Local agent: connect Claude Code and Codex side by side, and pick the model

New

Connect Claude Code and Codex together. The connect dialog now has agent tabs and only offers backends that are ready.

One agent per person per backend. Your Claude Code agent and your Codex agent sit side by side, and you pick one when you hand work off.

Pick the model and the reasoning level. What you picked is written on the chip next to the answer, and light requests run at a lower level so they come back faster.

Working folders are now one default folder plus per hub exceptions.

Borrow a teammate agent. A shared agent shows up in the / picker and in the assign menu.

When the agent gets stuck it asks with options, and continues in the same session once you answer.

An @name written by the agent is a real mention, notification included.

Result files open in the right panel. Markdown and text open as documents, and an uploaded HTML file opens as the page it is.

You can call the agent in DMs and group chats too.

How to use it

1. Install and open the desktop app. If you press connect in the browser, it points you to the app.

2. Connect your local agent. In the dialog, pick the Claude Code or Codex tab, check that it is installed and signed in, then name it.

3. Set the working folder. Pick one default folder, and add an exception for a hub that uses a different one.

4. In a thread or ticket, press / and pick the agent. Once the chip is attached, write the task the way you would tell a person.

5. Pick the model and the reasoning level. Raise it for long work, lower it for a quick check.

6. The result comes back to the same thread. Open the file in the right panel, and keep it as a to-do or a note.

Good to know

The agent runs on your computer. If the computer is off or the app is closed, nothing runs.

If your Claude sign-in drops, you get a sign in again chip instead of an answer. Sign in and it picks up from there.