OpenWork AI is a free, open-source desktop app for macOS, Windows and Linux where AI agents do real work on your own files, and it is built on OpenCode as an alternative to Claude Cowork.
It is made by Different AI, Inc., a Y Combinator company founded by Ben Shafii, and the desktop app is MIT-licensed, so you can use it locally without paying for it or creating an account.
You bring the model yourself, which means a ChatGPT sign-in, your own API key, or a local model through Ollama.
This page is the operator walkthrough, so I'll cover what OpenWork is, what it costs, how to install it, how to set up every main feature, where its limits are, and how it compares with Claude Cowork, Hermes Desktop and my own Agent OS.
I checked every fact here against the project's own GitHub repo, docs and pricing page on 10 October 2026.
What OpenWork Is In One Minute
OpenWork is a desktop workspace where you pick a folder, pick a model and hand an AI agent a job.
The agent reads your files, runs tools, drives a built-in browser and hands back finished work such as documents and spreadsheets.
It runs on top of OpenCode, the open-source coding agent, and the README says OpenWork works with 50+ providers.
The video above is my OpenCode walkthrough, because OpenCode is the engine underneath OpenWork and knowing it makes OpenWork much easier to understand.
| Question | Answer |
|---|---|
| Who makes it? | Different AI, Inc. makes it, and its founder and CEO is Ben Shafii. |
| Where is the code? | The code lives at github.com/different-ai/openwork, which had about 24,000 stars when I checked. |
| What does it cost? | The desktop app is free and MIT-licensed, and no OpenWork account is needed locally. |
| What do you pay for? | You pay for the model you connect, unless you use a local model. |
| Where does it run? | It runs on macOS, Windows and Linux as a desktop app. |
| What is the latest release? | Version 0.18.57 was published on 7 October 2026. |
| What is it built on? | It is built on OpenCode, and the desktop shell is an Electron app. |
Which OpenWork This Page Covers
Several projects share this name, so I want to be clear before you install anything.
This page covers OpenWork from Different AI, which lives at openworklabs.com and at different-ai/openwork on GitHub.
That project launched on Hacker News on 14 January 2026 as "an open-source alternative to Claude Cowork".
There was a second app called Openwork around the same time, distributed from accomplish.ai, and that is the one I tested on camera in January 2026.
That second project was later renamed, and its GitHub repo now says the project is no longer supported.
There is also an npm package called openwork from LangChain, which is an interface for its deepagents library and is not this app.
OpenWork's own docs say the same thing and tell you not to run npx openwork.
So I haven't done a full on-camera test of today's OpenWork yet, and everything below comes from its repo, its docs and its pricing page.
🔥 Want my exact OpenWork AI Setup setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
What Problem OpenWork Solves
Agent tools like OpenCode are powerful, but they live in a terminal.
Ben Shafii wrote in the launch post that he built OpenWork so his wife and he could run agent workflows without living in a terminal.
Claude Cowork solves the same problem for Claude users, but it needs a paid Claude plan and it is built around Anthropic's models.
OpenWork gives you the same kind of workspace with any model, on your own machine, with the source code open.
The README lists the reasons plainly.
- The desktop app is free and open source.
- You can use any model, including your own API key, a ChatGPT sign-in or a local model.
- It runs on macOS, Windows and Linux as an app rather than a command-line tool.
- Your files stay local, and the cloud is optional.
- Teams can share skills and MCP servers.
- You can self-host the team control plane.
- Skills, Claude-compatible plugins and MCP servers carry over from Cowork.
How To Install OpenWork
You have four ways to install it, and all four come from the project's own pages.
Option one is the download page.
Go to openworklabs.com/download and pick your system.
The page offers macOS builds for Apple Silicon and Intel, Windows installers for x64 and ARM64, and Linux AppImage and tar.gz builds for x64 and ARM64.
Option two is Homebrew on a Mac.
brew install --cask openwork
The Homebrew cask needs macOS 12 or newer, and it can sit one version behind the newest release.
Option three is GitHub releases.
Open the releases page of different-ai/openwork and download the file for your system.
Option four is letting your existing agent install it.
The README gives a prompt you can paste into Claude Code, Cursor or Codex, which points the agent at OpenWork's own start guide.
I'd read that guide yourself before you let any agent run install commands for you.
Set Up OpenWork In Four Steps
OpenWork's docs give the first run as four steps.
- Install the app from one of the options above.
- Pick a folder, because OpenWork only works in folders you authorise.
- Choose a model.
- Run a task, and the docs suggest starting with "Summarize this folder."
Start with a small folder that you've backed up.
That first summary proves the agent can read the files you expect before you let it change anything.
You do not need to sign in to OpenWork for any of this.
How To Connect A Model
The model is the part that costs money or doesn't, so pick it on purpose.
| Route | Where you set it up | What it costs |
|---|---|---|
| ChatGPT sign-in | Settings › AI Providers › Connect provider › OpenAI | It uses your paid ChatGPT Plus or Pro plan. |
| Anthropic API key | Settings › AI Providers | You pay Anthropic's API prices. |
| Custom or gateway | An OpenAI-compatible endpoint such as OpenRouter or LiteLLM | You pay that provider. |
| OpenCode Go or Zen | Settings › AI Providers › search "opencode" | OpenCode bills you separately. |
| Ollama | Settings › Ollama | It is free and local. |
| LM Studio or another local server | The workspace's opencode.jsonc file |
It is free and local. |
To use ChatGPT, open Settings › AI Providers, click Connect provider, choose OpenAI and then choose the ChatGPT subscription option.
A browser tab opens, you sign in with the email that holds your subscription, and the dialog closes by itself.
To use Ollama, install Ollama, open Settings in OpenWork and choose Ollama.
Wait for the status to say "Ollama running", select a model you've already downloaded or click "Pull a model", and then click "Add to workspace".
There is also a "Sync all models" button that matches OpenWork to everything in Ollama at once.
The docs say local models need no API key, no OpenWork Cloud account and no internet once the model is downloaded.
They also give an honest warning that small local models often chat well but struggle to call tools reliably.
Their advice is to keep a cloud model connected as a fallback for long, multi-step tasks, and I agree with it.
Three Checks To Run After Setup
I run the same three checks on every new agent desktop.
- Ask it to summarise the folder, and confirm it names real files.
- Ask it to create one small file, and confirm the file appears where you expect.
- Open the model picker under the message box, and confirm the model name is the one you meant to use.
If all three pass, the workspace, the permissions and the model are wired correctly.
🔥 Want help setting up your own agent desktop? Inside the AI Profit Boardroom, I share my agent trainings and my Agent OS, and 3,400+ members bring their setups to four coaching calls a week. → Get access here
How To Use Every Main OpenWork Feature
Skills
A skill is a set of written instructions the agent loads when a task matches.
OpenWork uses the same SKILL.md format as Claude.
The simplest way to make one is to finish a task in chat and then type, "Turn what we just did into a reusable skill for me."
If you're signed out, ask for a local skill, and OpenWork writes a SKILL.md file inside your workspace under .opencode/skills/.
Local skills are free and need no Cloud account.
Connectors, Plugins And The Library
OpenWork's docs split things into four units.
A connector is an MCP service, a skill is the playbook, a plugin is a bundle of skills and MCP servers, and the Library is the list of what you have.
You find all of it under Settings › Library.
Custom MCP Servers
You can add your own MCP server without signing in.
Open Settings › Library, expand Advanced settings and click "Add workspace MCP".
Enter an app name and a server URL, then click "Add App".
Managed connections such as Gmail, Google Calendar, Google Drive, Slack, Notion and Linear need an OpenWork Cloud sign-in and an organisation that has enabled them.
The Built-In Browser
OpenWork has its own browser that the agent can open pages in, click, fill forms, read content and take screenshots.
You allow browser control once per thread.
Clicks, typing and keyboard input still ask "Allow browser action?" each time, because they can change data on a site you're signed in to.
You can press "Take over" at any point to pause the agent and sign in yourself.
The docs say browser control does not give the agent access to your regular Chrome profile.
Cross-Chat Memory
OpenWork can answer questions about past sessions by searching your saved session history.
The docs are clear that this is not a separate long-term memory database, and it does not inject old chats into new prompts automatically.
You ask for it, with a prompt like "Find the session about the pricing page redesign and remind me what we decided."
Excel Workbooks
You can drop a .xlsx file into the chat, and the agent gets a compact preview of every sheet.
It can inspect a workbook, read any sheet or range, and write a new workbook back.
The built-in tools don't do charts, pivot tables or macros, and they don't read legacy .xls files.
Scheduled Tasks And Automations
OpenWork can run a prompt on a schedule.
Its roadmap marks scheduled tasks as "Partial", and the docs list Automations as a feature that needs an OpenWork Cloud sign-in.
The OpenWork MCP Gateway
This is the feature I find most interesting as an operator.
You can add one OpenWork MCP to Claude Code, Codex, Cursor or OpenCode and reuse the same skills and connections there.
claude mcp add --transport http openwork https://api.openworklabs.com/mcp/agent
Your client then opens a browser so you can sign in and choose your OpenWork organisation.
OpenWork Den For Teams
OpenWork Den is the control plane for teams.
Admins use it to invite people, share model providers, publish skills and plugins, and set desktop policies.
What Works Signed Out And What Needs A Sign-In
OpenWork's docs publish this split, and it is worth knowing before you plan around a feature.
| Works signed out | Needs an OpenWork Cloud sign-in |
|---|---|
| Chatting with files in folders you authorise | Connections your organisation shares, such as Gmail or Slack |
| Your own models, API keys, ChatGPT sign-in and local models | OpenWork inference models |
| Local workspace skills | Adding skills and plugins from the Library, and sharing your own |
| Custom MCP servers in a workspace | Automations |
| Dashboards and saved apps | |
| Cloud workspaces that keep tasks running while your computer sleeps |
What OpenWork Costs
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | You get the MIT desktop app, your own keys, the first 5 Cloud seats and self-hosting for up to 5 users. |
| Team | $10 per seat per month | You get unlimited users, SSO, the Extension Marketplace, shared keys and cloud automations. |
| Enterprise | Custom | You get SCIM, desktop policies, an audit log and internal white-labelling. |
There are two add-ons on the paid plans.
OpenWork Cloud Computer costs $50 per member per month, and OpenWork models cost $10 per user per month.
The licence is split by folder.
Everything outside the ee/ folder is MIT, and the ee/ folder, which holds OpenWork Den, uses a source-available licence.
OpenWork vs Claude Cowork vs Hermes Desktop vs My Agent OS
Here is how I'd place the four as an operator.
| What matters | OpenWork | Claude Cowork | Hermes Desktop | My Agent OS |
|---|---|---|---|---|
| Who makes it | Different AI | Anthropic | Nous Research | Me, for AI Profit Boardroom members |
| Cost of the app | Free, MIT desktop app | Needs a paid Claude plan | Free | Included with the AI Profit Boardroom |
| Models | Any provider or local | Built around Claude | The model you choose for Hermes | A crew of agents, including Claude Code, Hermes, Codex and Kimi |
| Platforms | macOS, Windows, Linux | Desktop, web and mobile | Desktop | macOS, Linux and web |
| Public source code | Yes | No | Hermes Agent is open source | No public repo |
| Best for | Cowork-style work on any model | People already paying for Claude | Hermes users who want a stable app | Running several agents under one shared memory |
Claude Cowork is the more finished product.
OpenWork's own comparison table marks OpenWork as "Partial" for memory across chats, sandboxing, automations and run history, and it marks starting work from your phone or Slack as planned.
Anthropic's help page says Cowork is available on paid plans and now runs on desktop, web and mobile.
Hermes Desktop is the official app for Hermes Agent, and it has a one-click plugin marketplace.
My Agent OS is a different kind of tool, because it is one dashboard running a crew of agents under one shared memory.
OpenWork could sit beside any of them as the place you do file work with a model you choose.
🔥 Want the Agent OS and the trainings behind it? The AI Profit Boardroom has my step-by-step agent courses, my automation blueprints and weekly coaching calls with 3,400+ members. → Join the AI Profit Boardroom
OpenWork Limits Before You Rely On It
- OpenWork is still on version 0.18, and the repo showed about 600 open issues and pull requests when I checked.
- The free app does not include a model, so "free" means the software and not the tokens.
- Small local models may stall on tool-heavy tasks.
- Memory across chats works only when you ask the agent to look something up.
- Scheduled tasks and sandbox workspaces are both marked "Partial" on the roadmap.
- Managed connections such as Gmail and Slack need a Cloud sign-in and an organisation.
- Conversation history does not transfer when you migrate from Cowork, and credentials do not move either.
- Every run costs tokens, because the docs say there is no result cache today.
My Verdict For Operators
OpenWork is the most complete open-source answer to Claude Cowork that I've looked at.
It has an active repo, a clear licence, honest docs and a real choice of models.
I'd use it if I wanted Cowork-style file work without being tied to one model provider.
I'd stay on Claude Cowork if I already paid for Claude and wanted the most polished experience today.
Install it on a test folder first, connect one model, and run the three checks before you trust it with client files.
Related Reading
- Herald OS is another free agent desktop I've covered, built around Hermes Agent.
- How to set up Hermes Agent in one click covers the fastest Hermes install.
- The best Hermes Agent alternative shows where tools like OpenWork fit.
- The best Hermes Agent UI ranks the interfaces I use.
Also On Our Network
- 🌐 How a business or agency would actually use OpenWork
- 🌐 Every OpenWork feature explained and scored
- 🌐 An OpenWork quick-start for your Agent OS
- 🌐 The best models to run inside OpenWork, with real benchmark scores
FAQ: OpenWork AI
What is OpenWork AI?
OpenWork is a free, open-source desktop app for macOS, Windows and Linux where AI agents work on your own files.
It is built on OpenCode by Different AI and is positioned as an open-source alternative to Claude Cowork.
Is OpenWork AI free?
The desktop app is free and MIT-licensed, and you don't need an account to use it locally.
You still pay for the model you connect unless you run a local one.
How do I install OpenWork?
Download it from openworklabs.com/download or from the GitHub releases page, or run brew install --cask openwork on a Mac.
Don't run npx openwork, because that npm package is a different project.
Can OpenWork run local models?
Yes, it has a built-in Ollama screen, and it connects to LM Studio through the workspace's opencode.jsonc file.
The docs warn that small local models often struggle to call tools reliably.
Who makes OpenWork?
Different AI, Inc. makes it, and its founder and CEO is Ben Shafii.
The company is listed in Y Combinator's Spring 2026 batch.
Is this the same OpenWork you tested in January 2026?
It isn't, because the app I tested then came from accomplish.ai and only shared the name.
That project was later renamed, and its repo now says it is no longer supported.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I run Goldie Agency, a 7-figure SEO and link building agency based in Bangkok.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery", which is available on Amazon.
- I run GoldieBench, where I score AI models on real one-shot builds.
→ Get my best AI training inside the AI Profit Boardroom
Pick a test folder, connect one model, run the three checks, and you'll know within ten minutes whether OpenWork AI belongs in your stack.
