jev-usecases
Collection of production-shaped Python harnesses implementing most of TypeSafe's listed Jev decisions with confidence-gated logic.
Overview
jev-usecases is an initial implementation of most of the decisions TypeSafe lists for Jev, written as production-shaped harnesses with confidence-gated decision logic. The README opens with a clear explanation of Jev: a System One model trained with reinforcement learning for calibrated decisions, one systemone endpoint taking state and a map of questions, and the semantics of Choice (1 to 255 options with probabilities and confidence), Score (2 to 10 ordered levels) and Noul (a single probability, no separate confidence).
Included harnesses span customer support routing, model routing, LLM guardrails for jailbreaks, injection and PII, RAG passage relevance with injection filtering, citation checks, a multi-stage security operations pipeline (triage, investigation, mitigation approval, escalation, recovery and closeout, which approve or hold actions but never perform them), invoice processing, recruiting fit, lead scoring, insurance claims routing, AML prioritisation, legal clause and claim compliance, e-commerce and trust-and-safety moderation, ad brand safety, gaming toxicity and churn, knowledge-graph entity merging, semantic linting, coding-agent tool-call guardrails, function calling and hierarchical classification.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | SOC decision: close, queue or contain this incident? | close · queue · contain |
| Choice | Invoice: pay, hold, dispute or fraud review? | pay · hold · dispute · fraud review |
| Choice | Moderation: allow, warn, remove or ban? | allow · warn · remove · ban |
| Choice | Coding agent tool call: allow, ask or block? | allow · ask · block |
How it works
- Pick a use-case harness
- Build state and closed questions
- Jev returns probabilities
- Confidence gates choose automatic action, review or escalation
Strengths
- Broad coverage of decision patterns in one repo
- Agents approve or hold sensitive actions but never execute them
- Clear explanation of question semantics
Caveats
- Initial implementations; maturity varies by harness
- Accuracy per use case not reported in the README excerpt
At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- CLI harnesses
- Has code
- yes
- License
- MIT
- Stars
- 22
Use cases and integrations
How curated lists describe it
Reference harnesses: a set of production-shaped use-case harnesses built around confidence-gated decision logic.
yibie/awesome-jevProduction TypeSafe Jev (System One) use-case harnesses with confidence-gated decision logic
logicrw/awesome-jev-projectsProduction TypeSafe Jev (System One) use-case harnesses with confidence-gated decision logic
wh000wh000/awesome-jev-liveProduction TypeSafe Jev (System One) use-case harnesses with confidence-gated decision logic.
hellogumbo/awesome-jev
Frequently asked questions
How many use cases are included?
Dozens, covering support, routing, guardrails, RAG, security operations, finance, compliance, moderation and more.
Do the SOC agents take actions?
No. They approve or hold actions such as isolation or credential resets but do not perform them.
Does Noul return a confidence?
No, Noul returns a single probability; Choice and Score include confidence.
Related projects
TypeScript/Python policy engine where Jev only answers declared questions and code decides routes, gates and actions.
ProjectbetaTypeScriptHackathon agent monitor that scores chains of sandboxed agent actions with Jev and escalates through a fixed L1-L5 containment ladder.
AppprototypePythonReference Telltale analyzer extension that asks Jev typed questions from configurable guard packs about coding-agent actions.
PluginprototypeRustCline's official curated plugin collection, which includes jev-browser, jev-guard and jev-playground plugins.
PluginproductionTypeScriptWorkflow DSL for existing agents that uses Jev for the steps that are decisions and calls an agent only when investigation is needed.
LibrarybetaTypeScriptCLI, MCP server and Claude Code plugin for closed-form yes/no, choice, score and rank decisions via Jev, OpenAI Decisions or Claude.
MCP serverbetaTypeScript