FastJev
Self-hosted open Jev implementation: SDK and System One API for choice, boolean and score on pinned open models across many runtimes.
Overview
FastJev is an open-source implementation of Jev for deployment on infrastructure you control, continuing the SemIf project as an independently maintained fork. Pinned open models run the Choice, Boolean and Score interface through Torch, vLLM, MLX, llama.cpp and optional EXL3 quantisation, and a WebGPU demo provides in-browser inference.
The Python SDK loads the model, validates 2 to 16 options, performs scoring and returns typed results with probabilities, token usage, timing, model revision and prompt version; the CLI and an optional System One-compatible HTTP API share the same result model, and most backends score in one pass with zero output tokens. Validated pinned checkpoints include Qwen3-0.6B, MiniCPM5-2B and Qwen3.5-4B, the recommended model, with the README reporting authored balanced accuracy of 0.440, 0.686 and 0.813 respectively. Published evaluations include row data, raw timings and checksums.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Runtime-defined Choice over 2 to 16 options | — |
| Yes / no | Runtime-defined Boolean question | true · false |
| Score | Runtime-defined Score question | — |
How it works
- pip install fastjev with a backend extra
- Load a pinned model revision
- Ask typed questions via SDK, CLI or HTTP API
- Read results with probabilities and provenance
Strengths
- Many runtimes including browser WebGPU
- Reproducible evaluations with checksums
- Zero output tokens on most backends
Caveats
- Open model accuracy below hosted Jev on hard tasks
- Options limited to 2 to 16
Getting started
pip install 'fastjev[torch]' and follow the quick start with a listed model ID and revision.At a glance
- Type
- Library
- Status
- beta
- Language
- Python
- Runtime
- Torch, vLLM, MLX, llama.cpp, WebGPU; HTTP API
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Local runtime: self-hosted Python SDK and System One-compatible API for runtime-defined `Choice`, `Boolean`, and `Score` decisions on pinned open models across Torch, vLLM, MLX, llama.cpp, and WebGPU, with committed row-level benchmarks and checksums.
yibie/awesome-jevAn independently maintained SemIf fork packaged SDK-first, for deploying an open decision model on your own infrastructure.
MrJev/awesome-jev
Frequently asked questions
Which model is recommended?
Qwen3.5-4B, with the highest measured balanced accuracy in the README.
Can it run in a browser?
Yes, via a WebGPU demo.
Is it compatible with Jev's API?
An optional System One-compatible HTTP API is included.
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.
ProjectbetaSwiftZero-dependency npm package that runs Laya natively or as WebAssembly in Node, Bun and browsers, drop-in for Jev's /v1/systemone.
ProjectbetaJavaScriptLocal-first desktop AI workbench for llama.cpp, vLLM and SGLang that includes a JEV typed-decision page and /v1/systemone gateway.
ServicebetaRuns the open LAYA decision model under Bun with native ONNX inference behind a TypeSafe-compatible API, plus an English/Persian playground.
LibrarybetaTypeScript