How is this different from a normal classifier?

It returns the full distribution with confidence and a conformal set that holds the right answer with high guarantee, saying when a case is ambiguous so you can automate single-option sets and send the rest to a person.

Does it need the cloud or pay per call?

No: it runs locally offline on CPU, NVIDIA GPU, or Apple Silicon with no external service and no per-call cost, needing free RAM for the quantized base and more under PyTorch.

What setup and memory should I expect?

Create a virtualenv, install requirements for PyTorch or MLX, then run the server script with an execution directory; MLX needs modest free RAM while PyTorch needs more, with a smaller model option for weak machines.

Can I trust the calibration and reproduce training?

Training targets calibration directly with soft targets and a scale penalty, and the repo documents data generation, majority-vote relabeling, and train scripts that require Apple Silicon plus an OpenAI-compatible key for synthesis.

Which Brier version should I use?

Use v3 by default, especially for text from third parties, since it is the most resistant to instruction injection; use v2 when the text is trusted and accuracy matters most, and rapido when speed matters more than accuracy.

What question types and limits does it support?

Three: choice over up to 26 options, score over 2 to 10 ordered levels, and noul for yes or no; confidence runs from 0 for a uniform distribution to 1 for outright certainty.

What is its instruction-injection resistance?

v3's malicious instruction changed the answer in 16.7% of 1000 cases against Jev's 21.4%, a significant difference; in a blind test of 57 unseen phrases it held 17.5% against 20.5%, smaller and not significant.

CommunityJev Tools & CLIs

Brier

Open local decision model on Qwen3-4B with LoRA adapters that answers choice, score, and noul questions with probabilities, calibrated confidence, and conformal sets, offline with no per-call cost.

See highlights

Open github.com

Stars: 8
Language: Python
Published by: Community
Author: Mangaba-ai
Last verified: Oct 2026

Mangaba-ai/brier

Organization repository on GitHub

View on GitHub

Brier: decisões tipadas e calibradas (choice/score/noul), locais e independentes. Qwen3-4B + LoRA em MLX; API /v1/decide, compatível com /v1/systemone.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Modelo aberto de decisão tipada e calibrada. Roda localmente e não depende de nenhum serviço externo. O Brier recebe um conteúdo (state) e uma ou mais perguntas com respostas pré-definidas. Para cada pergunta, devolve a opção escolhida e a probabilidade de cada opção. Ele não gera texto: responde sempre dentro das opções que você definiu, e diz o quanto tem certeza.

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Scoring

Used inJev for Software & DevTools

Languages: Python

Highlights

Takes state plus typed questions in one pass with block attention, reading option tokens directly so answers stay in type.
Returns choice with probabilities, score as a weighted average between levels, and noul as yes-probability, plus uncertainty and set fields.
Trained on synthetic Portuguese cases plus public sets with majority-vote labels, tuned for calibration with conformal, dropout, and permutation layers.
Runs on Windows, Linux, and macOS via PyTorch or MLX with the base downloading on first run; tests run without a model and CI covers all three systems.

Watch out

CPU runs but is slow for a 4B model, so real use prefers GPU; stochastic samples multiply latency. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

paper-radar-jev

Python CLI that pulls recent arXiv papers, scores relevance with TypeSafe and writes daily Markdown and JSON rankings with resume-safe storage.

Jev Tools & CLIs#community#github

Community

LYchoon/paper-radar-jevPython

rwkv-jev

RWKV-7 decision-model server exposing a Jev-compatible /v1/systemone endpoint, with constrained decoding over answer tokens and zero output tokens.

Jev Tools & CLIs#community#github

Community

XingQiPan/rwkv-jevPython

jev-bridge

Jev-style /v1/systemone API over any OpenAI-compatible backend, reading single-token label logprobs with a restricted softmax instead of generating JSON.

Jev Tools & CLIs#community#github

Community

TOSUKUi/jev-bridgePython

pyGPT

A desktop AI assistant with a Jev / System One inline plugin for classification, routing, verification, and scoring without chat-style output.

Jev Tools & CLIs#community#python#desktop

Community

szczyglis-dev/py-gptPython

judgekit

CLI and Python library that runs declarative YAML judgment tasks - classify, route, score, verify - natively on Jev Choice/Score/Noul, on any OpenAI-compatible backend, or on local keyword rules, with confidence gates, fallbacks and per-decision cost records.

Jev Tools & CLIs#community#yaml#classification

Community

lexingtonhibiki/judgekitPython

jev-tmmluplus-eval

Harness that sits Jev for the 66-subject Traditional Chinese TMMLU+ suite, asking every item as one four-way A/B/C/D choice with calibration reporting.

Jev Tools & CLIs#community#github

Community

lianghsun/jev-tmmluplus-evalPython

From the Jev guides

Original write-ups on what this entry is for.

Jev on the Gateways: Vercel, Workers AI, and OpenRouter

Oct 12, 2026

The request shape barely changes across surfaces; the surrounding contract does. When to call the API directly and when a gateway, a Worker binding, or a router entry earns its extra hop.

Jev Tools & CLIs#gateway#vercel#cloudflare

Read article

Exporting Decision Models: GGUF, ONNX, and Device Runtimes

Oct 10, 2026

Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.

Jev Tools & CLIs#open-models#local#onnx

Read article

Records at Rest: Documents, Inboxes, and Risk Queues

Sep 28, 2026

Jev labels a record that already exists. Code moves, flags, or holds it. How the document, mail, and transfer tools keep the action out of the model.

Jev Tools & CLIs#classification#email#security

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai