OpenJev
OpenJev 1.7B-8B: open decision models distilled for general decisions and browser-agent steps, served behind /v1/systemone.
Overview
OpenJev (package wev-ai) is an open, local alternative to Jev offering System One decision models for agents in 1.7B, 4B and 8B sizes. It answers the POST /v1/systemone request shape: free-form state plus any number of choice, yes/no or score questions, with no text generation and no API key. It covers general decisions like triage, routing, policy checks and agent monitoring, and browser-agent steps: which operation and which element.
The README reports that OpenJev-4B v0.2 gets browser steps on unseen websites right 78.8% of the time, scores 88.1 on Kev decision-v7 and 81.4 out of domain, and completes 39-41 of 153 held-out live web tasks as the System One of an open browser agent, against 27 for its qwen3-max teacher. Latency is 10-40 ms for general decisions and 91-322 ms per browser page on an RTX 5090, with ECE 0.01-0.04. Training code, data and models are Apache-2.0.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which operation should the browser agent perform? | — |
| Choice | Which page element should it act on? | — |
| Yes / no | Is the task done?(inferred) | yes · no |
How it works
- Run OpenJev locally (GPU or laptop).
- Send state and typed questions to /v1/systemone.
- Use calibrated probabilities to pick operations and elements.
- Raise the threshold for DONE to reduce early stops.
Strengths
- Strong browser-step accuracy on unseen sites per the README.
- Calibrated probabilities (ECE 0.01-0.04).
- Open weights, data and training code.
Caveats
- 8B and 1.7B are still v0.1.
- Results are author-reported.
Getting started
pip install wev-ai and download OpenJev-4B from Hugging Face; see the Quickstart in the README.At a glance
- Type
- Model
- Status
- beta
- Language
- Python
- Runtime
- PyTorch, HTTP server
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
OpenJev: open-source, local alternative to Jev.
wh000wh000/awesome-jev-liveLocal, open-weights decision model for browser agents: a drop-in for System One / Jev requests.
hellogumbo/awesome-jevOpen decision models (1.7B to 8B) trained by distillation for general decisions and browser-agent steps, served behind the `POST /v1/systemone` shape, with training code and results. Apache-2.0.
MrJev/awesome-jev
Frequently asked questions
What sizes are available?
OpenJev-1.7B, 4B (v0.2) and 8B.
How does it do on browsing?
The README reports 78.8% correct browser steps on unseen websites for OpenJev-4B.
Does it need an API key?
No, it runs locally.
Related projects
Local browser agent that picks each step's operation and element with a Jev-shaped model such as a fine-tuned Laya, no cloud needed.
MCP serverbetaPythonOpen 2B browser-agent decision model fine-tuned from MiniCPM5 that serves the System One protocol for jev-ultrafast.
ProjectresearchPythonRuns jev-browser entirely on a local Jev-style engine (Qwen2.5 1.5B decisions, 0.5B typing) on an RTX 2060 SUPER, with no cloud API.
BenchmarkprototypePython, TypeScriptMangaba AI's local, calibrated typed-decision model for Portuguese text, trained to optimize the Brier score; proprietary since v5.
ProjectproductionPython- blink★ 22
Open one-pass typed-decision model with a dependency-free C99 runtime that also builds to a 66 KB WebAssembly module.
ServicebetaC - Foq★ 11
Local open alternative to Jev using a 1.58-bit ternary 8B model to answer booleans, choices, scores and Pydantic extraction in ~25 ms.
ProjectbetaPython