Jev-Style runs on your own machine — the client starts it, so there is no endpoint to ping. 518 installs a week from pypi. Last commit 3 Oct 2026.
Local 2B and 0.8B decision models: calibrated yes/no, choice and score answers for your agent.
Today is the operative word: we check Jev-Style every 15 minutes and re-read its code on every release. Watch it and you find out the day that stops being true.
This server runs on your own machine — install it with the package manager and the client starts it for you. Package name taken from the official registry entry.
claude mcp add jev-style -- uvx jev-style
{
"mcpServers": {
"jev-style": {
"args": [
"jev-style"
],
"command": "uvx"
}
}
}
[mcp_servers.jev-style]
command = "uvx"
args = ["jev-style"]
{
"mcpServers": {
"jev-style": {
"args": [
"jev-style"
],
"command": "uvx"
}
}
}
{
"mcpServers": {
"jev-style": {
"args": [
"jev-style"
],
"command": "uvx"
}
}
}
This one needs environment variables set before it will start:
JEV_STYLE_URL (URL of a running `jev-style serve` (start it with `pip install "jev-style[torch]"` or `[mlx]` on Apple silicon, then `jev-style serve`).).
The author declared them in the registry entry; get the values from the project itself.
Calibrated decisions for agents: choice, yes/no and score questions answered with probabilities.
Typed decisions (choice, score, yes/no) from Laya, Julia-1 and clef-flash on your own GPU
Typed answers with calibrated confidence for agents. Local-first, no API key.
Typed Jev decisions in one batched tool: choice, score, and yes/no probabilities.
Ask TypeSafe Jev yes/no, choice and score questions and receive structured model judgments.
Typed noul / choice / score judgments from a local Laya model, with token-budget preflight.
Typed AI judgments for agents: choice, score, and yes/no probabilities instead of prose.
Local Issue triage with optional Jev judgments and calibrated acceptance gates.
Answers built from our own checks of this server.