Develop a browser extension that allows users to input natural text or voice commands, with immediate outcomes. The commands are passed to an LLM which determines what the user wants and runs the correct pre-configured browser APIS. I've already tested a prototype and it works really well with Google Gemini Flash model. You just need to pass to the prompt a JSON mapping of keywords/commands to APIs. For example:

User: "Clear history"

LLM: <Runs pre-made clear history API>

https://news.ycombinator.com/item?id=41334802