different-ai/openwork

The open-source alternative to Claude Cowork (powered by opencode)

OpenWork – Overview

What it is: OpenWork is an open‑source desktop application (macOS, Windows, Linux) that lets AI agents work directly on the files stored on your computer. It acts as a bridge between any LLM you choose (OpenAI‑compatible APIs, local models via Ollama, or 50+ other providers) and a set of skills and MCP (Model‑Control‑Plane) services that the agents can call.

Why it matters:

  • Provides a free, MIT‑licensed alternative to proprietary products such as Claude Cowork and Codex.
  • Keeps your data local – the cloud is optional.
  • Works with any model, so you can use hosted APIs, your own keys, or self‑hosted models.
  • Includes a team‑oriented control plane (OpenWork Den) for managing model access, permissions, and shared skills.

Core Features (as described in the README)

Feature What it does Why it’s useful
Cross‑platform desktop app Electron‑based UI built with React; runs on macOS, Windows, Linux. Gives a graphical workspace for agents without needing a CLI.
Model‑agnostic Connects to any OpenAI‑compatible endpoint, custom API keys, or local Ollama models. Freedom to pick the best‑performing or cheapest model for your workload.
MCP (Model‑Control‑Plane) integration Exposes four tools (search_capabilities, execute_capability, list_skills, get_skill) that any compatible agent can call. Allows agents like Claude Code, Cursor, Codex, or OpenCode to invoke shared skills and plugins stored in OpenWork.
Skill & plugin sharing Skills are markdown‑based (SKILL.md) and can be published to a team marketplace. Teams can reuse the same automation logic across projects and members.
OpenWork Den (org control plane) Web‑based admin UI for provisioning inference, managing teams, setting desktop policies, publishing skills, and importing Anthropic‑compatible plugins. Centralised governance for larger organisations; can self‑host the control plane.
Self‑hostable The core desktop app is MIT‑licensed; the Den component is source‑available under an EE license (free for ≤5 users, free trial, otherwise subscription). Gives you the option to run the entire stack on your own infrastructure.
Headless web mode Runs the UI in a browser (no Electron) for testing or CI environments. Faster feedback loops for developers; easier automated testing.
Developer tooling Uses Node 24, pnpm 11, and provides scripts for dev, worktree, headless, and world‑based environments. Streamlined onboarding for contributors.

Typical Use‑Case Flow

  1. Install – Download the binary or run the “install via agent” prompt from an existing AI assistant.
  2. Create a workspace – The app creates a local folder that becomes the root for all file operations.
  3. Add an MCP – Point your AI agent (Claude Code, Codex, etc.) at https://api.openworklabs.com/mcp/agent using the provided CLI snippets.
  4. Run capabilities – From the agent, call search_capabilities to discover what you can do, then execute_capability to have the model edit files, run scripts, query Google Workspace, etc.
  5. Team collaboration – In OpenWork Den, an admin publishes skills or plugins; teammates add the same MCP and instantly gain access.
  6. Self‑host (optional) – Deploy the Den control plane on your own servers for full data sovereignty.

Architecture Snapshot (from the repo layout)

  • apps/ – UI (apps/app), Electron shell (apps/desktop), and the backend server (apps/server). All MIT‑licensed.
  • packages/ – Shared core libraries used by both the desktop and server components.
  • ee/ – The Den control‑plane code, under a source‑available EE license.
  • evals/ – Test specifications built on @openwork/testkit for both unit and end‑to‑end validation.
  • worlds/ – Scripts that spin up declarative development/test environments (e.g., headless web, live desktop).
  • .opencode/skills/ – Repository‑wide skill definitions that agents can load.

Licensing & Pricing (as per README)

  • Desktop app & core platform – MIT license, free for any use.
  • OpenWork Den (EE) – Source‑available; free for up to 5 users, free 30‑day trial for larger orgs, otherwise requires a subscription. EE code becomes MIT‑licensed two years after each release.

Getting Started (for developers)

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

The repo also provides scripts for headless web (pnpm dev:headless-web) and worktree‑friendly development (pnpm dev:worktree). Tests live under evals/specs/**/*.test.ts and can be run with pnpm evals:e2e ….

Bottom Line

OpenWork is a genuine, production‑grade open‑source project that lets you attach any LLM to a local, file‑aware desktop environment and share reusable automation skills across a team. Its modular MCP design, cross‑platform UI, and optional self‑hosted control plane make it a solid alternative to commercial AI‑coworking platforms.

All information above is taken directly from the repository’s README; no external assumptions have been added.

Related

  • Project
    OpenCoworkAI/open-cowork

    Open Cowork is a free, open‑source desktop app that wraps Claude‑style (and other) LLMs into a local AI‑agent workspace. With one‑click installers for Windows/macOS, VM‑level sandboxing (WSL2/Lima), a built‑in “Skills” system for generating PPTX/DOCX/PDF/XLSX files, MCP connectors for browsers/Notion, and remote control via Feishu or Slack, it lets non‑technical users give natural‑language commands that read/write files, automate GUI apps, and produce professional documents—all running locally with optional VM isolation.

  • Project
    Devin-AXIS/iPolloWork

    iPolloWork is a local‑first, Electron‑based workbench that unifies multiple AI‑agent runtimes (OpenCode, Codex, DeepSeek Harness) into one collaborative UI for teams. It lets you manage agents, plugins, tasks and editable outputs (code, docs, designs, video) from a single workspace, with optional cloud integration for identity and hosted workers. Installers are provided for macOS, Windows and Linux; the source can be built with Node 22+, pnpm and Bun. The project is released under a source‑available license that permits personal use but requires permission for larger or commercial deployments.

  • Project
    zeenie-ai/OpenCompany

    OpenCompany is an open‑source desktop app that lets you create AI‑driven “employees” – multi‑agent teams that can automate tasks like coding, email handling, social‑media posting, payments, and research. You describe a job, assemble specialist agents on a visual canvas, connect them to tools (Gmail, GitHub, Stripe, browsers, Android emulator, etc.), and the system runs autonomously on your own computer using any LLM provider you supply.

  • Project
    CoWork-OS/CoWork-OS

    CoWork OS is an open‑source cross‑platform desktop that merges coding, terminals, browsers, document creation and AI agents into one workspace. Users plug in any LLM provider (cloud or local) and run governed, multi‑agent workflows via a GUI or the `cowork` CLI. Features include an Everything Workbench for generated files, visual Automation Studio, Inbox messaging, Box‑Brain knowledge indexing, and strong local‑first security. Distributed as an Electron app (MIT‑licensed) with installers for macOS 13+ and Windows.

  • Project
    MindWorkAI/AI-Studio

    MindWork AI Studio is a free cross‑platform desktop app that lets users chat with many LLM providers (cloud or local), use pre‑built assistants, run tool‑enabled RAG, and manage settings centrally for enterprises. It runs on low‑cost hardware, supports voice transcription, and is released under a functional source license that becomes MIT after two years.