Agents nowadays can research anything but complete nothing; practically every task ends with the model handing you a link to go click yourself. mkrrm is my attempt at fixing that Under the hood, everything runs locally on your machine. The agent gets their own browser; separate from yours, invisible, running on your hardware, which can optionally share your browser profile, so it's signed into what you're signed into without you setting anything up. Nothing is sent to us, nothing is stored on our servers, and you never type a password anywhere new. From the agent's side, services show up as discoverable capabilities that return structured results, instead of a screen it has to screenshot and guess at. The waitlist is a demo of the idea: there's no form. Paste "tell me all about mkrrm.com" into your agent, they'll fetch mkrrm.com/lIms.txt, explain the product back to you, and registers you itself with one POST. Watching different models interpret the same Ilms.txt has been half the fun....