OpenCowork

Open-source desktop platform for multi-agent collaboration
Empowering AI agents with local tools, parallel teamwork, and seamless workplace integration.

OpenCowork Screenshot

中文文档 • Why OpenCowork • Features • Inspiration • Quick Start

🚀 Why OpenCowork?

Traditional LLM interfaces are often "environment-isolated islands." Developers spend 50% of their time copy-pasting code, terminal logs, and file contents between the chat and their IDE.

OpenCowork solves this by providing:

Local Agency: Agents can directly read/write files and execute shell commands with your permission.
Context Awareness: No more manual context feeding. Agents explore your codebase and logs autonomously.
Task Orchestration: Complex tasks (like "Refactor this module and update tests") are broken down and handled by specialized sub-agents.
Human-in-the-loop: You stay in control with a transparent tool-call approval system.

💡 Inspiration

OpenCowork is deeply inspired by Claude CoWork. We believe the future of productivity lies in a "Co-Working" relationship where humans provide direction and AI handles the iterative execution, tool manipulation, and cross-platform communication.

✨ Key Features

Multi-Agent Loop: A lead agent coordinates parallel teammates to tackle multi-dimensional problems.
Native Toolbox: Built-in tools for File I/O, PowerShell/Bash, Code Search, and UI Previews.
Messaging Integration: Bridge your local agents to Feishu/Lark, DingTalk, Discord, and more.
Persistence: Cron-based scheduling for automated daily reports or monitoring tasks.
Extensible Skills: Load custom logic via simple Markdown-defined skills and agents.

🛠️ Quick Start

Prerequisites:

Node.js >= 18
npm >= 9

git clone https://github.com/AIDotNet/OpenCowork.git
cd OpenCowork
npm install
npm run dev

🏗️ Architecture Overview

OpenCowork follows a three-process Electron architecture to ensure security and performance.

graph TB
  subgraph "Renderer (Agent Logic)"
    A[React UI] --> B[Agent Loop]
    B --> C[Tool System]
  end

  subgraph "Main (System Access)"
    F[SQLite]
    G[File System]
    H[Shell]
    J[Plugins]
  end

  A -.->|IPC| F
  A -.->|IPC| G
  A -.->|IPC| H
  A -.->|IPC| J

🌟 Use Cases

Autonomous Coding: Let agents refactor code, fix bugs, and write tests directly in your workspace.
Automated Ops: Schedule agents to monitor logs or system status and report to Feishu/Slack.
Data Research: Agents can scrape web data, process local CSVs, and generate visual reports.

🔗 Ecosystem Pairing

Codex-Manager

Repository: qxcnm/Codex-Manager
Recommended pairing: use Codex-Manager as the management or orchestration entry point for Codex-centric workflows, while OpenCowork handles local file operations, multi-agent execution, workplace messaging, and desktop automation.
Best for: teams that want to separate task management and workflow organization from local execution and office integration.
A simple way to think about it: Codex-Manager organizes the work, OpenCowork executes it in the real workspace.

📈 Star History

🤝 Contributing

We welcome contributions! Please see our Development Guide for more details.

Special thanks

RoutinAI

[RoutinAI](https://routin.ai/) is an enterprise-grade unified LLM API gateway that provides a single, type-safe interface to access over 100 leading large language models from the GPT, Claude, and Gemini families, including models such as gpt-5.4, claude-opus-4-6, and gemini-3.1-pro-preview. It eliminates the complexity of managing multiple AI vendors by providing zero-latency edge routing, seamless model switching without code modifications, unified billing, and centralized governance with spending caps and access policies.

💝 Sponsors

📜 License

Licensed under the Apache License 2.0.

If this project helps you, please give it a star. ⭐

Made with ❤️ by the AIDotNet Team