Loops

How it works

1. Press Loop at the top of the sidebar, or ⌘N from anywhere, to open the workspace's new loop. Once you send, it is listed under Loops, over the tabs.
2. Origami drives Claude Code, Codex or GitHub Copilot as you signed it in, so no key is needed.
3. A loop's line shows where it works: its worktree, the checkout's branch, or its folder. A new loop in a repository can have a worktree of its own.
4. When the work is finished, press Close on its line. Reopen, typing in it, or a question reaching it brings it back, with the agent's memory of the work.
5. The arrow beside Loop has Find loop, which searches names and messages, and Closed loops.

Good to know

• A loop is named by its agent after your first message, or by you with Rename in its menu.
• A workspace has one draft. What you type or attach in it is kept, and the button reads Draft meanwhile. A loop left with a message unsent shows a yellow dot.
• Claude Code's terminal commands, such as /clear, are not offered in a loop, and one typed is not sent.
• The new page offers starters, such as a failed tab. A press fills the box and sends nothing.

Related

• Loop worktreesA new loop works in the workspace's checkout, or in a worktree and on a branch of its own.
• AttachmentsWhat you hand a loop beside your words, from artifacts and cards to files, tabs and pictures.
• ModelsThe model a loop runs, how hard it thinks, how fast it runs, and which agent runs it.