DEV.co
MCP Servers · TurixAI

TuriX-CUA

TuriX is an open-source desktop automation agent that uses AI to control your computer via GUI interactions—clicking, typing, scrolling—without needing app-specific APIs. It achieves 64.2% on OSWorld benchmarks and 80%+ on macOS-specific tests, with hot-swappable AI models and support for MCP integration.

Source: GitHub — github.com/TurixAI/TuriX-CUA
3.1k
GitHub stars
319
Forks
Python
Primary language
MIT
License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

FieldValue
RepositoryTurixAI/TuriX-CUA
OwnerTurixAI
Primary languagePython
LicenseMIT — OSI-approved
Stars3.1k
Forks319
Open issues4
Latest releasev0.4 (2026-03-06)
Last updated2026-06-08
Sourcehttps://github.com/TurixAI/TuriX-CUA

What TuriX-CUA is

Python-based agent framework for autonomous desktop control using vision language models (VLM) and computer vision. Supports model swapping via config, integrates with MCP for Claude Desktop, includes skill-based planning via markdown playbooks, and currently targets macOS (15+), Windows (multi-agent-windows branch), and Linux (multi-agent-linux branch).

Quickstart

Get the TuriX-CUA source

Clone the repository and explore it locally.

terminalbash
git clone https://github.com/TurixAI/TuriX-CUA.gitcd TuriX-CUA# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Multi-step office automation workflows

Orchestrate complex tasks spanning Excel, PowerPoint, email, and document tools—e.g., fetch data, generate charts, insert into presentations, send to contacts—without manual step-by-step intervention.

RPA and process automation for legacy systems

Automate interaction with in-house or third-party applications that lack APIs, using pure GUI automation. No app-specific connector code required.

AI agent integrations via MCP

Plug TuriX into Claude for Desktop or other MCP-compatible agents to extend their reach beyond text—enable Claude to execute desktop actions autonomously.

Implementation considerations

  • Requires Python 3.12 and macOS 15+ for primary branch; separate branches for Windows/Linux. Grant system accessibility and Safari automation permissions explicitly—TuriX cannot proceed without them.
  • Agent success rate is model-dependent; swapping VLMs in config.json can significantly change performance. Benchmark your chosen model on your specific tasks before deploying.
  • Skills (markdown playbooks) require authoring; planner integration means you must define task guides for domain-specific automation. Start simple, iterate based on failure logs.
  • Automation latency depends on vision inference time + screen update delays; not suitable for real-time or sub-second response requirements.
  • No built-in data collection telemetry (claimed), but ensure your VLM provider and underlying OS don't violate your compliance requirements.

When to avoid it — and what to weigh

  • You need production SLA guarantees — Project is actively developed but not production-hardened; 64% OSWorld success rate means 36% failure. Suitable for research and personal automation, not mission-critical enterprise workflows without extensive testing.
  • You require Windows or Linux as primary platform — macOS is the primary target (80%+ success rate). Windows and Linux support are on separate branches and not as mature; switching branches breaks a single codebase.
  • You need deterministic, reproducible automation — Vision-based agents are inherently stochastic; results depend on screen state, model inference, and timing. Not suitable for scenarios requiring 100% determinism or audit trails.
  • Strict data residency or air-gapped environment — Default models may require external API calls (configurable); ensure your chosen VLM aligns with your data governance policies.

License & commercial use

MIT License: permissive open-source license allowing use, modification, and distribution in commercial and private contexts, with only attribution required. No copyleft restrictions.

MIT License permits commercial use, modification, and redistribution without restriction or royalty. However, no warranty is offered; users assume all liability. Verify that your chosen VLM provider (configurable) permits commercial deployment. TuriX itself does not restrict commercial use, but your model license may. Requires review of end-to-end dependencies (OS, Python, VLM, data handling) to ensure compliance.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

SignalAssessment
MaintenanceActive
DocumentationAdequate
License clarityClear
Deployment complexityModerate
DEV.co fitGood
Assessment confidenceHigh
Security considerations

TuriX operates with elevated OS permissions (accessibility, GUI automation) and can access any GUI element on screen. Ensure the system running TuriX is trusted and isolated from untrusted users. Vision inference and VLM queries may log prompts or screen state depending on model provider—review your VLM's privacy policy. No explicit authentication, encryption, or audit logging documented for automation actions themselves. If used for sensitive tasks (banking, PII), verify data handling and consider air-gapped deployment with local VLMs (e.g., Ollama). Screen-based automation is inherently difficult to audit retroactively.

Alternatives to consider

UIPath or Blue Prism (enterprise RPA)

Production-grade RPA with formal SLA, governance, and extensive app connectors. Closed-source. Choose if you need 99.9% uptime and vendor support; TuriX better for research, cost-free automation, and open-source flexibility.

Claude for Desktop + native integrations

Closed-source agent with official OpenAI/Anthropic support, but limited to text and predefined tool calls. No GUI automation. Choose if you only need reasoning + API calls; TuriX needed if you must automate legacy GUIs.

Anthropic's Sonnet 3.5 computer-use (API)

Closed-source, commercial API for computer vision + GUI automation. Production support available but usage-based costs. Choose if you want managed service; TuriX better for self-hosted, free, or offline automation.

Software development agency

Build on TuriX-CUA with DEV.co software developers

Download TuriX SuperAgent or clone the repo on GitHub. Start with a simple workflow, tune your VLM in config.json, and scale. Join the Discord community for support.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

Related on DEV.co

Explore the category and the services that help you build with it.

TuriX-CUA FAQ

Can I use TuriX with my own AI model?
Yes. Edit config.json to swap the VLM. TuriX supports Qwen3-VL, Ollama, and other models. You manage the model endpoint and versioning.
Does TuriX collect telemetry or logs?
README states 'We never collect data,' but no detailed privacy policy or code audit provided. Verify your chosen VLM provider's data handling. OS-level automation logs may still be retained by macOS/Windows/Linux.
What's the success rate I should expect?
64.2% on OSWorld (Linux-based benchmark), 80%+ on TuriX's internal macOS benchmark. Real-world success depends on task complexity, screen design, and chosen VLM. Test on your specific workflows.
How do I use TuriX on Windows or Linux?
Switch to the multi-agent-windows or multi-agent-linux branch respectively. Each has separate setup instructions. Main branch is macOS-optimized.

Work with a software development agency

DEV.co helps companies turn open-source tools like TuriX-CUA into production software. Our software development services cover the full lifecycle — architecture, web development, integration, and maintenance — delivered by software developers and web developers who ship. Engage our software development agency to implement or customize it for your mcp servers stack.

Ready to automate your desktop?

Download TuriX SuperAgent or clone the repo on GitHub. Start with a simple workflow, tune your VLM in config.json, and scale. Join the Discord community for support.