CommunityJev Tools & CLIs
New in llama.cpp: Decision Models
ggml's announcement that llama.cpp server supports decision models through a /v1/systemone endpoint: send a state and typed questions, get a probability per option from one forward pass. It lists six GGUF decision models from Julia-1 (144M) to OpenJev (27B) with single-question latencies.
- Format
- Article
- Published by
- Community
- Last verified
- Oct 2026
Highlights
- Adds a /v1/systemone endpoint to llama.cpp server: a state (text, JSON, image) plus choice, score, or noul questions, answered with probabilities.
- Lists six supported GGUF models: Julia-1 (144M, 3 ms), Laya (421M, 5 ms), Kev-4B (12 ms), lev (36 ms), OpenJev (27B, 43 ms), and Clef (27B).
- A duplicate-charge example returns route billing 0.9049, angry 0.8208, and urgency score 2.2821 with per-level probabilities.
- Describing criteria changed Julia-1's route for a duplicate charge from shipping to billing at 0.99; timings are single-question medians on one RTX PRO 6000.
Watch out
Written by ggml-org contributors as a Hugging Face community article, and the speed column in its model table was measured on one GPU. Commenters report batch-size limits when inputs run into thousands of tokens.
Questions about this entry
Short answers drawn from the linked project or page.
Three: choice returns the top option with a probability for each option; score places the state on an ordered rubric of 2 to 10 levels and returns the expected level; noul returns the probability that a yes/no question is true.
Six GGUF models from ggml-org at launch: Julia-1, Laya, Kev-4B, lev, OpenJev (image-capable, CC BY-NC 4.0), and Clef (image-capable, Apache 2.0), with the others also Apache 2.0 licensed.
Get the latest llama.cpp from llama.app or run llama update, then start a model such as `llama serve -hf ggml-org/Kev-4B-GGUF`. The server exposes /v1/systemone; in router mode /v1/models lists ids and you set model per request.
Some can. OpenJev at the time of writing reads images such as documents or screenshots, and its vision projector downloads automatically with `llama serve -hf ggml-org/OpenJev-GGUF`. You pass an images array alongside the state and questions.
A model id and per-question answers. A choice returns the top option, a probability per option and a confidence; noul returns the probability of yes; score returns the expected level, a legend mapping levels to labels and per-level probabilities. A usage block reports input_tokens and output_tokens.
Describe your options rather than sending bare labels. The post reports Julia-1 routed a duplicate-charge message to shipping with bare labels, then to billing at 0.99 once each option had a description. It also advises picking a confidence cutoff per model and sending unconfident answers to a human.
Median time to answer one question on one NVIDIA RTX PRO 6000 ranges from 3 ms for Julia-1 (144M) and 5 ms for Laya (421M) up to 12 ms for Kev-4B, 36 ms for lev, 43 ms for OpenJev (27B) and no published figure for Clef (27B). Small models are faster, large ones know more.
More Jev resources like this
4GitHub stars
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
lexingtonhibiki/judgekitPython2kGitHub stars
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
szczyglis-dev/py-gptPython1GitHub stars
A Rust filter that reads JSONL security alerts on stdin and asks five Jev questions per alert — disposition, severity, false positive, escalation, ATT&CK class — then applies the routing policy in ordinary code you control.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
m0rphtail/triagedyRust
m0rphtail/triagedyRust0GitHub stars
Claude Code and ZCode skill that offloads classify, route, batch-screen, score and compliance-check judgments to Jev through OpenCode Zen's free tier, bundling a zero-dependency jev.py caller and a Taobao comment-triage pipeline that keeps raw items out of the agent context.
Jev Tools & CLIs#community#claude-code#skills
Community
yuyang2230/jev-agent-skillPython
yuyang2230/jev-agent-skillPython110.5kGitHub stars
Multi-agent trading research with optional Jev screening of StockTwits and Reddit posts for company relevance before sentiment analysis runs.
Jev Tools & CLIs#community#python#trading
Community
TauricResearch/TradingAgentsPython
TauricResearch/TradingAgentsPython2.6kGitHub stars
Cisco's security scanner for agent skills added a System One analyzer: it sends the full skill text to POST /v1/systemone as one advisory Noul question and records the probability, while being deliberately unable to emit a finding or change a severity.
Jev Tools & CLIs#community#security#agent-skills
Community
cisco-ai-defense/skill-scannerPython
cisco-ai-defense/skill-scannerPythonFrom the Jev guides
Original write-ups on what this entry is for.
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
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
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
From the Jev community
Posts from builders shipping with Jev right now.