Terminal and processes
A terminal in every thread, and the long-running processes, like dev servers, that agents start.
Open a terminal
Every conversation has a terminal of its own in the side panel. Use it to run commands yourself in the folder the agent works in, such as your tests or a dev server.
- Open a conversation and press ⌘J, or click Show panel in its header, to show the side panel.
- Click New tab (+) and choose Terminal.
The Terminal tab then stays in that conversation's panel. A task's side panel has a terminal too. Each further Terminal tab opens a shell of its own, named Terminal 2, Terminal 3, and so on, which ends when you close its tab. The first keeps the conversation's own shell.
If an agent signs out while you work, the transcript offers Sign in. It opens this terminal and runs the agent's sign-in command, such as claude auth login.
Shell and folder
The terminal starts in the folder the Changes tab reads: the conversation's worktree if it has one, otherwise the project folder. A Workspace conversation uses its own folder. The header shows the end of the path.
The shell is the one set for your user account, usually zsh on macOS. On macOS it starts as a login shell from the system's PATH, as Terminal does, so zsh reads ~/.zprofile and ~/.zshrc and builds the rest itself. On Linux it starts as an interactive, non-login shell, with the PATH OpenOrc found by asking your login shell, the same one agents get. On Windows it is the shell COMSPEC names, or PowerShell. OpenOrc also sets TERM=xterm-256color, COLORTERM=truecolor, and TERM_PROGRAM=OpenOrc.
When a command you typed finishes, the Changes tab refreshes, so edits you make by hand show up next to the agent's.
How long a shell runs
A shell keeps running when you switch tabs, close the panel, move to another conversation, or reload the window. When you come back, the terminal replays its recent output, up to the last 256 KB.
| Button | What it does |
|---|---|
| End shell | Stops the shell and forgets its output. |
| Restart shell | Stops the shell and starts a new one in the conversation's folder. |
When a shell exits, the terminal says so and offers Start a new one. Quitting OpenOrc ends every shell.
Processes agents start
Agents run commands with their own tools, outside your terminal. Most commands end within the turn. Some keep running, such as a dev server or a file watcher.
Claude Code reports the commands it leaves running in the background. OpenOrc lists them above the composer, each with its description, Running, and a Stop button. Codex and OpenCode don't report theirs, so OpenOrc can't list them or stop them one by one. Ask the agent to stop a server, or end it yourself.
When the turn ends
A background command that OpenOrc lists outlives the turn that started it. The turn ends and the conversation stops showing as working, while the command keeps running. A dev server stays up for you, for the Preview, and for the agent's next turn.
The command belongs to the agent's process, and closing that process would end it. So OpenOrc keeps the process open while any of its commands run, even past the idle time that normally closes it. A command ends when:
- you click Stop next to it,
- it exits on its own, or
- the agent's process closes. That happens when you quit OpenOrc, delete the conversation, or send a message after switching the conversation's agent or mode, which starts a new process.
If OpenOrc quits unexpectedly on macOS or Linux, it stops the agent processes left over from that session the next time it starts.
Next steps
Checked against OpenOrc's source on November 9, 2026.