clowder-ai: Coordinate AI Agents as a Persistent Team

Overview

Clowder AI is a TypeScript platform layer for turning separate AI agents into a persistent team. Instead of manually routing work and copying context between agents, developers can coordinate agents through shared identity, memory, and agent-to-agent communication. It works with existing agent CLIs such as Claude Code, Codex CLI, Gemini CLI, and opencode rather than replacing them.

The project is aimed at people building or operating software with several AI agents, especially when work needs review, handoffs, or continuity across sessions. It offers a desktop installer for Windows and macOS, as well as a source-based setup that opens a local web interface.

Key Features

• Route work between supported agent CLIs, with agents retaining roles and personality across sessions.
• Enable cross-model review, such as having one agent produce work and another review it.
• Send asynchronous agent-to-agent messages with @mention routing and structured handoffs.
• Keep shared evidence, lessons, and decision logs as persistent team memory.
• Create conversation proposals that split a larger task into reviewable pieces of work.
• Steer an agent while it is replying, or bring an existing local agent session into a conversation.
• Connect MCP tools, messaging adapters, and on-demand skills through a plugin framework.
• Run scheduled tasks and respond to GitHub events to support work that continues without an operator present.

Use Cases

• A software developer can assign implementation and review to different agent CLIs when they want cross-model feedback.
• A team using agents across multiple sessions can keep decisions and lessons in shared memory instead of rebuilding context each time.
• An operator can hand off a long-running task when an agent reaches a quota limit, rather than restarting the work manually.
• A project maintainer can configure scheduled work or GitHub event responses for agent-assisted repository tasks.

Getting Started

The README documents a source setup with pnpm:

git clone https://github.com/zts212653/clowder-ai.git
cd clowder-ai
pnpm install
pnpm start

See the README and setup guide for prerequisites and further instructions. Desktop installers for Windows and macOS are linked from the releases page.

Alternatives

• oh-my-claudecode: It orchestrates parallel Claude Code work through in-session skills or terminal workers, rather than coordinating a broader set of agent CLIs.
• ruflo: It provides agent teams, persistent memory, workflows, and MCP integrations specifically as a Claude Code and Codex harness.
• cli-agent-orchestrator: It delegates coding tasks to agents running in isolated tmux sessions, with a supervisor managing parallel or sequential work.
• omnigent: It governs multiple coding agents through a shared harness across terminal, browser, phone, and desktop interfaces.

Considerations

The repository is active, with regular commits and releases, and contributions are not concentrated in one person. Its reported issue and pull-request handling is quick, though the project is still relatively young. The available project facts do not indicate an automated test suite, so teams should evaluate their own validation needs before relying on it in critical workflows.

There is a requirements discrepancy to resolve before installation: the README quick start says Node.js 20+, while the package metadata specifies Node.js 24 or later. Redis 7+ is optional, and the README says --memory can be used to skip it. The project is MIT-licensed; its README separately identifies the name, logos, and cat character designs as brand assets, with details in TRADEMARKS.md.