TuriX Computer-Use Agent (CUA)

TuriX is an AI-driven desktop automation agent with the following core capabilities:

• SOTA Default Models: High success rates and speed on macOS, outperforming previous open-source agents like UI-TARS.
• No App-Specific APIs Required: Operates via GUI interaction (clicks/typing), meaning it can use any application a human can (WhatsApp, Excel, Outlook, etc.).
• Hot-swappable 'Brain': Easily replace VLM strategies via config.json.
• MCP Ready: Can be integrated with Claude for Desktop or any agent supporting the Model Context Protocol (MCP).
• Skills (Markdown Manuals): The Planner selects skills based on names/descriptions, while the Brain uses the full skill content to guide execution steps.

To set up TuriX on macOS 15+, follow these steps:

1. Download the App: Obtain the application from the official website.
2. Create a Python 3.12 Environment: Ensure you have a dedicated Python 3.12 environment prepared.
3. Grant macOS Permissions: You must grant specific permissions for the agent to operate:
   • Accessibility: Required for controlling the UI.
   • Safari Automation: Required if the agent needs to interact with Safari.
4. Configure and Run: Set up your config.json and execute the application.
5. Skills (Optional): You can optionally add skills to enhance capabilities.

To run TuriX on Linux (e.g., Ubuntu), you must switch to the multi-agent-linux branch.

git checkout multi-agent-linux

To use the faster, lighter setup for macOS (OpenClaw Flash/Fast Mode skill), switch to the mac_legacy branch.

git checkout mac_legacy

Once your environment is set up and examples/config.json is configured, start the agent using the following command:

python examples/main.py

You can use TuriX via OpenClaw using the published ClawHub skill. The repository also contains local OpenClaw skill packages for different operating systems:

• macOS: Located in the main branch. Includes SKILL.md and scripts/run_turix.sh.
• Windows: Located in the multi-agent-windows branch. Includes SKILL.md, scripts/run_turix.ps1, and agents/openai.yaml.

For specific installation and permission instructions for OpenClaw, refer to OpenCLaw_TuriX_skill/README.md within the repository.

If a task is interrupted, you can resume from the last position by configuring a fixed agent_id and enabling resume in examples/config.json.

Requirements for Resuming:

• Use the same agent_id used in the previous run.
• Keep the same task string.
• A memory file must exist at src/agent/temp_files/<agent_id>/memory.jsonl.

To start fresh, set resume to false, change the agent_id, or delete the directory src/agent/temp_files/<agent_id>.

{
    "agent": {
         "resume": true,
         "agent_id": "my-task-001"
    }
}

To immediately stop the agent, use the following methods:

• Hotkey: Cmd+Shift+2
• Command Line: Use pkill to terminate the python process.

pkill -f "python examples/main.py"

Once configuration is complete, start the agent by running the main entry point from the root directory.

python examples/main.py

Skills are Markdown files located in a directory (default skills/) that provide instructions for specific tasks.

1. Format: Each file must start with YAML frontmatter containing name and description.
2. Usage: The Planner uses the name/description to select the skill, while the Brain uses the full content for step-by-step guidance.
3. Requirement: You must enable planning (agent.use_plan: true) to use skills.

Example Skill File (skills/github-web-actions.md):

---
name: github-web-actions
description: 用于在浏览器中操作 GitHub（搜索仓库、点 Star 等）。
---
# GitHub Web Actions
- 打开 GitHub，使用站内搜索并进入仓库页面。
- 若需要登录，先向用户确认再继续。
- 在继续之前确认 Star 按钮状态。

Enable in examples/config.json:

{
  "agent": {
    "use_plan": true,
    "use_skills": true,
    "skills_dir": "skills",
    "skills_max_chars": 4000
  }
}

To enable Clawdbot to control your macOS desktop visually, you must set up the TuriX Core and configure the skill directory.

1. TuriX Core Setup: Install the core requirements from the official TuriX-CUA repository:

conda activate your_env
pip install -r requirements.txt

2. Skill Directory Structure: Place the skill files in your local Clawdbot skills directory following this structure:

your_dir/clawd/skills/local/turix-mac/
├── README.md
├── SKILL.md
└── scripts/
    └── run_turix.sh