CLI Agent Orchestrator (CAO) is an Apache-2.0 Python tool that runs a local cao-server and coordinates multiple AI coding CLIs, such as Claude Code, Kiro, Codex and Cursor CLI, as supervisor and worker agents inside isolated tmux sessions, and it is aimed at developers and teams who already have at least one of those provider CLIs installed and authenticated and want them working together.

What it is

CAO coordinates multiple AI coding CLIs so that a supervisor can delegate work to specialist agents in parallel or in sequence. It runs a local cao-server, starts provider CLIs in isolated terminal sessions, and gives the supervisor tools for coordinating workers. The agents remain full CLI processes with their native authentication and capabilities, so CAO sits above the provider CLIs rather than replacing them. Runtime architecture and package layout are documented in CODEBASE.md.

The concrete problem it solves is the manual work of running several coding CLIs at once on one project. Without a coordinator, a person has to open each terminal, start each CLI, route tasks between them by hand and track which agent is doing what. CAO makes that routing explicit: the supervisor holds the delegation tools, each worker runs in its own tmux session, and the sessions can be listed, attached to, or shut down as named units. It lives in the ecosystem of terminal-based AI coding assistants, and what it replaces is the improvised, person-in-the-middle arrangement of separate interactive CLI sessions.

Key capabilities

• Runs a local cao-server control plane that starts provider CLIs in isolated tmux sessions, leaving each agent as a full CLI process with native authentication.
• Provides supervisor-to-worker delegation for parallel or sequential work through a supervisor profile installed with cao install code_supervisor and launched with cao launch --agents code_supervisor.
• Supports twelve provider CLIs: Kiro CLI, Claude Code, Codex CLI, Antigravity CLI, Hermes, Kimi CLI, MiniMax Code, GitHub Copilot CLI, OpenCode CLI, Oh My Pi (OMP) CLI, Cursor CLI and Grok Build CLI, each with a focused provider guide.
• Offers several control planes: a Web UI at http://localhost:9889, a shell CLI, an operations MCP server, and plugins.
• Defines flows and workflows for scheduled runs and multi-step pipelines.
• Includes skills for installing, scoping and authoring reusable agent guidance, plus a memory component.
• Manages session lifecycle with cao shutdown --session {session-name} for one named session and cao shutdown --all for every session, and updates installs with cao update.

Who uses it and how

• Developers who already run one or more of the supported provider CLIs locally and want several of them working on the same project instead of one at a time.
• Teams that split work by specialty, using a supervisor profile to hand tasks to worker agents in parallel or in sequence while keeping the same codebase directory.
• Operators who want a browser view of the agent fleet through the Web UI or MCP Apps, rather than watching only terminal output.
• Users running scheduled or multi-step pipelines through flows and workflows, where agent runs are triggered rather than typed by hand.
• Container-based users, who can install CAO through the documented devcontainer feature instead of a local tool install.

Getting started

Install CAO as a uv tool with uv tool install git+https://github.com/awslabs/cli-agent-orchestrator.git@main --upgrade, or install the tagged release cli-agent-orchestrator from PyPI. Then install the built-in supervisor profile with cao install code_supervisor, leave cao-server running in one terminal, and launch the supervisor from the project directory with cao launch --agents code_supervisor.

How it compares

The tools CAO names are the provider CLIs it coordinates, not competitors: Claude Code, Kiro, Codex CLI and the rest continue to run as themselves, with their own authentication and capabilities, and CAO adds the supervisor and worker layer around them. No peer orchestration project is named in the facts, so within this registry CAO stands alone in its category rather than as an alternative to a listed paid product.

When to use it — and when not to

A self-hoster must operate a local cao-server and satisfy the prerequisites: Python 3.10 or later, tmux 3.3 or later, uv, and at least one supported provider CLI authenticated before launching CAO, so anyone unwilling to run a long-lived local server or install tmux should look elsewhere. The project carries 151 open issues against 1,312 stars and 272 forks, which is a large backlog for a tool this size and worth weighing before adopting it in a critical workflow. It is also a poor fit for anyone who wants a single self-contained coding assistant, since CAO adds coordination only and provides no provider CLI of its own.

uv tool install git+https://github.com/awslabs/cli-agent-orchestrator.git@main --upgrade
cao --help

cao update