agentos.guide › blog

OpenWork AI Quick-Start: What It Is, How To Set It Up, And Where It Fits Beside Your Agent OS

By Julian Goldie · 10 October 2026 · agentos.guide

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.

Different AI, Inc. makes it, the desktop app is MIT-licensed, and you can use it locally without an account.

You bring the model, which can be a ChatGPT sign-in, your own API key or a local model through Ollama.

Here is the four-step quick-start, every main feature, what it costs, its limits, and the lane I'd give it next to an Agent OS.

What OpenWork is, in one minute

OpenWork is an app where you pick a folder, pick a model and hand an AI agent a job.

The agent reads your files, runs tools, calls MCP servers and drives a built-in browser.

The README calls it "the free, open-source alternative to Claude Cowork and Codex".

It runs on OpenCode, which is why I've put my OpenCode walkthrough at the top of this page.

Different AI is listed in Y Combinator's Spring 2026 batch, and its site is openworklabs.com.

OpenWork first launched on Hacker News on 14 January 2026.

Which OpenWork this guide covers

More than one project uses this name, so check before you install.

This guide covers OpenWork from Different AI at openworklabs.com and different-ai/openwork on GitHub.

In January 2026 I tested a different app called Openwork, which was distributed from accomplish.ai.

That project was later renamed, and its GitHub repo now says it is no longer supported.

There is also an npm package called openwork from LangChain, and OpenWork's docs tell you not to run npx openwork for that reason.

I haven't run a full on-camera test of today's OpenWork yet, so this guide is built from its repo, docs and pricing page.

The problem OpenWork solves

OpenCode is a strong agent, but it lives in a terminal.

Ben Shafii wrote in his launch post that he built OpenWork to bring terminal agent workflows into an interface that non-technical people could use.

Claude Cowork does that job for Claude users, but it needs a paid Claude plan.

OpenWork does it with any model, with the source code public, and with your files kept on your machine.

OpenWork quick-start in four steps

OpenWork's docs give the first run as four steps, and none of them needs an account.

• First, install it from openworklabs.com/download, which has builds for macOS, Windows and Linux.
• Second, pick a folder, because OpenWork only works in folders you authorise.
• Third, choose a model under Settings › AI Providers.
• Fourth, run a task, and the docs suggest "Summarize this folder."

On a Mac you can install it with one command instead.

brew install --cask openwork

The Homebrew cask needs macOS 12 or newer, and it can sit one version behind the newest release.

You can also download any release from the GitHub releases page.

The README even gives a prompt you can paste into an agent you already use, so the agent installs OpenWork for you.

I'd read OpenWork's start guide yourself before you let any agent run install commands.

Start with a small folder you've backed up, so the first job is read-only.

How to connect a model to OpenWork

The app is free, so the model is where the cost sits.

For ChatGPT, you choose the ChatGPT subscription option, sign in on the browser tab that opens, and return to the app.

For Ollama, you open Settings › Ollama, wait for "Ollama running", pick or pull a model, and click "Add to workspace".

A "Sync all models" button adds every model Ollama has in one go.

The docs say a local model needs no API key, no OpenWork Cloud account and no internet once it has downloaded.

They also warn 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 tasks, and I'd follow it.

🔥 Want the exact open-source agent desktops beside the Agent OS setup I run inside my Agent OS?

Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.

→ Get access here

Every main OpenWork feature

A few of these need a closer look.

OpenWork uses the same SKILL.md format as Claude, so skills you already have are not wasted.

To make a new skill, finish a job in chat and type, "Turn what we just did into a reusable skill for me."

If you ask for a local skill, OpenWork writes the file under .opencode/skills/ in your workspace, and that needs no account.

The built-in browser asks once per thread for permission to navigate and read.

Clicks, typing and keyboard input still ask each time, because they can change data on a site you're signed in to.

You can press "Take over" to pause the agent and sign in yourself.

The docs say the browser does not get access to your regular Chrome profile.

For spreadsheets, the built-in tools read and write .xlsx files, but they don't make charts, pivot tables or macros.

Cross-chat memory is a search over your saved sessions, and the docs say it is not a separate long-term memory database.

The OpenWork MCP Gateway

This is the feature that matters most if you already run agents.

You add one OpenWork MCP to Claude Code, Codex, Cursor or OpenCode, and your skills and connections travel with it.

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.

The README says the gateway exposes four tools, which find a capability, run it, list your skills and read one skill.

So the desktop app is optional, because you can use OpenWork from the agent you already have.

What OpenWork costs

There are two paid add-ons, which are OpenWork Cloud Computer at $50 per member per month and OpenWork models at $10 per user per month.

The licence is split by folder.

Everything outside the ee/ folder is MIT, which covers the desktop app and core platform.

The ee/ folder holds OpenWork Den, the team control plane, under a source-available licence.

That licence is free for organisations of up to 5 users, and each release converts to MIT two years after it is published.

🔥 Want the exact open-source agent desktops beside the Agent OS setup I run inside my Agent OS?

Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.

→ Get access here

How OpenWork fits an Agent OS workflow

I don't replace my whole stack every time a new agent desktop appears.

I give each tool one lane.

The Agent OS is one dashboard where my agents share one memory and keep working while I'm away.

OpenWork is a workspace for a single job in a single folder.

Those two things sit side by side quite happily.

The bridge between them is the skill file.

Because OpenWork uses SKILL.md, a process you write once can be reused by other agents that read the same format.

The second bridge is the MCP Gateway, which puts your OpenWork skills inside Claude Code and Codex.

The third bridge is the model, because OpenWork can use the same providers and local models you already run elsewhere.

OpenWork vs Claude Cowork vs Hermes Desktop vs my Agent OS

Claude Cowork is the more finished product today.

Anthropic's help page says Cowork is available on paid plans across desktop, web and mobile, with scheduled tasks and projects.

OpenWork publishes its own comparison, and it marks OpenWork as partial for memory across chats, sandboxing, automations and run history.

It also lists starting work from your phone or Slack as planned.

OpenWork is ahead on model choice, on Linux support, on price and on self-hosting.

Hermes Desktop is the official app for Hermes Agent, and it has a one-click plugin marketplace.

My Agent OS is not on public GitHub, and it is only available inside the AI Profit Boardroom.

OpenWork limits to know first

• The free app does not include a model, so you still pay for tokens unless you go local.
• OpenWork is on version 0.18, and the repo showed about 600 open issues and pull requests when I checked.
• Scheduled tasks and sandbox workspaces are both marked Partial on the roadmap.
• Memory across chats only works when you ask the agent to look something up.
• Managed connectors, Library installs, automations and the MCP Gateway need an OpenWork Cloud sign-in.
• The docs say there is no result cache, so rerunning a task costs tokens again.
• If you migrate from Claude Cowork, conversation history and credentials do not come with you.

Three tips before you rely on OpenWork

• Run "Summarize this folder" first, and confirm the agent names real files.
• Check the model name under the message box before a long job, so you know what you're paying for.
• Turn the first job that works into a skill, because OpenWork's docs say a good skill makes repeat work shorter and cheaper.

FAQ

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. Different AI, Inc. builds it on top of OpenCode and positions it as an open-source alternative to Claude Cowork. It works with 50+ model providers, your own API keys, a ChatGPT sign-in or local models through Ollama.

Is OpenWork AI free?

The desktop app is free and MIT-licensed, and no OpenWork account is needed to use it locally. You still pay for the model you connect unless you run a local model. The Team plan costs $10 per seat per month, and the first 5 Cloud seats are free.

How do I install OpenWork?

Download it from openworklabs.com/download or the GitHub releases page of different-ai/openwork, or run brew install --cask openwork on a Mac. Then pick a folder, choose a model and run a first task. Do not run npx openwork, because that npm package is a different project.

Can OpenWork replace my Agent OS?

Not really, because they do different jobs. OpenWork is a single workspace for file and browser work with one model at a time. An Agent OS runs several agents under one shared memory. OpenWork works well beside an Agent OS as the lane for one-off file work.

Does OpenWork work with Claude Code and Codex?

Yes. The OpenWork MCP Gateway is one remote MCP server that brings your OpenWork skills, plugins and connections into Claude Code, Codex, Cursor, OpenCode and other MCP clients. It needs an OpenWork sign-in.

Is this the OpenWork Julian tested in January 2026?

No. That video tested a different app that used the Openwork name and was distributed from accomplish.ai. That project was later renamed and its repo now says it is no longer supported. This guide covers OpenWork from Different AI at openworklabs.com.

Final word on OpenWork

OpenWork is the open-source Cowork alternative I'd try first, as long as you go in knowing which features are still partial.

Install it on a test folder, connect one model, and give it one read-only job today.

Keep your long-running agents and your shared memory in your Agent OS, and give the one-off file work to OpenWork AI.

Related guides on agentos.guide

→ How to Build Your Own AI Agent OS

→ OpenCode × Hermes: The Free AI Dev Team On A Kanban Board

→ What Is Herald OS? Free Hermes Agent OS Quick-Start

→ The Hermes Command Desk: One AI Employee In Every Room Of The Agent OS

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

← all guides on agentos.guide