OpenWork

OpenWork is the free, open-source alternative to Claude Cowork and Codex: a desktop app for macOS, Windows, and Linux where AI agents do real work on your own files. It is built on OpenCode, works with any model — 50+ providers, your own API keys, or local models via Ollama — and lets teams share skills and MCP servers.

Download OpenWork · GitHub releases · Docs

Read this in: 简体中文 · 繁體中文 · 日本語

Add one OpenWork MCP to Codex, Claude Code, Cursor, or another compatible agent and reuse the same skills, MCPs, and connected services across your tools, teammates, and machines. Create something once, share it with coworkers or friends, or keep it for yourself.

The desktop app is there when you want a dedicated workspace, but it is not required. You can use OpenWork from the agent you already have. For larger organizations, the admin interface lets you publish capabilities, manage access, and configure shared or per-user connections.

Why OpenWork instead of Claude Cowork

• Free and open source. The desktop app is MIT-licensed; no OpenWork account is required to use it locally.
• Any model. Bring your own API key, sign in with ChatGPT, or run local models through Ollama or any OpenAI-compatible server. See Add a custom LLM.
• Runs on macOS, Windows, and Linux as a desktop app — not a CLI.
• Your files stay local. Cloud is optional.
• Share skills and MCP servers with your team. See Share skills with your team and Shared MCP connections.
• Self-host the team control plane on your own infrastructure. See Self-host.
• Bring your Cowork setup. Skills, Claude-compatible plugins, and MCP servers carry over. See Migrate from Claude Cowork.

Install with your AI agent

Already use an AI agent? Copy this prompt and paste it into Claude Code, Cursor, Codex, ChatGPT, or any agent that can run commands on your computer.

Install OpenWork on my computer, set up my first workspace, and open it ready to use. Follow the steps in https://openworklabs.com/start.md?v=hero

1. Installs OpenWork
2. Creates your workspace
3. Opens it ready to run

Use OpenWork from any agent

The OpenWork MCP brings your assigned skills, plugins, MCP connections, Google Workspace, and Microsoft 365 capabilities into any compatible agent.

It exposes four tools: search_capabilities finds what you can use, execute_capability runs it, and list_skills / get_skill list your skills and read one SKILL.md directly. After adding the MCP, your client opens a browser so you can sign in and choose your OpenWork organization.

Codex

codex mcp add openwork --url https://api.openworklabs.com/mcp/agent

Claude Code

claude mcp add --transport http openwork https://api.openworklabs.com/mcp/agent

OpenCode

Add this to opencode.json:

{
  "mcp": {
    "openwork": {
      "type": "remote",
      "enabled": true,
      "url": "https://api.openworklabs.com/mcp/agent",
      "oauth": {}
    }
  }
}

Any MCP client

Use this remote MCP server URL:

https://api.openworklabs.com/mcp/agent

OpenWork Den

OpenWork Den is the control plane for managing OpenWork across a team or organization.

• Provision inference at scale and control which members and teams can use each model provider.
• Invite teammates, create teams, and manage access from one place.
• Set desktop policies, restrict local model access, and control which app versions your organization can use.
• Publish skills and plugins through marketplaces, then assign them to the organization, a team, or specific people.
• Import Agent Plugins or Anthropic-compatible plugins and make their supported skills and remote MCPs available through the OpenWork MCP.

Licensing

This repository uses a directory-split license, similar to GitLab:

• Everything outside ee/ is MIT — the desktop app and core platform are open source, free for any use.
• Everything under ee/ (OpenWork Den — the org control plane) is under the OpenWork EE License, a source-available license. The code is public so you can audit exactly what you deploy. Production use requires an OpenWork subscription, except that it is free for organizations with up to 5 users, free to evaluate for 30 days at any size, and always free for development and testing. Each ee/ release additionally converts to MIT two years after publication.

Versions released before this license was adopted remain under their original license (FSL-1.1-MIT). See pricing and the subscription terms.

Documentation

Read the OpenWork docs.

Getting started (contributors)

The fastest path from a fresh clone to a running dev build.

Prerequisites

• Node 24 — pinned in .nvmrc (nvm use picks it up).
• pnpm 11 — pinned in package.json (packageManager); run corepack enable to use the pinned version automatically. Never use npm or yarn.
• Git with DCO sign-off — every commit needs a Signed-off-by trailer (git commit -s). See CONTRIBUTING.md.

First run

git clone https://github.com/different-ai/openwork.git
cd openwork
corepack enable
pnpm install
pnpm dev   # launches the Electron desktop app with hot reload

Repository layout

Testing

All executable coverage lives in evals/specs/**/*.test.ts; app-driving journeys use .e2e.test.ts.

pnpm --dir evals install --frozen-lockfile   # once
pnpm evals:pr specs/<name>.test.ts           # app-less PR-lane spec
pnpm evals:e2e <name> --local                # app-driving E2E journey, run locally

Runtime-observable changes need test evidence on the PR. AGENTS.md and evals/README.md describe the verification contract and vocabulary.

Sending a pull request

1. Branch from dev (the default branch) and open your PR against dev.
2. Sign off every commit: git commit -s.
3. Keep the diff as small as possible, and include or update test evidence for runtime-observable changes.
4. Read CONTRIBUTING.md for the DCO and licensing rules — contributions under ee/ additionally require a CLA.

Local development

For one checkout, keep using pnpm dev; with no extra environment variables it reuses the existing shared dev profile.

To run multiple git worktrees at once, use:

pnpm dev:worktree

That sets OPENWORK_DEV_PROFILE=auto, derives a stable profile name from the worktree path, lets Electron choose a free CDP port, and asks Vite for a free dev-server port. You can also choose a named profile, for example OPENWORK_DEV_PROFILE=my-feature OPENWORK_ELECTRON_REMOTE_DEBUG_PORT=0 PORT=0 pnpm dev.

dev:worktree also defaults OPENWORK_ELECTRON_USE_MOCK_KEYCHAIN=1. A brand-new profile has no stored credentials, so on macOS the real keychain prompts as soon as Chromium persists an authenticated cookie, and that modal blocks Electron's main loop until it is dismissed. Set OPENWORK_ELECTRON_USE_MOCK_KEYCHAIN=0 if you specifically want the system keychain in an isolated profile.

Dev startup prints a banner like [openwork] dev profile=... cdp=http://127.0.0.1:9823; use it to find the profile directory and pass the CDP URL to local tooling.

If a second instance cannot get the profile lock it now says so and exits, instead of lingering with an open CDP port and no window.

Headless web (no Electron)

To run the OpenWork UI in a browser against a local openwork-server (no desktop shell):

pnpm world up dev-headless --detach

pnpm dev:headless-web is a compatibility alias for the same script. The alias remains foreground by default and accepts --detach; world up is foreground unless --detach is explicit.

This is an isolated launcher: