OpenCoworkAI/open-cowork

Open-source AI agent desktop app for Windows & macOS. One-click install Claude Code, MCP tools, and Skills — with sandbox isolation, multi-model support, and Feishu/Slack integration.

Open Cowork – A Desktop AI‑Agent Workspace

What it is – Open Cowork is an open‑source desktop application (Windows & macOS) that lets you run a Claude‑style AI agent locally. The app bundles a GUI, one‑click installers and a sandboxed execution environment so the model can read/write files, generate Office documents, control other programs, and be reached from remote chat tools (Feishu/Lark, Slack). It is essentially a “personal AI coworker” that you can point at your own data and let it act on your behalf.

Key capabilities

Feature What you get
One‑click install Pre‑built .exe (Windows) and .dmg (macOS) installers; also Homebrew cask for macOS.
Model flexibility Supports Anthropic/Claude, OpenAI‑compatible APIs, and Chinese models (GLM, MiniMax, Kimi) via configurable API keys and base URLs.
Sandbox isolation Basic path‑guarding plus optional VM‑level isolation – WSL2 on Windows, Lima on macOS – so commands run inside an isolated Linux VM.
Skills library Built‑in workflows (pptx, docx, pdf, xlsx) that turn prompts into PowerPoint, Word, PDF or Excel files; a skill‑creator tool lets you add custom skills.
MCP (Model Context Protocol) Connectors for browsers, Notion and other desktop apps, extending the agent’s toolset beyond the file system.
Remote control Integration with Feishu (Lark) and Slack so you can issue commands from those collaboration platforms.
GUI automation The agent can drive ordinary desktop applications; the README recommends Gemini‑3‑Pro for the best GUI‑understanding.
Multimodal input Drag‑and‑drop files or images directly into the chat window.
Trace panel Real‑time view of the model’s reasoning and tool calls.

How to get started

  1. Install – either brew tap OpenCoworkAI/tap && brew install --cask --no-quarantine open-cowork (macOS) or download the installer from the Releases page. Linux users can build from source (npm install && npm run dev).
  2. Configure an API key – open the Settings pane, paste an OpenRouter/Anthropic/GLM/etc. key, set the base URL (required for non‑Anthropic providers) and choose a model (e.g., claude-4-5-sonnet or gemini-3-pro).
  3. Pick a workspace folder – this directory is the only place the agent may read/write files.
  4. Prompt the agent – type natural‑language commands such as “Read financial_report.csv and create a 5‑slide PowerPoint summary.” The agent will use the appropriate Skill, run any needed commands inside the sandbox, and stream results back to the UI.
  5. Optional extras – enable WSL2 (Windows) or Lima (macOS) for stronger isolation, connect a Notion integration token, or set up Feishu/Slack remote control via the app’s settings.

Safety notes

  • The sandbox restricts file access to the chosen workspace, but when a VM is not present commands run directly on the host, so users should still be cautious with prompts that delete or modify files.
  • All data stays on the local machine; the only external traffic is the API calls to the chosen LLM provider.

Community & contribution

  • Discord: https://discord.gg/pynjtQDf
  • WeChat: QR code in the repo’s resources/WeChat.jpg
  • Contributions are welcomed via typical fork‑branch‑PR workflow.

License – MIT.

All details above are taken directly from the repository’s README.

Related

  • Project
    iOfficeAI/AionUi

    A free, open-source Cowork platform that integrates built-in and external AI agents to automate file operations, document creation, and multi-step tasks on a user's computer.

  • Project
    musistudio/claude-code-router

    Claude Code Router (CCR) is a local model gateway that unifies many coding‑focused AI agents (Claude Code, Codex, Grok, Kimi, etc.) behind a single endpoint. It lets you manage providers, API keys, routing rules, retries, and tool extensions from a desktop UI, CLI, or Docker container, and provides detailed logs and cost estimates for every request.

  • Project
    AIDotNet/OpenCowork

    OpenCowork is an open‑source desktop (Electron + .NET) platform that lets LLM‑based agents work directly on your local files, shell, and repositories. It supports multi‑agent teams, integrates with many chat services, offers five interaction modes (chat, clarify, cowork, code, acp), and provides a rich toolbox (file/shell access, browser automation, task/cron management, CodeGraph indexing, skill extensions). The same native .NET worker powers both the GUI and a terminal‑based CLI, keeping a single authoritative agent loop. Quick start requires Node ≥ 18 and .NET 11 SDK; builds are available for Windows, macOS, and Linux. Use cases include autonomous coding, scheduled ops, research/report generation, remote SSH tasks, and media‑pipeline design. Licensed under Apache 2.0.

  • Project
    CoWork-OS/CoWork-OS

    An open-source AI super app that integrates coding, email, research, and document creation into a single workspace with support for multiple AI providers.

  • Project
    chenhg5/cc-connect

    cc‑connect is an open‑source bridge that lets you run AI agents (Kimi, Claude, Copilot CLI, etc.) on your own machine and control them from chat apps like Slack, Discord, Telegram, Feishu, QQ, and more. It forwards messages to the local agent, returns the agent’s response, and supports many platforms and agents out‑of‑the‑box.