Action is a macOS launcher that can take actions (click, type, and more) on your Mac using Claude’s Computer Use API.
The interface is a floating window triggered by a keyboard shortcut, similar to Spotlight. This lets you see what the model outputs as it performs actions. It’s written in Swift, and the DMG size is relatively small at roughly 3 MB. You can download the macOS client today at https://action.new (requires Anthropic API key).
To be frank, LLMs’ computer use capabilities (including Claude) are still a little slow and error prone. I find Action most valuable as an LLM that has immediate context of my screen.
I’ll be adding support for other providers—and hopefully local computer use models—as they become available.
Anthropic Computer Use however does it for me - have you guys chosen to restrict these types of tasks? if so, why?