Practical overview
When it fits
Use it when a task requires an agent to act through desktop interfaces rather than depend on an app-specific API, and when you can configure the model roles and grant the required system permissions.
Who it is for
- People who want an AI agent to carry out computer tasks across desktop apps.
- Developers configuring model roles or connecting a desktop agent to another agent through MCP.
- Researchers exploring computer-use agents and desktop automation.
Documented uses
- Automate a task that involves interacting with visible desktop controls.
- Search for information, create a document, and share it through a desktop app.
- Transfer information between desktop applications, such as creating a chart from spreadsheet data for a presentation.
- Connect desktop actions to an MCP-capable agent or use Markdown skill playbooks to guide task planning.
Requirements & limits
The repository is public and MIT-licensed. Its README describes the project as cost-free for personal and research use.
- The main multi-agent branch is described as Mac-only; its quick-start specifies macOS 15 or later.
- The setup calls for Python 3.12 and macOS Accessibility and Safari Automation permissions.
- The main configuration distinguishes brain, actor, and memory models, with a planner model also needed when planning is enabled.
- The README documents local Ollama configuration and provider customization; custom providers may need code added in main.py.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Configure a desktop task and model roles
Prepare the main-branch configuration for a specified desktop task and selected model provider.
Before you start
- A Python 3.12 environment
- macOS 15 or later for the main-branch quick-start
- The TuriX repository and its examples/config.json
- Model-provider credentials and endpoint details, or a configured local Ollama server
Steps
Create the Python 3.12 environment described in the quick-start.
Set the task in examples/config.json using a clear, specific instruction.
Configure the brain, actor, and memory model roles in that file; configure a planner model too if planning is enabled.
For local Ollama, set the model roles to the Ollama provider and point them to the Ollama server.
Expected outcome
A task and model-role configuration prepared for the next documented stage, Configure & Run.
Guide limitations
- This recipe covers configuration, not execution.
- The README says the main multi-agent branch supports Mac only.
- Custom providers may require defining the provider in main.py before configuring it.
Prepare macOS permissions for desktop control
Grant the permissions identified by the macOS quick-start before using TuriX to control the desktop.
Before you start
- macOS 15 or later
- The TuriX app
Steps
Open the macOS Accessibility permission settings and grant TuriX the permission described in the quick-start.
Grant TuriX Safari Automation permission using the quick-start instructions.
Expected outcome
The two macOS permission areas named by the quick-start are configured for the subsequent run stage.
Guide limitations
- This is a permissions setup guide, not a task-execution recipe.
- The supplied setup text identifies the permission areas but does not reproduce their full click-by-click instructions.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 9 October 2026
- Original website title
- TurixAI/TuriX-CUA · Official homepage title · retained in its source language