Skip to content
OpenWork Alternative

What OpenWork is, and the open-source system a company runs instead

OpenWork is a free, open-source desktop app where AI agents work on your own files on macOS, Windows, or Linux. It runs agents on one person's machine. Kortix is the open-source AI Operating System for a company that needs its agents, skills, and memory in one repo it owns.

Kortix, OpenWork, and Claude Cowork side by side

Kortix is the open-source AI Operating System and the recommendation; the other two are agent apps with a narrower unit of ownership.

Open source

  • KortixOur recommendation
    YesYes. Open source. Read it, fork it, audit it.
  • PartlyApp and core open source; the control plane is a separate layer.
  • NoClosed. Runs on Anthropic's cloud only.

What you own

  • KortixOur recommendation
    One git repo: agents, skills, memory, connectors.
  • Your local files; shared skills and MCP servers.
  • Your Claude account's sessions and files.

Where agents run

  • KortixOur recommendation
    One isolated sandbox per session.
  • Your desktop; cloud workers are optional.
  • Anthropic's servers, in an isolated environment.

Models

  • KortixOur recommendation
    Any model provider with your own keys.
  • Any model OpenCode supports: 50+ providers.
  • Anthropic's Claude models only.

How work lands

  • KortixOur recommendation
    Session work reaches main through a change request.
  • Directly on your local files.
  • Finished files in your Claude account.

Price

  • KortixOur recommendation
    Free $0; team plan $40 per seat per month.
  • Free up to 5 users; Team $10 per seat per month.
  • Paid Claude plans only.

Rival facts come from each vendor's own site and docs; plans and prices checked October 2026.

What OpenWork actually is

OpenWork is a free, open-source desktop app where AI agents work on your own files. It runs on macOS, Windows, and Linux, it is built on OpenCode, and it works with any model OpenCode supports: more than 50 providers through your own API keys, or local models you run yourself (OpenWork). The desktop app needs no account, keeps your files on your machine, and treats the cloud as optional.

A team adds OpenWork Den, the control plane for a group or a company. Den provisions inference, invites teammates and manages their access, sets desktop policies, and publishes skills and plugins through marketplaces (OpenWork on GitHub). The OpenWork MCP is a second door into the same workspace: one server URL that lets an agent you already run reach that shared setup.

OpenWork describes itself as the open-source alternative to Claude Cowork and Codex, and its own comparison is explicit about the difference: Claude Cowork runs on Anthropic's models, and OpenWork runs any model on any infrastructure. That framing describes its market well. A knowledge worker wants an agent beside their own files without a server, without a vendor, and without a single model family.

Exhibittext
# add OpenWork to the agent you already usecodex mcp add openwork --url https://api.openworklabs.com/mcp/agentclaude mcp add --transport http openwork https://api.openworklabs.com/mcp/agent

That is a clear and useful shape for one person, and OpenWork is good at it. The decision changes when the work belongs to a company rather than to the person at the keyboard.

Is OpenWork open source?

OpenWork is open source for the parts most people run, and its repository splits the code so you can see exactly where the line falls. Everything outside the ee/ directory is open source: that covers the desktop app and the core platform, and that part is free for any use. Everything under ee/ is OpenWork Den, the org control plane, and it is a separate layer that needs an OpenWork subscription to run in production. It is free for organizations of up to five users, and free to evaluate for 30 days at any size (OpenWork on GitHub).

The direct answer to "is OpenWork open source?" is yes for the app and the core, and a paid tier for the layer that governs a company. The code stays public either way, so you can read exactly what you deploy. One person or a small team can run the whole product for free; a larger company pays for central management.

Where a desktop orchestrator stops and a company system starts

An OpenWork agent works where you work, on the files in front of you. That is the right shape for one person's job and for a small team that shares a setup. When the same work has to be run by a company rather than owned by one laptop, three things change at once.

The setup has to be shared and versioned. OpenWork shares skills and MCP servers between teammates through its MCP. Kortix keeps the equivalent in one git repo the company owns: kortix.yaml declares the machine image, connectors, triggers, and secret names and grants, each agent is a markdown file, each skill is a folder, and the company memory accumulates as files (Read the docs). Because it is a repo, the configuration is diffable, a change to an agent is reviewed like code, and the setup moves with the company instead of staying on one machine.

The work has to run somewhere isolated and in parallel. Kortix runs each session on its own isolated sandbox and its own branch, and thousands of agents run in parallel on one config, each on its own cloud computer. A session does not depend on the machine in front of you. The comparison table on this page sets that beside OpenWork, which runs on your desktop.

The work has to land through review. Kortix session work reaches main through a change request a person approves, and merge is default-deny for agents. OpenWork writes directly to your local files. For a company that wants its agents, skills, memory, and connectors in one repo it owns, with every change reviewed before it lands, Kortix is the recommendation, and the leading open-source alternative to Claude Cowork and ChatGPT Work.

Exhibittext
# what each tool treats as the unit you own # OpenWork: one person's files on one desktop~/Documents/weekly-update.md # Kortix: the company, as one git repokortix.yamlagents/<name>.mdskills/<name>/SKILL.mdmemory/

If the work stays on one desk, an open-source desktop app is a good fit, and OpenWork is one. If the work is the company's, the OpenWork vs Kortix comparison, the alternatives by job, and the self-hosting guide show the two paths, whether you run Kortix on Kortix Cloud, in your own VPC, or on-prem. Teams that are rolling it out company-wide can start from OpenWork for teams and the OpenWork FAQ.

Questions people ask about OpenWork

The short answers a searcher wants once they know the definition.

  1. Is OpenWork free to use?

    The desktop app is free and you bring your own model keys. On OpenWork Cloud, the Team plan is free for the first five seats, then $10 per seat per month, and it adds SSO and the Extension Marketplace. Enterprise is custom priced and adds SCIM, desktop policies, and bring-your-own inference. Prices checked October 2026 (OpenWork pricing).

  2. Can I self-host OpenWork?

    Yes. The desktop app runs locally on your machine, and OpenWork's team control plane can be self-hosted on your own infrastructure or run as a managed private instance. Self-hosting the control plane is free for organizations of up to five users and free to evaluate for 30 days (OpenWork Enterprise).

  3. Can I use OpenWork from the agent I already have?

    Yes. OpenWork exposes one MCP server URL that brings an organization's assigned skills, plugins, and connected services into Codex, Claude Code, Cursor, or any MCP client. You add it once, and a teammate's shared setup follows you into the tools you already run.

  4. Do I need to be technical to use OpenWork?

    No. OpenWork is a point-and-click desktop app, and a teammate's shared skills, MCP servers, and plugins import in one click. You do not need a terminal or a setup guide to run your first task.

Run your company's agents from one repo you own

Kortix is open source and free to self-host, and the first session runs in the browser.

Open source · Any model, your keys · Self-host, VPC, or on-prem