Summary

TuriX-CUA enables AI models to directly control desktop applications and actions via natural language. It targets users seeking automated workflows, offering a state-of-the-art, open-source agent that interacts with any UI element. The project provides a cost-free solution for personal and research use, allowing flexible integration of custom AI models.

How It Works

This project implements a computer-use agent driven by AI, enabling powerful models to interpret commands and execute real-world desktop actions. Its core design features "hot-swappable brains," allowing users to replace the underlying AI model (VLM policy) through config.json modifications without code changes. This approach facilitates easy integration of diverse AI models and enables interaction with any clickable UI element, eliminating the need for app-specific APIs.

Quick Start & Requirements

• Install: Clone repo, create Python 3.12 environment (conda create -n turix_env python=3.12), activate, run pip install -r requirements.txt.
• Prerequisites: Python 3.12, macOS 15 permissions (Accessibility, Safari Automation), API key (Turix or local Ollama). A windows branch is available.
• Links: Discord community, contact@turix.ai.

Highlighted Details

• Achieves state-of-the-art performance on macOS desktop automation tasks.
• Interacts with any UI element a human can click, supporting various applications without specific APIs.
• Features "hot-swappable brains" for easy VLM policy replacement via config.json.
• Supports Model Context Protocol (MCP) for third-party agent integration.
• Recent Qwen3-VL integration enhances task success rates by up to 15%.

Maintenance & Community

Active development is indicated by frequent updates (e.g., Dec 2025, Oct 2025) and a roadmap extending into Q2 2026. Community support is available via Discord.

Licensing & Compatibility

Described as "100% open-source and cost-free for personal & research use." Commercial use implications are not detailed, suggesting potential restrictions.

Limitations & Caveats

Primary quick-start is macOS 15-specific; a windows branch exists. Requires significant system permissions (Accessibility, Safari Automation) and relies on external API keys or local LLM setups, which may incur costs or require specific hardware.