# Jev AI

> Jev AI is an independent developer guide, primitives explorer, and community project directory for TypeSafe's Jev System One decision model.

Jev is built for structured judgments in software automation: input application state and focused questions, receive structured choices, scores, and probabilities with sub-500ms latency and calibrated confidence.

## Core Concepts & Primitives

- [Choice Primitive](https://jevai.me/ai): Single candidate selection from a defined set with calibrated probability distribution.
- [Score Primitive](https://jevai.me/ai): Numerical evaluation against a described rubric or scale.
- [Noul Primitive](https://jevai.me/ai): Probability judgment evaluating whether a specific truth statement or condition holds.
- [Model Architecture & Guide](https://jevai.me/blog/hello-world): Technical breakdown of System One decision models, RLCD training, and workflow composition.
- [Getting Started Guide](https://jevai.me/blog/getting-started): Integration guide for routing, retrieval ranking, and agent guardrails.

## Directory & Tools

- [Community Projects](https://jevai.me/projects): Catalog of browser agents, routing implementations, context compaction tools, and MCP servers.
- [Interactive Decision Demos](https://jevai.me/ai): Local educational examples demonstrating typed outputs for support routing and verification.
- [Changelog](https://jevai.me/changelog): Version releases and update history.

## Technical Specifications

- Latency profile: 70ms to 500ms published end-to-end response time.
- Input format: Text state and focused questions.
- Training methodology: Reinforcement Learning for Calibrated Decisions (RLCD).
- Output guarantees: Valid typed JSON output matching specified primitives.

## Full Documentation

- [Full Content Ingestion](https://jevai.me/llms-full.txt): Comprehensive documentation and guide texts for LLMs with extended context windows.