Mechanical-Jev
Rust client, CLI, terminal UI and eval harness for asking System One questions of a local model served on Xeon Phi cards.
Overview
Mechanical Jev is the asking side of a System One setup in Rust: a client library, the mjev command line, a terminal interface and an evaluation harness. You send a state and typed questions (noul, choice with up to 255 options, or score with 2 to 10 levels) and get typed answers with probabilities. The server is Intel Phi Jev, a sibling project that runs a local model on the host and its Xeon Phi accelerator cards and answers the System One wire format; this repository holds no model and runs no inference.
make setup runs a doctor that checks every step a question needs, from the CLI to the llama.cpp build, payload, cards and server, builds and links what it can, and names anything requiring a card boot, model download or sudo rather than doing it. The first question starts the server, which stops itself after 30 minutes idle and returns memory and cards. Extra targets run long real-session evaluations, compare local answers with Jev's published questions, fetch pinned tokenizers and reproduce reverse-engineering fits offline. It is a hobbyist and research setup for unusual hardware.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this command read-only? (example) | — |
| Choice | Which option fits (up to 255)? | — |
| Score | Where on a 2-10 level rubric does this fall? | — |
How it works
- Clone the Intel Phi Jev server alongside
- make setup checks and builds what it can
- mjev query or the TUI sends a question
- Server starts on demand and answers locally
- Server stops itself after 30 idle minutes
Strengths
- Fully local inference on owned hardware
- Thorough setup doctor
- Closeness checks against Jev's published answers
Caveats
- Requires Xeon Phi cards and a sibling server repo
- Niche hardware setup
- Questions take seconds after startup
Getting started
git clone https://github.com/Lasimeri/Intel-Phi-Jev ../Intel-Phi-Jev && make setup, then run mjev.At a glance
- Type
- Library
- Status
- research
- Language
- Rust
- Runtime
- CLI and TUI against a local System One server
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
The asking side of Jev (TypeSafe System One: noul, choice, score) in Rust…
wh000wh000/awesome-jev-live
Frequently asked questions
Does this run a model?
No, it is the client side; Intel Phi Jev serves the model.
What hardware is needed?
The sibling server runs on a host with Xeon Phi cards.
How is it compared to Jev?
A closeness target asks Jev's published questions locally and compares answers.
Related projects
Go harness for System One models: hosted Jev, open Kev in-process via llama.cpp, and Cloudflare Clef, with typed helpers and a CLI.
ServicebetaGoTypeScript library for one-line local decisions with open models (kev-0.6b, kev-4b, open-jev), no API keys or cloud.
ProjectbetaTypeScriptSwift 6 bridge that runs System One decision models behind Apple's Foundation Models API, on device with Laya or via hosted endpoints.
ProjectbetaSwiftPHP library, extracted from LLPhant, for typed Jev Noul, Choice and Score calls, also usable with a Laya endpoint.
LibrarybetaPHPUnofficial Rust gateway routing /v1/systemone requests to TypeSafe, Vercel, OpenRouter or local Laya via an editable DAG.
ProxybetaRustZero-dependency local gateway and live visualiser that proxies Jev calls, logs every decision to SQLite and draws them on a real-time map.
ProxybetaJavaScript