How to get Jev-style typed decisions out of any hosted LLM provider today, no new model needed

notjev is a Node library that makes any OpenAI-compatible model produce fast multiple-choice decisions. It requests one token with logprobs enabled, renormalizes the letter probabilities, and returns a verdict, probability, margin, and abstention threshold. This gives Jev-style Choice/Score/Noul outputs to developers running local models (vLLM, llama.cpp, Ollama) without runtime dependencies.

Products·launched

Jevless: Jev-style typed decisions (Choice/Noul/Score) from any logprobs-exposing model API

Jevless is an open-source tool that produces Jev-style typed decisions — Choice, Noul, and Score outputs — using any LLM whose API exposes logprobs, rather than requiring a single supported backend. It also ships a local /v1/systemone-compatible server. The author built it to keep Jev's low cost, speed, and bias-canceling accuracy while working across as many model backends as possible.

Jevless is an open-source tool that produces Jev-style typed decisions — Choice, Noul, and Score outputs — using any LLM whose API exposes logprobs, rather than requiring a single supported backend. It also ships a local /v1/systemone-compatible server. The author built it to keep Jev's low cost, speed, and bias-canceling accuracy while working across as many model backends as possible.

Products·launched

hearim: open-source gateway turning ordinary local LLMs into a Jev-compatible decision API

hearim is an open-source gateway that lets existing local or hosted LLMs act as a Jev-compatible decision API. Given data and a set of choices, it returns a decision with probabilities without task-specific fine-tuning or custom weights. It targets developers running models via llama.cpp, vLLM, SGLang, Ollama, or OpenAI-compatible servers.

hearim is an open-source gateway that lets existing local or hosted LLMs act as a Jev-compatible decision API. Given data and a set of choices, it returns a decision with probabilities without task-specific fine-tuning or custom weights. It targets developers running models via llama.cpp, vLLM, SGLang, Ollama, or OpenAI-compatible servers.

Releases·Tool Release

Nokia Applied Research releases AnyJev, turning any LLM into a Jev-style typed-decision model with real probabilities and no training

Nokia Applied Research, with a Tencent Hunyuan collaborator, released AnyJev, an open-source Python package (Apache-2.0, on PyPI) that converts any LLM into a Jev-style decision model producing typed decisions and real probabilities with no training. The release demonstrates that raw logit readout is order-sensitive — Qwen3-8B flips a BANKING77 answer with 1.00 confidence when options are reversed — while AnyJev's L0 mode stays consistent with zero labels.

Nokia Applied Research, with a Tencent Hunyuan collaborator, released AnyJev, an open-source Python package (Apache-2.0, on PyPI) that converts any LLM into a Jev-style decision model producing typed decisions and real probabilities with no training. The release demonstrates that raw logit readout is order-sensitive — Qwen3-8B flips a BANKING77 answer with 1.00 confidence when options are reversed — while AnyJev's L0 mode stays consistent with zero labels.

Social·Community

JEV explained: TypeSafe's decision model reads once and scores all allowed answers — no hidden 'none of these' option

A Diamant AI newsletter post explains the internals of JEV, a decision model from TypeSafe. Unlike an LLM loop, JEV reads once and scores every allowed answer in parallel. Its output probabilities must land on one of the supplied options, so practitioners should always include an escape answer.

A Diamant AI newsletter post explains the internals of JEV, a decision model from TypeSafe. Unlike an LLM loop, JEV reads once and scores every allowed answer in parallel. Its output probabilities must land on one of the supplied options, so practitioners should always include an escape answer.