User repository on GitHub
turn any llama-server into a jev system one endpoint
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Models & Reimplementations
Verdict — khimaros
A Python server in front of llama-server that reads one-token label probabilities and exposes them as POST /v1/systemone.
- Stars
- 8
- Language
- Python
- License
- GPL-3.0
- Reading time
- 3 min
- Published by
- Community
- Author
- khimaros
- Last verified
- Sep 2026
About this project
an open source, self-hosted alternative to typesafe's jev and its system one api. verdict serves the same wire protocol on your own hardware, so an existing jev client changes its base url and nothing else. a small python server sits in front of your existing llama-server or llama-swap endpoint with any model of your choosing and serves a jev compatible API. answer typed questions about a piece of text without generating any text.
Quoted from the project's README. Read the full README.
Highlights
- An existing Jev client changes TYPESAFE_BASE_URL. browser-use/jev-ultrafast is the conformance check.
- Option mass is the probability on the label tokens before renormalising, kept as a health signal.
- On a quiet server, eight models reproduced scores exactly. Under load, 5 of 16 prompts differed by up to 3.2e-02.
- A model that fails an alphabet check is refused rather than served.
- The README says gemma-4-e2b reported 1.000 confidence on wrong answers, so raw confidence is not safe to gate on.
Quickstart
bash
PYTHONPATH=python python3 -m llama_verdict.server \
--base-url "$LLAMA_VERDICT_URL" --model "$LLAMA_VERDICT_MODEL" --port 8477Watch out
GPL-3.0-or-later. Independent of TypeSafe and makes no accuracy claim against hosted Jev. Needs a running llama-server. Linking the planned library would carry the GPL.
Questions about this entry
Short answers drawn from the linked project or page.
That is the design. verdict serves the same wire protocol on your own hardware, POST /v1/systemone, so an existing client changes its base URL and nothing else. It is a small Python server in front of your own llama-server or llama-swap endpoint with any model you choose, and it needs a running llama-server with a chat model.
verdict makes no accuracy claim against hosted Jev. What it measures is jevbench's 231 public decisions per model, each read the way it was trained and without debiasing: jev-omni:Q8_0 leads at 204 of 231, jevk5-4b at 199, and qwen3.5-0.8b trails at 131. These ran on an iGPU reached over a roughly 150 ms VPN, so the latencies are not comparable with the datacenter jevbench board.
Not on raw confidence. gemma-4-e2b reported 1.000 confidence on wrong answers, and under server load 5 of 16 prompts came back different by up to 3.2e-02 even though a quiet server reproduced scores exactly across eight models. verdict keeps option mass, the probability landing on the label tokens before renormalising, as the health signal, because a model can rank options perfectly on an option mass of 1.7e-08.
Past A-Za-z, labels come from the model's own single-character vocabulary under the opt-in --wide-alphabet flag, and it is not free. With every label unfamiliar, only qwen3.5-9b held 16 of 16; minicpm5-2b fell to 9 of 16 with mass dropping to 0.6991. Median decision latency also rose from 1033 ms to 2596 ms. A model that fails the alphabet check is refused rather than served.
GPL-3.0-or-later. The README warns that linking the planned library would carry the GPL, which matters if you want to embed it rather than run the server beside your application.
One forward pass reads the probability the model puts on each declared option label and renormalises over them. The answer is a distribution over option IDs plus the confidence signals, and no tokens are sampled, so there is no seed to set.
Your data does not leave the machine and the cost is electricity rather than a per-call price, because verdict runs on your hardware with any gguf you already have. The wire protocol stays the same, so a Jev client only changes its base URL.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsOther listings of this project
Forks and copies of this project in the directory. This entry is the canonical one.
Python decision models meant to replace LLM calls for routing, triage, guardrails and policy checks: ask choose, score or check questions and get a typed answer with a temperature-scaled probability on CPU.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
Manavarya09/verdictPython
Manavarya09/verdictPythonMore Jev resources like this
A local /v1/systemone server that reads label probabilities from a GGUF chat model, with prompt repetition and a confidence shrink toward uniform.
Jev Models & Reimplementations#community#python#open-models
Community
Mushroom-Systems/lichenPython
Mushroom-Systems/lichenPythonA local yes/no decision model that speaks Jev's wire format on a laptop CPU, refusing Choice and Score until those question types ship.
Jev Models & Reimplementations#community#python#open-models
Community
feder-cr/jevC++
feder-cr/jevC++A local server that reads next-token probabilities from a fine-tuned Spark model and returns Choice, Score, and Noul answers with zero generated tokens.
Jev Models & Reimplementations#community#python#open-models
Community
Rizzo-AI-Academy/rizzo-flowPython
Rizzo-AI-Academy/rizzo-flowPythonA local daemon that pulls open decision models and serves them on TypeSafe's /v1/systemone, so a Jev client switches hosts with TYPESAFE_BASE_URL.
Jev Models & Reimplementations#community#rust#open-models
Community
ollaya-dev/ollayaRust
ollaya-dev/ollayaRustAn Apache-2.0 decision engine with a TypeSafe-compatible endpoint, document answers with source passages, and a ViZDoom demo that picks actions from structured game state.
Jev Models & Reimplementations#community#python#local
Community
deepanwadhwa/OpenDecisionPython
deepanwadhwa/OpenDecisionPythonA self-hosted System One server: one /v1/systemone contract, a Docker image each for Laya and kev, and a playground where snake, dino, and tetris call that endpoint.
Jev Models & Reimplementations#community#python#local
Community
chaitin/DecisPython
chaitin/DecisPythonFrom 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.