jevbooks

← All projects

stuntd

bladedevoff/stuntd · Homepage

Local proxy that learns your typed LLM decisions.

What it is

stuntd is a local, self-hosted proxy for developers who make typed LLM decisions. It records Choice, Score and Noul questions, distils each decision site into a small head on a frozen Laya encoder, and answers locally with calibrated confidence.

How it uses Jev

stuntd speaks the Jev System One protocol (POST /v1/systemone) and the OpenAI Chat Completions API. It records every Choice, Score and Noul decision, trains a head per site, and answers live with calibrated confidence, falling back to the upstream below its threshold.

Primitives:choicescorenoul

Technique worth stealing

Distil each decision site into a small head on a frozen Laya encoder, serving locally with calibrated confidence and falling back to the provider when unsure.

Try it

pip install "stuntd[train]" && stuntd config init && stuntd serve

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-centricyes0.88
  • Shows a System One patternyes0.91
  • Handles uncertaintyyes0.92
  • Measuredyes0.97
  • Runnableyes0.96
  • Worth recommendingunclear0.54
  • Model replicayes0.84
  • Problem scopescore on a 0–2 scale1.18
  • About Jevyes0.99

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.