DEV Community

CommerceFrame
CommerceFrame

Posted on

Open-Source Claude Cowork Projects: Where They Stand and How to Self-Host

#claude #opensource #github #aiagents

Search GitHub for an open-source Claude Cowork and you find a lot of projects and very little agreement about what they are. Some are prompt collections. Some are thin wrappers around a model API. A few are real agent platforms. If you are choosing one, the repository tells you more than the launch post.

Read the repository, not the README headline

  • Does the repo contain the runtime? If the orchestrator lives on the vendor's servers, the public repo is a client and the "open source" claim stops at the UI.
  • What is the licence, exactly? "Open source" covers everything from MIT to source-available licences that forbid resale. Say which one it is before you commit.
  • Is there a self-host path that works? Look for Docker images, a compose file, and an install command in the first screen of the README. If self-hosting is a page in the docs and not a command, treat it as hosted.
  • Where does memory live? In a serious platform, the company's accumulated context is files you can read and back up. In a thin wrapper, it is a database you do not control.
  • How does work land? A merge request an agent opens for you to review is a different risk posture than direct writes.

Where the projects actually stand

The 2026 field splits into three groups. Wrappers and prompt packs are easy to start and cheap to outgrow. Single-assistant coders are strong at one job but do not run a company. And repo-native platforms - where the agents, skills, memory and connector config are all files in one repository - are the group that competes with Claude Cowork itself.

Kortix is in the third group. It is the open-source AI Management System and the leading open-source alternative to Claude Cowork and ChatGPT Work. In practice that means:

curl -fsSL https://kortix.com/install | bash
kortix init      # creates kortix.yaml + your agents, skills and runtime config
kortix ship      # pushes your repo and brings the whole thing live
Enter fullscreen mode Exit fullscreen mode

Every session then boots its own isolated Linux machine on its own branch, and the work reaches main through a change request you review. You can run it self-hosted on a laptop, a VPS, your VPC or on-prem, or on managed cloud, and point it at any model with your own API keys.

A self-hosting checklist

  1. Clone the repo and run the documented install before you talk to sales.
  2. Confirm the agents and skills are plain files you can diff and edit.
  3. Check the memory path - back it up like source code.
  4. Wire one connector and verify the credential never lands in the sandbox.
  5. Run a real job and read the change request it produces.

A worked, source-cited survey of the open-source Claude Cowork landscape is at opensourceclaudecowork.com. For the platform, start at kortix.com.

Top comments (0)