# JEV API — Developer Guide

> An independent developer guide to the Jev API, covering typed decisions, agent routing, and evaluation with tutorials and examples; API access is provided via BeatAPI.

jevapi.io is an independent developer resource maintained by the BeatAPI team, focused on helping developers understand and build with the Jev decision model API. It is not the official Jev website and is not affiliated with TypeSafe AI, which develops Jev. The site provides tutorials, request/response examples, use-case patterns, and production safeguards guidance, with actual API access, accounts, and billing handled on BeatAPI.

## What It Is

Jev is a decision model for software, developed by TypeSafe AI as its first "System One" model. Unlike general-purpose LLMs that generate open-ended text, Jev accepts application state and a set of developer-defined questions, then returns typed answers with probability distributions. The three core decision primitives are:

- **Noul** — a boolean probability (0 to 1) for whether a statement is true
- **Choice** — selects one option from a defined set and returns the full probability distribution
- **Score** — evaluates an ordered rubric and returns a numeric score with uncertainty across levels

Requests are sent to `POST /v1/systemone` with model `jev-1.13` via BeatAPI. The response returns a typed answer and probability for every question in the request.

## How It Fits in a Software Stack

The site positions Jev as a complement to LLMs, not a replacement. The intended pattern is: deterministic code handles permissions, thresholds, retries, logs, and side effects; Jev handles the bounded judgment in the middle. The comparison table on the site contrasts Jev (typed value output, defined outcomes, probabilities) against LLMs (generated text, open-ended outcomes, inferred uncertainty).

## Use Cases and Patterns

The site documents 18 patterns across a field guide, with six having dedicated pages:

- **Tool and model router** — choose the next tool, subagent, or model tier from a fixed roster
- **Confidence-gated workflow** — act automatically above a threshold, review the middle band, stop below it
- **Agent completion verifier** — check a claimed "done" against acceptance criteria before stopping
- **LLM-as-a-judge alternative** — grade outputs against a fixed rubric without a written critique
- **RAG evidence filter** — screen retrieved passages for relevance and injection risk before generation
- **Semantic policy check** — enforce a natural-language rule that ordinary code cannot parse

Additional patterns cover support triage, risk scoring, output verification, and guardrails for high-volume decisions where possible outcomes are known in advance.

## JEV Signals: Conversation Analyzer

The site also documents JEV Signals, a working tool hosted on BeatAPI that uses Jev to analyze conversations and estimate four signals: intent, momentum, reciprocity, and next-step readiness. A separate DeepSeek call then drafts replies that fit those probability estimates. The site notes that jevapi.io itself does not run inference, hold accounts, or store balances — all of that lives on BeatAPI.

## Editorial Approach and Accuracy Standards

The About page describes a structured evidence-labeling system: every claim carries a label of "documented," "reported," "illustrative," or "live verified." Example payloads are explicitly marked illustrative with invented numbers. Performance and cost claims from vendors are attributed and treated as reported until independently reproduced. The site is updated when primary sources change, and corrections replace old claims rather than appending them.

## Features
- API tutorials with runnable examples
- Three decision primitives: Noul, Choice, Score
- Agent routing patterns
- Confidence-gated workflow guidance
- RAG evidence filter pattern
- LLM-as-a-judge alternative pattern
- Agent completion verifier pattern
- Semantic policy check pattern
- 18-pattern field guide
- JEV Signals conversation analyzer
- Multilingual site (EN, 中文, 日本語, 한국어, Português)
- Evidence-labeled claims (documented, reported, illustrative, live verified)
- Jev vs LLMs comparison
- Production safeguards and threshold guidance

## Integrations
BeatAPI, DeepSeek, TypeSafe AI Jev model

## Platforms
WEB, API

## Pricing
Free

## Version
jev-1.13

## Links
- Website: https://jevapi.io
- Documentation: https://jevapi.io/jev-api/
- EveryDev.ai: https://www.everydev.ai/tools/jev-api-developer-guide