poorjev

poorjev</h1>

<p align="center"><b>The poor man's Jev.</b> An open source, local-first "System One" decision layer for LLM apps: typed decisions with <b>provably calibrated confidence</b>. No API key. No waitlist.</p>

<p align="center">
  <a href="https://pypi.org/project/poorjev/"><img src="https://img.shields.io/pypi/v/poorjev" alt="PyPI version"></a>
  <img src="https://img.shields.io/badge/license-MIT-green" alt="MIT license">
  <img src="https://img.shields.io/badge/python-3.10%2B-blue" alt="Python 3.10+">
  <a href="https://github.com/rupeshpoojary9/poorjev/actions/workflows/tests.yml"><img src="https://github.com/rupeshpoojary9/poorjev/actions/workflows/tests.yml/badge.svg" alt="tests"></a>
  <img src="https://img.shields.io/badge/ECE-0.170%20%E2%86%92%200.071-orange" alt="ECE 0.170 to 0.071">
  <img src="https://img.shields.io/badge/API%20key-not%20required-blueviolet" alt="no API key required">
</p>

---

**Your model's `0.9` is a vibe. poorjev's `0.9` is a measurement.**

Every LLM-in-JSON-mode hands you a confidence score and hopes you don't check it. poorjev checks it. On the shipped eval set it cuts calibration error (ECE) from **0.170 to 0.071** with zero loss of accuracy, and it runs on your laptop with no API key.

<p align="center">
  <img src="https://raw.githubusercontent.com/rupeshpoojary9/poorjev/main/docs/reliability_before_after.png" alt="Reliability diagram: raw confidences are overconfident, calibrated confidences hug the diagonal" width="760">
</p>

<p align="center"><i>Left: raw confidences, overconfident. Right: calibrated, a stated 0.8 really is right about 80% of the time.</i></p>

## Quickstart

```bash
pip install "poorjev[local]"
```

```python
from poorjev import Client, Choice, Score, Noul

client = Client()  # local model, no key, of