Ulexite

Ulexite is a specialized programming language designed for seamless interactions with conversational AI. It prioritizes the structure of conversations over the complexities found in typical prompt-based systems. By allowing developers to conceptualize conversations as flows, Ulexite ensures deterministic execution, reproducible traces, and built-in testing capabilities.

Key Features

1. Conversation-Centric Approach: Focuses on conversations rather than prompts, making it easier to craft and verify interactions as if they were traditional software programs.
2. Deterministic Execution: Ulexite offers predictable outcomes, ensuring that conversations can be reliably conducted and subsequently reviewed.
3. Extensive Testing and Reproducibility: The language facilitates rigorous testing, allowing developers to trace back through interactions to evaluate outputs.

Language Structure

A typical Ulexite program (.ulx file) outlines a conversation's path, detailing inputs from users, responses from models, and intermediary judgments as shown in the following flow chart:

flowchart LR
    U[You ask] --> M[Model answers]
    M --> J{Judge grades it}
    J -->|Pass| R[Result]
    J -->|Fail| Retry[Retry with feedback] --> M
    J -->|Escalate| H[Human approval] --> R

For example, the following Greet conversation showcases a simple interaction:

conversation Greet(name: text) -> text {
  user: """Say hello to {name}."""
  assistant -> greeting: text
  greeting
}

Interactive Playground

Experience Ulexite in action without the need for installation through its Interactive Playground, where users can compile and execute .ulx programs in-browser against a lightweight local model.

Markdown Integration

Users can directly execute plain Markdown as Ulexite scripts, further simplifying the writing process:

# Greet
Say hello to {name} and ask how their day is going.

This can be executed with the command:

ulx run greet.md Greet --arg name=Ana

Ulexite is positioned as a versatile tool for developers looking to integrate AI conversations more naturally and effectively into their applications, while significantly enhancing testing and validation workflows.