jev
Rust CLI that sends typed questions to Jev, Cloudflare Clef, Ollama Clef, OpenAI Decisions or other System One providers.
Overview
jev is a command-line tool for System One decision models. It sends a piece of state plus typed questions and prints the typed answers as JSON, hiding differences between providers that share the same model/state/questions request shape. Supported providers include TypeSafe Jev (default, jev-latest), Cloudflare Workers AI serving Clef and Clef-flash, Clef on a local Ollama, OpenAI's Decisions API in public beta, OpenAI chat models read through their log-probabilities, and Celeris.
It is aimed at developers who want to use decision models from scripts or the shell for tasks otherwise handled by brittle parsing or hand-written rules. Credentials can come from the system keychain or environment variables, a provider and model can be set per call or by environment, and a custom URL can point at a gateway or mock. Provider-specific envelopes, such as Cloudflare's result wrapper, are removed so output looks the same everywhere, and unsupported model names are rejected before sending. It is written in Rust 2024 with CI.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Other | User-supplied Noul, Choice and Score questions over arbitrary state | — |
How it works
- Choose provider and model
- Pass state and typed questions on the command line
- CLI translates and sends the request
- Print normalised JSON answers for scripts to branch on
Strengths
- One interface across several decision-model providers
- Local option via Ollama
- Keychain credential storage
- Normalised output shape
Caveats
- OpenAI chat provider uses logprob approximation, not a native decision model
- Provider availability and pricing vary
At a glance
- Type
- CLI
- Status
- beta
- Language
- Rust
- Runtime
- CLI
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
CLI for System One decision models: typed questions in, calibrated…
wh000wh000/awesome-jev-live
Frequently asked questions
Which providers are supported?
TypeSafe Jev, Cloudflare Clef, Clef on Ollama, OpenAI Decisions, OpenAI chat via logprobs and Celeris.
Can I run it locally?
Yes, with Clef on a local Ollama and no credential.
Does output differ by provider?
No, provider envelopes are removed so every provider prints the same answers shape.
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