One of my biggest LLM use cases right now: getting them to take over my browser for tedious tasks.
I’m getting lazier and lazier. The moment I hit a new interface, I log in once, then hand the rest to the agent.
The setup: a VPS running Docker, a real browser (not headless) running 24/7 inside it, and the agent connecting to it over CDP via websocket — like it’s grabbing the keyboard and mouse of a browser that’s already open, same profile, same cookies, same history. A freshly launched headless browser screams “bot” the moment a site pays any attention. Here, as far as the site is concerned, it’s just me coming back.
Save the cookies once = no more logging back in, and it often skips the login-time captcha entirely.
The one catch: some sites throw a random captcha mid-session, no warning. When that happens the agent stops everything, sends me a message, I go solve the challenge by hand, and it picks up exactly where it left off.
In practice it does everything I’d do myself in a browser: post on X, search a search engine without going through an API, reverse-engineer a piracy streaming site, set up an account in a new workout-tracking app, etc.
Video proof: the agent posting a tweet and deleting it itself, without me touching anything. 👇
Repo (docker-compose + everything an LLM needs to connect to it): github.com/sepiropht/llm-browser-agent
Originally posted on X