OpenJevPro
OpenJevPro: typed Choice/Noul/Score API over open LLMs, Laya or Jev with order-invariant scoring and a System 2-to-1 distillation flywheel.
Overview
OpenJevPro presents itself as a high-throughput decision engine for AI agents and a Jev alternative. It offers Choice, Noul and Score primitives over open LLMs, Laya (ModernBERT) or commercial Jev, with two headline features: order invariance by scoring each candidate in isolation, and an explore-first, crystallise-later flywheel that distils System 2 reasoning traces into cached System 1 fast paths. Deployment can be fully local (vLLM, SGLang, Laya) or a hybrid gateway with Jev behind a circuit breaker.
The README makes strong marketing claims such as sub-35 ms decisions. A list summary describes the mechanism more plainly: it asks an Ollama or OpenAI-compatible model to write a likelihood score per candidate and softmaxes them with a fixed temperature, and notes the licence is PolyForm Noncommercial, not an open-source licence. It installs via pip and includes a CPU-only CLI demo.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which candidate fits the state? | — |
| Yes / no | Is this statement true? | yes · no |
| Score | Level on an ordered scale | — |
How it works
- Score each candidate in isolation with the selected backend.
- Softmax the scores with a fixed temperature.
- Cache distilled decisions for repeated situations.
- Fall back between local and Jev via the gateway.
Strengths
- Order invariance by isolated candidate scoring.
- Multiple backend options including local.
- Caching of repeated decisions.
Caveats
- PolyForm Noncommercial licence per the list summary.
- Performance claims are marketing-style and unverified here.
- LLM-written likelihoods are not inherently calibrated.
Getting started
pip install openjevpro, then run the zero-GPU CLI demo.At a glance
- Type
- Service
- Status
- prototype
- Language
- Python
- Runtime
- library, CLI, gateway
- Has code
- yes
- License
- PolyForm-Noncommercial-1.0.0
- Stars
- 42
Use cases and integrations
How curated lists describe it
Self-evolving cognitive decision engine & TypeSafe Jev alternative. Deliberative decision flywheel ('explore first, crystallize later' System 2→1) with 100% option-order invariance. Typed probabilistic API (Choice, Noul, Score) for Open LLMs, Laya (ModernBERT), & commercial Jev with calibrated logprobs and adaptive safety guards.
wh000wh000/awesome-jev-liveOpenJev: Open-source alternative to TypeSafe Jev. Typed probabilistic decision API (Choice, Noul, Score) powered by open LLMs & constrained logprob calibration.
hellogumbo/awesome-jevAsks an Ollama or OpenAI-compatible model to write a likelihood score per candidate, then softmaxes them with a fixed temperature. PolyForm Noncommercial, not an open-source licence.
MrJev/awesome-jev
Frequently asked questions
How are probabilities produced?
Per the list summary, an LLM writes a likelihood per candidate and these are softmaxed with a fixed temperature.
Is it open source?
The list summary says it uses PolyForm Noncommercial, which is not an open-source licence.
Can it use real Jev?
Yes, as a backend in its hybrid gateway mode.
Related projects
llama.cpp-based C++ server that turns any LLM into a System One decision model, reading answers from next-token logits.
ProjectbetaC++MLX 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.
ProjectresearchRustEducational C11 library and CLI that turns DiffusionGemma 26B into a CPU-first OpenJev-style decision engine reading candidate logits.
ProjectresearchCExperimental engine that makes an existing LLM do Jev-like classification and scoring by reusing context and building JSON.
ProjectresearchPythonReads typed decisions from one forward pass of PrismML's 1.75-bit Bonsai 27B under llama.cpp, using the OpenJev/SemIf logit method.
ProjectresearchPython