Mairex

An experimental orchestration language for AI models and system commands

Mairex is an orchestration language that allows you to coordinate AI models, shell commands, and data flows using a JSON-based syntax with specialized operators. It's designed for developers who want to prototype AI workflows and automation scripts.

⚠️ Alpha Status: Mairex is in early development (v0.1.0). Expect bugs, missing features, and potential breaking changes in future versions.

Current Limitations

• Sequential execution only (parallel execution syntax exists but runs sequentially)
• Basic error handling
• Limited debugging capabilities
• Many planned features not yet implemented
• Documentation may be ahead of implementation in some areas

What It Does

Mairex lets you write scripts that combine shell commands and AI model calls in a declarative way. For example, you can download websites, process them with AI models, and save outputs to files - all coordinated through a single .jsom file.

✨ Features

• 🤖 Native AI Integration - Call Ollama, OpenAI, Anthropic, Gemini, and XAI models directly
• 🔄 Parallel Execution - Run multiple shells and AI models concurrently
• 🔗 Chainable Operations - Flow data between commands, files, and AI models
• 📦 Variable Scoping - Custom and AI-specific variable management
• 🎯 JSON-Based Syntax - Familiar structure with powerful extensions
• 🛠️ Shell Integration - Execute any terminal command with persistent shell sessions

🚀 Quick Start

Installation

pip install mairex

Your First Mairex Script

Create a file hello.jsom:

{
  "greeting": {
    "set_input": [
      "~| A&I <&¤S- 'World' |~"
    ],
    "set_prompt": [
      "~| A&P <&¤S- 'Say hello to the input' |~"
    ],
    "call_ai": [
      "~| A&O -$S> |>echo '<$>'<| |~"
    ]
  }
}

Run it:

mairex hello.jsom

What this does:

1. Sets AI input to "World"
2. Sets AI prompt to "Say hello to the input"
3. Calls the AI model and echoes the response

📚 Core Concepts

JSOM Files

Mairex scripts use .jsom files (JSON + Mairex). They follow standard JSON syntax with one rule:

All leaf nodes must be arrays:

{
  "step": {
    "action": ["value"]
  }
}

NOT:

{
  "step": {
    "action": "value"
  }
}

Instructions

Instructions are declared between ~| |~ specifiers:

["~| |>echo 'Hello'<| |~"]

Shell commands go between |> <|:

["~| |>ls -la<| |~"]

Variables

Custom Variables (shared across shells, scoped to function):

["~| VAR&V <&¤S- 'my value' |~"]

AI Variables (shell-specific, persistent across tree levels):

["~| A&I <&¤S- 'AI input' |~"]

Data Flow

Left to right:

["~| |>echo 'output'<| -&#> FILE&V -€S> result.txt |~"]

Right to left:

["~| FILE&V <&€- result.txt <&#- |>cat file.txt<| |~"]

Parallel Execution

Separate shells (parallel):

{
  "parallel_tasks": [
    "~| |>echo 'Shell 1'<| |~",
    "~| |>echo 'Shell 2'<| |~",
    "~| |>echo 'Shell 3'<| |~"
  ]
}

Each array element runs in its own independent shell session.

🎓 Learn More

• Syntax Reference - Complete language specification
• Tutorial - Step-by-step guide
• Examples - Real-world use cases

🔧 Requirements

• Python 3.8+
• Dependencies (auto-installed):
  • ollama - Local AI model support
  • litellm - Multi-provider AI API support
  • lizard - Code analysis for function extraction
  • whats_that_code - Programming language detection

🌐 AI Provider Setup

Using Ollama (Local Models)

1. Install Ollama: https://ollama.ai
2. Pull a model: ollama pull llama3
3. No API keys needed - works out of the box!

Using Cloud AI Providers

Create API_keys.json in your working directory:

{
  "openai": "sk-your-key-here",
  "anthropic": "sk-ant-your-key-here",
  "gemini": "your-gemini-key",
  "xai": "your-xai-key"
}

Set the provider in your JSOM file:

["~| A&S <&¤S- 'openai' |~"]
["~| A&M <&¤S- 'gpt-4o' |~"]

📄 License

MIT License

Development Status

This is an early alpha release. The project is not currently accepting outside contributions. Bug reports and feedback are welcome via GitHub issues.

---

An experimental tool for AI orchestration