Jevs-Garage
Gallery of small Python demos where Jev answers typed questions on realistic state and explicit policy decides the action or fallback.
Overview
Jev's Garage is a workshop of small, inspectable experiments for TypeSafe's System One models. Each 'bay' gives Jev a realistic state, asks Choice, Score and Noul questions, and lets ordinary Python policy decide what happens next with explicit thresholds: act when confident, fall back to human review when uncertain. Critical demos never let the model directly perform an irreversible side effect.
Critical bays include fraud screening (route suspicious card activity to authorisation, verification or reversible holds), cyber triage (investigation or containment recommendations), safety guardrails for assistant output, conservative medical routing without diagnosis, and industrial cases, with further non-critical bays. A local gallery app has input, signals, visuals and decision tabs, validates edited input before a live call, and streams progress. It is for developers learning the state, Jev, typed result, policy and action pattern across domains. CI includes a live-test workflow.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Should this card activity be authorised, verified or held for review? | authorise · verify · review hold |
| Choice | Should this endpoint evidence lead to investigation or containment? | investigation · containment |
| Choice | Should this drafted assistant output be released, transformed or suppressed? | release · transform · suppress |
How it works
- Choose a bay and edit its JSON input
- Input is validated locally and on the server
- Jev answers the bay's typed questions
- Deterministic policy applies thresholds
- Confident results produce a bounded action; uncertain ones fall back to human review
Strengths
- Many domains shown with the same safe pattern
- Model never directly performs irreversible actions
- Inspectable gallery UI with signal visualisation
Caveats
- Demos, not production systems
- Requires a TypeSafe key for live calls
- Medical and security bays are illustrative only
Getting started
uv sync --locked --all-groups, cp .env.example .env (add TYPESAFE_API_KEY), then just gallery.At a glance
- Type
- App
- Status
- prototype
- Language
- Python
- Runtime
- local web gallery
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
System One turns unstructured or structured state into fast probabilistic judgments. Instead of asking for free-form prose, these demos ask \`Choice\`, \`Score\`, and \`Noul\` questions and receive typed values with uncertainty that application code can reason about.
logicrw/awesome-jev-projects
Frequently asked questions
What is a bay?
A self-contained demo with a realistic state, typed Jev questions and a deterministic policy.
Can the model take actions directly?
No. Policy decides, and critical demos never let the model trigger irreversible effects.
What do I need?
Python 3.11+, uv and a TypeSafe API key.
Related projects
Companion code for a hands-on book on building decision engines with hosted Jev and open-source Kev, with two capstone projects.
ArticleproductionPythonZero-dependency HTTP traffic and bot defense gate that can escalate requests to Jev for five typed risk questions.
ProjectbetaTypeScript, PythonEvent-stream layer labelling signups, logins and checkouts as human, AI agent, crawler or bot, with a bad-intent probability.
AppprototypeJavaScriptAwesome list and guide to System One models (Jev, Laya and others) organized around a five-stage agent lifecycle, with comparison tables.
ProjectproductionOpinionated catalog of 84 Jev projects in 11 categories, each with a hand-written reason it is worth looking at.
ProjectproductionUnofficial collection of Jev state formats, question patterns, cookbook index and case studies, plus a ranked Top 30 of community builds.
Projectproduction