jev-starter
TypeScript library of decision contracts, confidence thresholds, fallbacks, human review and evaluation patterns on top of the TypeSafe SDK.
Overview
jev-starter is not another SDK wrapper; it sits on the official @typesafe-ai/sdk and provides the layer applications need around a Jev call. It turns application state into a Jev decision, takes the typed probabilities, and passes them through a decision policy that routes to automatic action, fallback or human review.
It targets TypeScript developers moving Jev into production. Stated principles include separating inference from policy, keeping side effects in the host application (the starter returns a recommendation, never executes business operations), treating uncertainty as a first-class path, supporting two-sided thresholds so confident negatives are distinguished from ambiguity, and choosing thresholds from task-specific evaluation data rather than copying examples. It also includes retrieval-augmented and evaluation examples. Version 0.1.1 is published on npm with provenance, has CI, and the repository is usable as a GitHub template.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Other | Application-defined Noul, Choice or Score questions passed through a confidence policy(inferred) | — |
How it works
- Build state from application data
- Ask Jev via the official SDK and receive typed probabilities
- Apply policy thresholds (two-sided for binary)
- Return auto, fallback or human-review recommendation to the host
Strengths
- Clear separation of model estimation and business policy
- Explicit uncertainty and review routes
- Evaluation tooling for choosing thresholds
- Published on npm with provenance and CI
Caveats
- Early 0.1.x release
- Thresholds must be tuned on your own data
Getting started
pnpm add jev-starter (Node.js 20+).At a glance
- Type
- Library
- Status
- beta
- Language
- TypeScript
- Runtime
- Node
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
TypeScript patterns for thresholds, fallbacks, human review and evaluation on top of the TypeSafe SDK.
logicrw/awesome-jev-projectsTyped, policy-driven decision workflows on top of TypeSafe AI Jev: confidence routing, fallbacks, evaluation, and RAG patterns for TypeScript apps.
hellogumbo/awesome-jev
Frequently asked questions
Is this a Jev SDK?
No. It uses the official @typesafe-ai/sdk and adds policies, fallbacks, review routing and evaluation.
Does it execute actions?
No. It returns an action recommendation; side effects stay in your application.
How should I set thresholds?
From task-specific evaluation data, which the library helps you measure, not by copying example values.
Related projects
Rust CLI and library that asks Jev typed questions via OpenRouter and combines the probabilities with fuzzy-logic rules into decisions.
CLIbetaRustUnofficial standard-library Go client for Jev with generic typed Choice values and local confidence thresholds.
ProjectbetaGoRust CLI for Jev: ask yes/no, choice and rubric questions about text and branch on exit codes, JSON or MCP tools.
MCP serverbetaRustpip-installable Jev kit: a CLI for verify, screen, classify, extract, route and rerank, plus an MCP server for the primitives.
MCP serverbetaPythonTypeScript library for typed semantic decisions via Zod schemas or native batches, defaulting to Jev, with confidence-gated policies.
LibrarybetaTypeScriptRubyLLM provider gem for TypeSafe Jev: block-style typed evaluations with noul, choice and score, plus RubyLLM.rerank support.
ProjectbetaRuby