CoCo

Run parallel Codex agents. Keep their work organized.

CoCo — Codex Coordinator — gives each agent a named workspace with its own checkout and conversation. Start work across repositories, see which agents need attention, and return to any conversation from your terminal.

Work on several things at once

Let one agent fix a bug while another investigates a different part of the project. Each works in a separate Git checkout, called a worktree, so your current checkout stays available.

Create and navigate workspaces

Find the work you need

See activity across repositories with coco status -a. Use coco jump to pick a workspace and open its conversation in the Codex terminal UI. You can leave that UI while work continues and return later.

Start your first agents

Connect agents to your workflow

An agent can publish a signal such as “ready for review,” with fields you define. CoCo saves it for you and other tools to read. A hook runs your program when that signal arrives or a workspace is created, closed, reopened, or deleted.

Use these building blocks to send notifications, update another tool, or connect your own automation. Applications can also inspect workspaces and send instructions through CoCo’s MCP server.

Publish agent updates · Run hooks · Connect through MCP

Control actions and resource use

Add a guard to check your own rules before a workspace is closed or deleted. Watch CPU, memory, token usage, and your remaining Codex account quota. Set resource limits on supported Linux systems.

Protect workspace actions · Monitor usage and set limits

Get started

CoCo uses Codex App Server and your existing Codex login and configuration. It runs on Linux and macOS. Keep the cocod coordinator running during active work; stopping it interrupts running turns while preserving saved conversations and workspaces.

Install CoCo, then follow the quickstart.