Watch it run. Then copy it.
A good AI result rarely comes from one prompt. It comes from everything around it: the instructions, the tools, who checks what, and what the person stopped doing because it didn’t work. A setup is all of that, packaged so you can see it working and run it yourself.
Pick a job.
Start with what you want AI to do better. Every listing says what it does, who made it, what it costs to run and what you have to bring.
Watch it run.
A listing with a run shows the real thing: who did what, where a person stepped in, and what came out. Every run says who ran it. No run yet? The page says that too.
Copy it, then check it.
Preview every file before it lands in your project. Nothing of yours is overwritten unless you say so. Then run the first task and see if it passes on your machine.
It goes where you work.
A setup is stored in a format that isn’t tied to one tool. Today it exports instructions for Claude Code, Codex, Cursor and OpenCode.
You connect your own accounts and keys in your own tools. They never pass through LiveSetup.
Find a setupBuilt one worth copying? Sell it.
Bring the files you already have. You get a draft listing and a few questions about what the files can’t say. You set the price and publish updates, and the people who copied it see what changed.
Some listings were written by AI and some were indexed from public repositories. Both are labeled. If an indexed one is yours, claim it.
Bring your setupWhat we don’t do
See the sourcesWe don’t invent reviews, sales or numbers. A claim says who made it and a run says who ran it. If nobody has run a setup, its page says so. Public sources keep their license and their author’s name, and nothing gets packaged until the license has been read. We don’t run setups on our servers: the runs you see were recorded by the people who ran them.