whose key is missing. Copy reviewed against the repository README on 2026-10-04, not from running it.
rajdhakad9826/jev-routerTypeScript
BeLazy167/typesafe-modTypeScript
tylerjharden/ailerixTypeScript
bestagentkits/jev-skillfulTypeScript
iefnaf/pi-jevTypeScriptPosts from builders shipping with Jev right now.
