Chats: your agents, running on your machine, opened here
Last reviewed: 16 Sep 2026
Chats is where the agents that run on your own computer live: Claude Code, Codex CLI, or a plain shell. They run on your machine, started by the local console the installer put there, and you open them here. Nothing moves to our servers; what you type and what the agent prints passes through encrypted in transit and is not stored. ChatGPT and Claude on the web are different: they have nothing to attach to, so they connect from Connect and reach your files through the tunnel instead.
A shell is the exception on this page. It is a plain command line with no agent in it, kept because you sometimes want one.
What you need
Two things on the machine, both from the one-click install on Connect: the tunnel, which is how the platform reaches your machine at all, and the console, which runs the agents. If either is missing the page says which, and shows the one command to start it.
You also pair the browser with the machine once, as a step on Connect. A login alone cannot open a shell on your computer; the machine decides, by you typing a short code in the tunnel window. Pair once and you never see it again on that browser.
Chats run on this computer. Your understanding and your work are the same everywhere you sign in, but a running chat lives on the machine that started it. Another computer's page is empty until you install there.
What the page shows
Open now is what is running on the machine right now, live or idle, with the folder each one is in. Open fills the page with it. Beside Work opens it in the split instead.
Start begins a new one: Claude Code, Codex CLI, or a shell, in the project's folder.
Saved on this machine is every chat the agents have kept in that folder, newest first, whether or not it is running. Resume picks one up exactly where it left off. This is the part people miss: a conversation from last week is still there, and one click brings it back.
A chat keeps running when you leave. Close the tab and it carries on; come back and it is still there. Closing the tab and stopping the chat are different things, and the page asks which you meant.
Beside Work
With a chat open, going to Work or Understanding puts it in a split beside them, so you can watch what the agent is doing against what the project knows. Ctrl + ` shows and hides it. That pane is a terminal in the ordinary sense, and the word is used there deliberately.
Your own terminal still works
None of this replaces working in your own terminal. A Claude Code window you start yourself connects the same way, appears on Work under Who's here, and commits understanding like any other. This page is a convenience, never the only way in.
See also
- The console and the naetive command, the small program that makes this possible
- Connecting Claude Code