User repository on GitHub
- Forks
- 9
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Tools & CLIs
System One Lite
A proof-of-concept HTTP service for constrained decisions with a local MLX model: send text or JSON plus typed choice, score, or noul questions and it returns probabilities over the answers you declared — no generated text, no parsed model JSON.
- Stars
- 50
- Language
- Python
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- snellingio
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceRuns locally
Highlights
- README: takes a state plus typed questions and returns probabilities over your declared answers, reading the model's scores at a fixed answer position.
- Three question types: choice (winning key, probability per option), score (probability-weighted score per level), and noul, a yes-or-no judgment.
- Questions are evaluated independently — each gets its own prompt and a fresh copy of the state — preventing influence between answers while growing token use.
- Named uses: routing a support request, scoring urgency, deciding whether a record needs human review, and combining judgments in application code.
- Requirements: an Apple silicon Mac (MLX), Python 3.12+, and uv; MIT, 50 stars, 9 forks, created 2026-09-16, last pushed 2026-09-18.
Watch out
README reviewed 2026-10-17 and truncated inside the requirements list, so install commands, the specific model used, and any benchmarks are not in the digest. The excerpt never mentions Jev or TypeSafe: it is an independent experiment with stock open-weight models, and the connection is System One semantics — typed questions, probability answers, and its own noul boolean type.
Questions about this entry
Short answers drawn from the linked project or page.
A proof-of-concept HTTP service from snellingio for constrained decisions with a local language model: send text or JSON plus typed questions, and it returns probabilities over the answers you declared instead of generating a text response.
The README says it is an independent experiment built with stock open-weight models and MLX and never mentions Jev, TypeSafe, or an API. The connection is the System One pattern: typed questions in, probabilities out, and noul as its boolean type.
Choice returns the winning key, a probability for every option, and confidence; score returns a probability-weighted score plus probability per level; noul returns the probability of yes — the project's name for a boolean judgment that carries uncertainty.
No: the README says 0.9 means the model assigned 90% of the allowed probability mass to yes, not that the answer has been calibrated to be correct 90% of the time. Treat it as mass on yes, not a correctness rate.
It formats the state, instructions, and allowed answers into a prompt, maps answers to token codes such as A, B, and C, runs the model to the fixed answer position without decoding text, keeps only the logits for valid codes, and applies softmax.
The README says it is not a general chat API, a text generator, or a multi-step reasoning system. Each question re-sends the state in its own prompt, so runtime and input-token use grow with the number and length of questions, and it requires an Apple silicon Mac.
An Apple silicon Mac running MLX, Python 3.12 or newer, uv, and network access on first run to download the model. MIT licensed; 50 stars, 9 forks; created 2026-09-16 and last pushed 2026-09-18, two days later.
More Jev resources like this
A local workbench for System One-style models on Apple Silicon Macs: it serves Laya through MLX or Core ML behind a Jev-compatible /v1/systemone endpoint, proxies other Jev-compatible servers like Jeff and Kev, recommends models and evaluates them on Japanese data.
Jev Tools & CLIs#community#github#python
Community
katya4oyu/systemone-workbenchPython
katya4oyu/systemone-workbenchPythonAn open-source platform for creating, distributing, and running sovereign EU-compliant LLMs: one Rust binary with OpenAI- and Ollama-compatible APIs, decisions answered as probabilities in milliseconds, and an audit trail designed for the EU AI Act.
Jev Tools & CLIs#community#github#local-llm
Community
eullm/eullmPython
eullm/eullmPythonA local workbench that pairs OpenCV vision with Laya or Jev decision models: images are segmented and measured, hard rules settle the numbers, and the model gives the verdict — with one-click installers and an "open-source JEV" training path served at /v1/systemone.
Jev Tools & CLIs#community#github#python
Community
uu889/laya-opencvPython
uu889/laya-opencvPythonA Chinese-language decision model that returns a probability for each fixed candidate option in one forward pass, with no generated text or chain of thought. It is a LoRA fine-tune of Qwen3.5-0.8B on 13,048,000 CJ-Dataset rows, benchmarked on CJ-Bench against Chinese-Jev.
Jev Models & Reimplementations#community#github#python
Community
yanqiangmiffy/chinese-decisionPython
yanqiangmiffy/chinese-decisionPythonAn Ollama-style local runner for System One decision models from Hugging Face: pull, serve and query them behind the same /v1/systemone API as TypeSafe's Jev, with a terminal model manager and a browser playground.
Jev Tools & CLIs#community#catalog-import#madewithjev
Community
nvkudva/ollajevPython
nvkudva/ollajevPythonA local, non-autoregressive security guardrail suite powered by Convai Innovations' Laya model, returning typed decision probabilities for security inspection. It positions itself against Jev on latency and against generative LLMs on hallucination risk.
Jev Tools & CLIs#community#github#security
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.