openjev (lookski)
Turns any local LLM into a Jev-style engine via masked-logit softmax, or proxies OpenAI-compatible APIs or Jev; includes a chat crush radar.
Overview
OpenJev (lookski) turns a local LLM into a Jev-style decision engine running fully offline. It reads raw logits at the first answer position, masks them to the allowed answer tokens and applies a softmax, giving type-safe answers with probabilities in one forward pass and no decoding. The same interface can instead point at any OpenAI-compatible API (OpenAI, OpenRouter, DeepSeek, Groq, self-hosted vLLM) or at the official Jev cloud, switched with one flag.
The README shows a CPU-only Qwen3-0.6B run classifying a Stripe support message as technical at 0.999 with urgency and frustration answers, taking about 1.2 s for three forward passes, and compares local latency of about 0.1-3 s at no cost with Jev's 70-500 ms and per-token price. Per the list summary, it also includes a crush radar for WeChat, QQ and Telegram chats that returns a verdict and next-move advice. It is MIT licensed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | department | billing · technical · sales |
| Score | frustration | — |
| Yes / no | is_urgent | Yes · No |
How it works
- Pick a backend: local model, OpenAI-compatible API or Jev cloud.
- Send state and typed questions.
- Locally, mask answer-token logits and softmax.
- Act on the type-safe answer and probability.
Strengths
- Zero cost and private when local.
- One interface across local, API and Jev backends.
- Runs on CPU with small models.
Caveats
- Local latency is slower than Jev (about 0.1-3 s).
- Probabilities from raw logits are not necessarily calibrated.
Getting started
Clone the repo, install the package and run python demo.py with a local model such as Qwen3-0.6B.At a glance
- Type
- Service
- Status
- prototype
- Language
- Python
- Runtime
- library, HTTP service
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Turn any local LLM into a Jev-style System One decision engine: type-safe answers with raw softmax probabilities. 100% offline, zero API cost.
logicrw/awesome-jev-projectsOpen decision engine: type-safe answers with raw softmax probabilities from a local LLM or any OpenAI-compatible API. Crush radar for WeChat/QQ/Telegram chats - verdict + next-move advice. | 开放决策引擎: 本地 LLM 或任意 OpenAI 兼容 API, 原始 softmax 概率 + 类型安全回答. 微信/QQ/Telegram 追人雷达: 冲/稳/缓/停 + 下一步建议.
hellogumbo/awesome-jev
Frequently asked questions
How does it work locally?
It masks the logits at the first answer position to allowed tokens and applies a softmax.
Can it call Jev itself?
Yes, the official Jev cloud is one of the selectable backends.
What is the crush radar?
Per the list summary, a chat analyser for WeChat, QQ and Telegram that gives a verdict and next-move advice.
Related projects
llama.cpp-based C++ server that turns any LLM into a System One decision model, reading answers from next-token logits.
ProjectbetaC++Python library that imitates Jev-style typed decisions (choice, score, yes/no) on local LLMs via Ollama, vLLM or llama.cpp.
ProjectprototypePythonMLX library that turns an existing local model into a decision engine by reading candidate logits, with no fine-tuning, on Apple Silicon.
ProjectprototypePythonLocal server (xks) implementing Jev's /v1/systemone contract on an open-weight LLM, accelerated with Xeon Phi cards.
ProjectresearchRustA chatbot with no language model: every word is chosen by Jev Choice questions from thousands of options, with lookahead and self-critique.
ServiceresearchPythonEducational C11 library and CLI that turns DiffusionGemma 26B into a CPU-first OpenJev-style decision engine reading candidate logits.
ProjectresearchC