What is JEV-Local and when would I use it?

A lightweight Python library and server that mimics Jev's System-1 pattern on local LLMs through Ollama, vLLM or llama.cpp. It returns structured decisions only, choice, score or noul, with probabilities and confidence instead of prose.

Does JEV-Local use the real Jev model?

No. The README describes the cloud Jev API as returning structured decisions in about 70-500 ms and says this library brings that pattern to local models; the judgments come from whatever local model you point it at.

Which backends does it support?

Any OpenAI-compatible endpoint: Ollama, vLLM and llama.cpp are called out. It auto-detects the backend from the endpoint port, and the README's example runs llama-server with a Qwen2.5-Coder 7B GGUF on port 8080.

How do I install and start the server?

Install httpx and pydantic, then jev-setup asks for a JEV port (default 3030) and installs a systemd user service that starts with your session. Check it with curl on /health and /v1/models, and manage it with systemctl --user.

What question types can I ask?

Choice (pick one), score (0-10) and noul (yes/no gate), and several questions can run in parallel in a single LLM call. The README's criteria example shows fill_email, fill_password and click_submit as a login-state Choice.

How does it connect to editors and agents?

Through the OpenAI-compatible gateway: set the provider to OpenAI Compatible, base URL http://127.0.0.1:3030/v1, model jev-local, and an empty or any local value as the API key.

What are the limitations of JEV-Local?

It only returns JSON with no explanations, and result quality depends on the local model you choose. Parts of the README are Turkish, the digest's copy cuts off in the question-types section, and GitHub lists no licence for the repo.

Jev directory

CommunityJev Repos & SDKs

JEV-Local

A lightweight Python engine that mimics Jev's structured decisions on local LLMs via Ollama, vLLM or llama.cpp: choice, score and noul questions come back as JSON with probabilities and confidence, plus an OpenAI-compatible gateway.

See highlights Open github.com

Stars: 1
Language: Python
Reading time: 2 min
Published by: Community
Author: tapsin
Last verified: Oct 2026

tapsin/jev-local

User repository on GitHub

View on GitHub

JEV-Local: System-1 decision engine for local LLMs. Mimics TypefAI JEV: structured choices only, no chatter, calibrated confidence. Works with Ollama, vLLM, llama.cpp, LM Studio.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity catalog-import logicrw local-llm ollama openai-compatible

Works forJev for Classification Jev for Routing Jev for Scoring

Used inJev for Software & DevTools

Languages: Python
Platforms: API CLI
Capabilities: Open sourceRuns locally

Highlights

Returns only JSON with choice, probabilities and confidence, never explanations; probabilities sum to 1.0 and confidence stays in [0,1].
Backend-agnostic over OpenAI-compatible endpoints: auto-detects the backend from the port; works with Ollama, vLLM and llama.cpp.
Serves a jev-local model on /v1/models and /v1/chat/completions for editors and agents; jev-setup installs a systemd user service on port 3030 by default.
Only dependencies are httpx and pydantic; JSON parsing handles fenced blocks, thinking tokens and partial output.
Supports parallel questions in one LLM call and a decide_action helper that picks from a list of actions.

Quickstart

bash

python3 jev_local.py \
  --state "Browser on login page with email, password, submit button" \
  --model qwen2.5-coder-7b-instruct-q4_k_m \
  --endpoint http://localhost:8080 \
  --backend openai

Watch out

Grounded in the truncated README and GitHub metadata reviewed on 2026-10-15; the README cuts off in the question-types section, parts of the documentation are Turkish, the GitHub licence field is empty, and output quality depends entirely on the local model you point it at.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

The Open Reimplementations: Local Models That Speak System One

Sep 22, 2026

Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.

Jev Repos & SDKs#open-models#research#local

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai