Can I swap this for hosted Jev later?

Yes by design. POST /v1/systemone matches the TypeSafe contract so a later API key is a base-URL swap; response fields stay limited to model, answers, and usage tokens.

When does a request fail over to cloud?

On missing weights, load failure, timeout, 503, or exception within a 12s hop timeout and short breaker; bad client requests as 4xx never fail over.

Should I add the Laya hop for quality?

No for general quality. The digest calls Laya honest but near chance zero-shot and a specialist for synthetic workflows, keeping official cloud as the backup to trust.

What breaks calibration most often?

Changing model, prompt, or label mix without refitting; temperature applies at softmax only and cannot fix systematically wrong ranking, so refit after eval exists.

What do I need installed before it will answer anything?

Apple Silicon, Python 3.11 to 3.13 with 3.12 pinned, and uv sync --extra mlx --extra dev. The weights are not in the repo: the default LFM2.5-1.2B MLX 8-bit checkpoint is about 1.2 GB from Hugging Face, or mlx-lm fetches the Hub id on first serve. The code is MIT, but you need an LFM license from Liquid AI for the weights.

How is an answer produced without decoding?

A shared state prefix is followed by an isolated prompt per question, and the model is prefilled to the Answer: boundary. Allowed continuations are scored straight from logits: yes/no sequence logprobs give P(yes) for a Noul, option-key sequence logprobs give a Choice, and level numbers 0 to n-1 give an expected value for a Score. Nothing is generated and no free JSON is parsed.

Can I test it before downloading the model?

Yes. The contract tests mock the scorer and are required to stay green without the LLM weights; a live smoke is opt-in through JEVONS_LIVE=1 pytest -m live or uv run jevons smoke. The bundled eval seed is about 95 hand-authored items with static teacher labels, scored on modal agreement, Brier, ECE, schema validity and accuracy at confidence 0.8 or above.

CommunityJev Models & Reimplementations

Jevons

Embeddable System One server matching TypeSafe HTTP shape, scoring LFM2.5-1.2B logits in code with optional Laya and cloud failover.

See highlights

Open github.com

Stars: 0
Language: Python
License: MIT
Published by: Community
Author: gopalanj
Last verified: Oct 2026

gopalanj/jevons

User repository on GitHub

View on GitHub

Local System One for typed decisions. Scores noul, choice, and score from option logits — never free JSON. TypeSafe-compatible POST /v1/systemone with layered routing: on-device LFM2.5 first, optional second encoder, cloud Jev only if local fails. Fine-tune on proprietary labels. Not official Jev.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Local, embeddable System One server. POST /v1/systemone matches the TypeSafe HTTP contract so a later API key is a base-URL swap. This is not official Jev. Official Jev cannot be cloned. Jevons scores allowed outcomes from logits and assembles {choice, probabilities, confidence, noul, score, legend} in code. It never asks the model to write JSON.

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

Tagscommunity github

Works forJev for Routing Jev for Scoring

Used inJev for Software & DevTools

Languages: Python

Highlights

Default LFM2.5-1.2B via MLX 8-bit on Apple Silicon; repo ships code only and needs a Liquid AI license for weights.
Scores allowed continuations from logits with no decoding; aliases jev-latest and jev-local map to local scoring.
Router tries local, laya, cloud in order with timeout and breaker; 4xx validation errors never fail over.
Ships contract tests that stay green without weights plus eval, temperature calibration, and optional LoRA path.

Watch out

Not official Jev; zero-shot logprobs rank without calibrated training, and Laya hop is near chance zero-shot per model card. 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

ComfyUI-ScriptFlow

ComfyUI script nodes offering a restricted Python-like interpreter with multiple text and numeric ports, plus a Jev variant that branches on yes, choice, and score answers.

Jev Models & Reimplementations#community#github

Community

kantan-kanto/ComfyUI-ScriptFlowPython

Bespoke Nimble

2.2k

Open recipe for a local Jev-style decision model: a Qwen3.5-9B LoRA that scores one allowed answer token per Choice, boolean or rubric Score question, with data pipeline, training config and eval harness. Reports 90.1% on its 324-example holdout against Jev's 93.2%.

Jev Models & Reimplementations#community#lora#qwen

Community

bespokelabsai/nimblePython

Julia 1

Supersonic Labs' 144.3M-parameter decision model on mmBERT-small: choice, score, and noul answers from supplied options in one pass, 73.15% on 2,000 typed decisions, Apache-2.0.

Jev Models & Reimplementations#community#open-models#local

Community

Liquid AI d1-3B

Liquid AI's 3.12B decision model built on LFM2.5-VL: state plus typed questions in one forward pass with zero output tokens, 48.57 on Decision Index 0.2.1 and 8 ms per decision on an RTX 4090.

Jev Models & Reimplementations#community#open-models#local

Community

Decider Chat Gemma-4-31B

An untrained readout: stock Gemma-4-31B weights plus a decider config that reads option letters at the answer slot, scored #2 of 70 on Decision Index v0.2.1 at 57.33 against Jev 1.13.0's 57.91.

Jev Models & Reimplementations#community#open-models#local

Community

Drex DLM

Nace.AI's 8B decision model on NVIDIA's Efficient-DLM diffusion backbone: a 256-dim pointer head scores every option in one pass, behind the same POST /v1/systemone API as its hosted service.

Jev Models & Reimplementations#community#open-models#local

Community

From the Jev guides

Original write-ups on what this entry is for.

Multimodal Decision Models: What Clef and d1 Change

Oct 12, 2026

Jev takes text only; Clef-omni adds image, audio, and video input and Liquid d1 adds images, each keeping the same typed-answer shape. Where that matters, where it does not, and the costs the vendors underplay.

Jev Models & Reimplementations#multimodal#open-models#decision-model

Read article

The Decision-Model Price War

Oct 11, 2026

One model in September, a dozen by October. Clef, Liquid d1, Perplexity, and OpenAI all ship the same typed-answer shape now. How to read the rival benchmarks and pick on your own task.

Jev Models & Reimplementations#decision-model#benchmarks#pricing

Read article

What Is Actually on Hugging Face

Oct 8, 2026

Hundreds of repositories carry the system-one tag, but the ecosystem is a dozen ideas in different sizes. A tour of the families, the licenses, and what a download does not include.

Jev Models & Reimplementations#open-models#hugging-face#local

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai