cua
cua copied to clipboard
trycua
Open-source infrastructure for Computer-Use Agents. Sandboxes, SDKs, and benchmarks to train and evaluate AI agents that can control full desktops (macOS, Linux, Windows).
Fixed: - Fixed agent trajectories sometimes missing the text & reasoning in the tool call results Added: - `screenshot_cua` tool so an MCP agent can observe the VM state before...
- The agent codebase uses `StrEnum` which was introduced in python 3.11
Will there be Gemini support for the Cua Computer Use Agents? Gemini 2.5 Flash is currently the best tradeoff on cost / quality for vision modeling and reasoning.
This section in the `Computer.__init__` method: ```python if shared_directories: for path in shared_directories: abs_path = os.path.abspath(os.path.expanduser(path)) # type: ignore[attr-defined] if not os.path.exists(abs_path): # type: ignore[attr-defined] raise ValueError(f"Shared directory does not...
Unable to complete a task to install Chrome, even after spending several minutes and $0.77 on OpenAI
[trajectories.zip](https://github.com/user-attachments/files/19874219/trajectories.zip)
I did everything from README and Claude Desktop is returning this 2025-04-21T08:26:10.326Z [cua-agent] [info] Initializing server... 2025-04-21T08:26:10.353Z [cua-agent] [info] Server started and connected successfully 2025-04-21T08:26:10.354Z [cua-agent] [info] Message from client:...
Double click doesn't work for some reason. I've tried changing the interval as well, no luck.
Hi C/UA team! While investigating this awesome project I discovered that the quickstart notebook (`notebooks/blog/build-your-own-operator-on-macos-1.ipynb`) has what I think is an outdated shape for the "wait" action. I attempted to...
Adds @emgeee as a contributor for code. This was requested by f-trycua [in this comment](https://github.com/trycua/cua/issues/132#issuecomment-2832673077)
![allcontributors[bot] avatar](https://avatars.githubusercontent.com/in/23186?v=4 "allcontributors[bot] avatar"){kind=avatar}