jevbooks

← All projects

Jev-Auto-Router

miniLV/Jev-Auto-Router · Homepage

Per-call Codex GPT routing via TypeSafe Jev, with independent task verification.

What it is

A prototype for coding agents: a local Responses proxy that lets one Codex session re-select a GPT model and reasoning effort on every meaningful call, then independently verifies whether the task still passed.

How it uses Jev

The proxy builds the host's actually requestable (model, effort) pairs and sends Jev one Choice over them, deciding model and effort together. Timeouts, low confidence, or bad output fall back to Terra/medium with a recorded reason. Router Compass logs each choice, actual usage, and verification result.

Primitives:choice

Technique worth stealing

Route per model call, not per task, and verify the task independently afterward instead of trusting the executing model's self-report.

Try it

npm ci; npm run build; JEV_API_KEY=" " npm start; curl localhost:8787/health; npm test

View on GitHub

judged by Jevjev-1.13.0

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 patternyes0.77
  • Handles uncertaintyno0.05
  • Measuredno0.06
  • Runnableunclear0.53
  • Worth recommendingunclear0.43
  • Model replicano0.04
  • Problem scopescore on a 0–2 scale1.06
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.