Quickstart

Give two agents separate workspaces, then move between their conversations.

Before you begin, install CoCo and run cocod in another terminal. Leave it running throughout this walkthrough. Then open a Git repository with at least one commit. CoCo adds it automatically when you create the first workspace.

Start two agents

coco create fix/login -s "Investigate the login flow and fix a bug if you find one"
coco create review/cache -s "Review the cache and report possible correctness issues"

Choose instructions that fit your project. Each agent gets a separate checkout and conversation. The commands return once Codex accepts the work.

Watch their progress

coco status -f

The display updates as agents work, become ready, or need your input. Press Ctrl+C to leave the display; their work continues.

When a workspace needs approval or an answer, status shows a coco decide <decision-id> command. Run it in another terminal, or open the conversation as shown below.

Jump into a conversation

coco jump fix/login

This opens the workspace in the Codex terminal UI. Read its response, answer questions, and continue working. Use /quit or /exit to leave the UI while keeping active work running.

Run coco jump without a name to choose a workspace interactively.

Continue from your terminal

Once the agent is ready, send another instruction:

coco send fix/login "Add a regression test for the change" --wait

--wait prints the response when the turn finishes. Omit it when you want the command to return while the agent works.

Next steps

• Run and navigate agents, including across repositories.
• Choose code and context for a new workspace.
• Monitor usage and set limits.
• Review and clean up when work is finished.
• Publish agent updates for your own automation.