very cool stuff!
5 karma · joined February 9, 2025
Run models like Gemma3, Phi4, Qwen2.5, Llama3.1 & more supported by Ollama, keeping your data completely private - no cloud required.
Agent with local LLMs combines UI grounding with pixel-detection for accurate computer control with Pyautogen OmniParser - all in Cua's sandboxed environment for security.
Privacy-conscious? Local models through Ollama run entirely on your machine while maintaining the same powerful agent capabilities backed by MPS on Apple Silicon.
Simple to set up: just `pip install "cua-agent[all]"` and connect to your local Ollama models. Check out our examples on our repo: https://github.com/trycua/cua
Today we're launching a new blogpost series to help. First up: Build Your Own Operator on macOS
A step-by-step guide to pairing OpenAI's computer-use-preview model with a macOS VM sandbox.
Why build your own instead of using ChatGPT's Operator?
• Control native macOS apps, not just web
• Better privacy with local VMs
• Full access to system-level operations
• Superior performance on your hardware
Our guide covers everything you need:
• VM setup with Lume CLI
• Connecting to OpenAI's model
• Building the action loop
• Complete working Python code
Link here: https://www.trycua.com/blog/build-your-own-operator-on-macos...
Discord for questions: https://discord.com/invite/mVnXXpdE85
Stay tuned for Part 2 introducing our high-level Agent framework.
- Locally secure and isolated from your main system
- Pyautogui-compatible (screenshot, left_click ..)
- Configure and customize in just 2 commands
- Near-native performance on Apple Silicon
Maximize your host system’s resources, share only the files and folders you need, and leverage an open-source solution that’s compatible with any AI agent library (OpenAI Agents SDK, LangChain, CrewAI, AutoGen).
Quickstart: https://t.co/tTUDtHCl8H