v1.4.11

The template now supports give the assistant "computer use"! It uses the tool to take screenshots and to simulate mouse and keyboard interactions on the pixel grid.

Our default implementation gives the agent a simple headless web browser and the ability to automate that browser with playwright, but the optimal implementation would be something more like a remote connection to a full cloud VM. Implementing this is left as an exercise to the end user, but some helpful customization instructions are provided in the README.