OpenWork AI: Every Feature Explained + Scored (2026)

OpenWork AI is a free, open-source desktop app 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, it runs on macOS, Windows and Linux, the desktop app is MIT-licensed, and it works with any model you connect.

My short verdict is that it is the strongest open-source Cowork alternative on paper, with a few features that its own makers still mark as partial.

This page is the reference version, so I'll explain every main feature, give each one a score, lay out the pricing and the licence, and compare it with Claude Cowork, Hermes Desktop and my own Agent OS.

I need to be upfront that these are on-paper scores from OpenWork's own repo, docs, roadmap and pricing page as of 10 October 2026, because I have not yet run a full hands-on test of this product.

OpenWork At A Glance

The video above is my OpenCode walkthrough, which I've put first because OpenCode is the engine OpenWork runs on.

What OpenWork Is And Why It Exists

OpenWork gives you a Cowork-style workspace without tying you to one model company.

You authorise a folder, connect a model and describe a job.

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

Ben Shafii's launch post said the original goal was to bring terminal agent workflows into an interface that non-technical people could use.

The README now pitches it as the free, open-source alternative to Claude Cowork and Codex.

How I Scored OpenWork

I score out of 10 on three things for each feature.

The first is what the feature does according to the docs.

The second is how complete OpenWork itself says it is, using the Live, Preview and Partial labels on its public roadmap.

The third is how much of it you get for free and signed out.

A hands-on test could move any of these numbers, so treat them as a buying guide and not a lab result.

🔥 Want my exact OpenWork AI 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

Every OpenWork Feature, Scored

Model Choice: 9 Out Of 10

This is OpenWork's headline feature.

You connect models under Settings › AI Providers.

You can sign in with a paid ChatGPT Plus or Pro plan, add an Anthropic API key, or point it at any OpenAI-compatible endpoint such as OpenRouter or LiteLLM.

For local models, OpenWork has a built-in Ollama screen that pulls a model and adds it to your workspace.

It reads which models support images and which thinking levels they offer.

LM Studio and other local servers connect through a provider entry in the workspace's opencode.jsonc file.

OpenCode Go and Zen models are a separate provider with separate billing, and OpenWork's own managed inference is a third option.

I took one point off because the docs admit that small local models often struggle to call tools.

Skills, Plugins And The Library: 8 And 7

OpenWork's docs define four units, and the definitions are useful.

You create a skill by finishing a job and typing, "Turn what we just did into a reusable skill for me."

When you're signed in, that creates a private skill in your OpenWork Cloud organisation.

When you ask for a local skill, OpenWork writes a SKILL.md file under .opencode/skills/ in your workspace, and that path needs no account.

The docs note that creating a skill never publishes it, and sharing is a separate step.

The Built-In Browser: 8 Out Of 10

The agent can open pages, click, fill forms, read content and take screenshots in OpenWork's own browser.

You allow browser control once per conversation thread.

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

A "Take over" button pauses the agent so you can sign in yourself, and "Resume browser" hands control back.

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

They also say the browser only exists in the desktop app, so cloud workers and OpenWork Web don't have one.

Excel And Artifacts: 7 Out Of 10

OpenWork gives the agent three workbook tools.

One inspects a workbook, one reads a sheet or range, and one writes a new .xlsx file.

Numbers stay numeric, so sums and filters work when you open the result.

Password-protected workbooks, legacy .xls files, .numbers files and .ods files are not read.

Charts, pivot tables and macros are outside the built-in tools.

Finished files show up in the artifact panel, where you can preview and download them.

Memory, Schedules And Sandboxing: 5 Out Of 10 Each

These three are where OpenWork is weakest today, and its own pages say so.

Cross-chat memory is a search over saved sessions that the agent runs when you ask.

It is not a long-term memory store, and nothing is recalled automatically.

Scheduled tasks are marked "Partial" on the roadmap, and Automations need an OpenWork Cloud sign-in.

Sandbox workspaces use Docker or microsandbox, and the roadmap says they have platform and setup limits.

If those three features are why you want an agent desktop, wait or pick something more mature.

🔥 Want to know which AI tools I actually keep using? Inside the AI Profit Boardroom, I share my tested setups and my Agent OS, and 3,400+ members get four coaching calls a week. → Get access here

The MCP Gateway: 8 Out Of 10

This is the feature that makes OpenWork more than a Cowork clone.

You add one OpenWork MCP to an agent you already use, and it carries your skills, plugins and connections with it.

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

It lists setup commands for Codex, Claude Code and OpenCode, and one URL for any other MCP client.

The docs also cover Cursor, VS Code, ChatGPT, Claude Desktop, Windsurf, Zed and Gemini CLI.

It needs an OpenWork sign-in, because your client opens a browser to pick your organisation.

OpenWork Den And Team Features: 7 Out Of 10

OpenWork Den is the control plane for a team or company.

The README says admins use it to provision inference, invite teammates, set desktop policies and publish skills and plugins through marketplaces.

It can also import Agent Plugins and Anthropic-compatible plugins.

Team adds SSO, and Enterprise adds SCIM, an audit log and internal white-labelling.

You can self-host it with Docker Compose or Helm, and the pricing page says the price is the same on cloud and self-hosted.

OpenWork Pricing And Licence

The licence is the part to read carefully.

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

Everything inside ee/ is OpenWork Den, under the source-available OpenWork EE License.

That licence allows free production use for organisations of up to 5 users, free 30-day evaluation at any size, and free development and testing.

Each ee/ release converts to MIT two years after it is published.

OpenWork publishes a stewardship page that promises the MIT core stays MIT.

I score the pricing and licence 8 out of 10, because it is clear and fair, but "open source" does not cover every folder.

How To Install And Set Up OpenWork

1. Download it from openworklabs.com/download, from GitHub releases, or with brew install --cask openwork on a Mac.
2. Pick a folder, because OpenWork only works in folders you authorise.
3. Connect a model under Settings › AI Providers, or add a local one under Settings › Ollama.
4. Run "Summarize this folder" as your first task.

The docs warn against npx openwork, because that npm package is a different project.

No OpenWork account is needed for these four steps.

OpenWork vs Claude Cowork vs Hermes Desktop vs My Agent OS

OpenWork publishes its own comparison with Claude, checked against each vendor's docs.

In that table OpenWork counts 12 full "Yes" answers out of 22 capabilities, with 9 marked partial.

It marks OpenWork as partial for memory across chats, sandboxing, scheduled tasks, run history and desktop policies.

It marks starting work from your phone or Slack as planned, and OpenWork in Slack is a private alpha.

That table is written by OpenWork, so read it as a vendor's view, but I respect that it lists its own gaps.

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

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

My Agent OS runs a crew of agents under one shared memory, so it solves a different problem from a single workspace.

A Note On My January 2026 OpenWork Video

My January 2026 video tested a different app that used the Openwork name and was distributed from accomplish.ai.

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

In that video Claude Cowork was clearly faster on the same tasks, and I said Cowork won.

Those results belong to that other app, so they don't count for or against today's OpenWork.

There is also an npm package called openwork from LangChain, which is unrelated.

🔥 Want my honest take on new AI tools every week? The AI Profit Boardroom has my tool breakdowns, my automation blueprints and weekly coaching calls with 3,400+ members. → Join the AI Profit Boardroom

OpenWork Limits

• The app is free, but the model is not, unless you run a local one.
• It is on version 0.18, and the repo showed about 600 open issues and pull requests when I checked.
• Memory, scheduling and sandboxing are all partial.
• Managed connectors and Library installs need a Cloud sign-in.
• The docs say there is no result cache, so every rerun costs tokens.
• The Homebrew cask can lag one version behind the newest release.
• OpenWork Web is in preview as a paid add-on, according to OpenWork's comparison page.

My Verdict On OpenWork

OpenWork is the open-source Cowork alternative I'd try first.

It suits people who want to choose their own model, people on Linux, and teams that want to self-host.

It does not yet suit people who need automatic memory, reliable automations or a phone app.

Claude Cowork stays ahead on polish, and it is the easier choice if you already pay for Claude.

I'll update these scores when I've run a full hands-on test.

Related Reading

• My Herald OS review covers another free agent desktop.
• The best AI models for coding helps you choose what to connect to OpenWork.
• The best Hermes Agent alternative ranks OpenWork-style tools against Hermes.
• The best Hermes Agent UI compares the interfaces I use.

Also On Our Network

• 🌐 Install OpenWork and run your first task, step by step
• 🌐 Seven real jobs a business can give OpenWork
• 🌐 The OpenWork quick-start on agentos.guide
• 🌐 How the models you'd run in OpenWork score on GoldieBench

FAQ: OpenWork AI

What is OpenWork AI?

OpenWork is a free, open-source desktop app where AI agents work on your own files.

It is built on OpenCode by Different AI and positioned as an open-source alternative to Claude Cowork.

Is OpenWork really open source?

The desktop app and core platform are MIT.

The team control plane in the ee/ folder is source-available, and it is free for organisations of up to 5 users.

How much does OpenWork cost?

The desktop app is free, and Team costs $10 per seat per month.

Model usage through your own provider is billed by that provider.

Is OpenWork better than Claude Cowork?

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

Claude Cowork is more finished, and it needs a paid Claude plan.

Does OpenWork have memory?

The agent can search and read past sessions when you ask.

The docs say it is not a separate long-term memory database.

Are there other tools called OpenWork?

There are, including a renamed app once distributed from accomplish.ai and an npm package from LangChain.

This page covers OpenWork from Different AI at openworklabs.com.

📺 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

Download the free app, test it on one folder with one model, and decide for yourself whether my on-paper 7.5 is fair for OpenWork AI.