TuriX-CUA is a Python-based desktop automation agent that integrates AI models for controlling GUI workflows, with multi-branch Linux/Windows/macOS support and OpenClaw Skill integration. It provides config-based brain/actor/memory setup and optional skills, with latest releases updating search, memory, and multi-agent features.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
TuriX-CUA is a desktop actions platform driven by AI, designed to let powerful AI models perform actions on the user's computer via GUI automation. It ships with a state‑of‑the‑art computer‑use agent and supports switching in different model brains via a configurable setup. It includes OpenClaw skill packages for macOS and Windows, and provides optional markdown-playbook skills.
How it works
The project uses a multi-model architecture where brain_llm, actor_llm, memory_llm, and planner_llm can be configured to use either built-in Turix models or external providers (e.g., Ollama). Skills are stored as Markdown files under a skills directory; the planner selects relevant skills by name/description and the brain uses the full skill content to guide steps. The README shows guidance for editing task configuration and API configuration in JSON, and how to enable/disable planning and skills.
Getting started
- Quick-start guidance is provided for macOS, including:
- Create a Python 3.12 environment and install dependencies
- Grant macOS permissions (Accessibility and Safari Automation)
- Edit and configure task and API settings in examples/config.json
- Start the agent with
python examples/main.py
- Sample commands and configuration blocks are included for brain/actor/memory/planner setups and optional Ollama integration.
Recent releases
Latest releases (latest 4):
- v0.4 (2026-03-06): release v0.4, support brain-actor mode multi-agent
- v0.3 Release V0.3 (2026-01-27): DuckDuckGo search, Ollama support, advanced recoverable memory compression, and Skills
- v0.2 V0.2 (2025-12-01): more advanced output structure and memory control
- 0.1.0 First beta release version (2025-08-04): first beta of TuriX-CUA model
Traction
Stars: 3144. Forks: 333. Open issues: 4.
Behind the repo
This section not present in the provided facts, so omitted.
Caveats
- License: MIT
- Created: 2025-08-04
- Last push: 2026-07-30
- Language: Python
- Branch guidance for Linux/Windows/macOS is provided in the Quick-Start notes
- OpenClaw skill packages exist for macOS (main) and Windows (multi-agent-windows) with specific run scripts and YAML files in those branches