It could be used by AI agents on servers. People check and work with an agent on the same browser.
It sync DOM/CSS between mini runtime and the desktop browser.
I’m building mini-runtime: a web page’s JS and DOM run outside the browser. Browse through a viewer, close the tab, then reconnect and pick up where you left off.
An experiment in reducing browser memory use. Early days, partial web compatibility.
https://t.co/L8ur6eLsy6
However, it may fail several times to find a right way. So, this is what predefined scripts are for. Once it figures out a right way, you can ask it to create a shell script to generate JS script for this purpose. Then, next time, it can/will reuse it without trial and error.
https://t.co/UiHr4dJUCN
Expose windows, tabs, history, bookmark functions to AI model through MCP. Be careful about potential security issues. This project is to provide a platform to explore the potential power of AI and browser.
About injecting JS code, it usually happens automatically. You may said "Please reply the email about XXX in my gmail. Tell him, I will be there." Then, Claude Code would be smart enough to find the gmail tab and find the message, and inject script to reply the email.
@TheFigen_ Right one has a better strategy, however she is slow. For each ball, you need to move from the bucket to the ball. The distances are never changed no mater the order, except the first one you pick up. The left girl wast the move for her first ball, but she is quicker.
@khuey_ Parking is a good business in urban areas. Many property owners in Taiwan make their life by running unmanned parking lots. My dad has a property next to an area just planned to develop. I suggest doing a parking business if that happens.