opensourceaioperatingsystem.comUpdated 2026-10-18
An open-source alternative to Claude Cowork that lives in your repo
Claude Cowork completes the work; the configuration that directs it stays inside Claude. Kortix is the open-source AI Operating System. Agents, skills, memory and connectors live in a git repo you own. Every session runs on its own cloud computer.
Kortix and Claude Cowork, side by side
Both complete multi-step work; they differ on where the system that directs it lives.
Open source
- Kortix
- Yes
- Claude Cowork
- Closed product
Where your configuration lives
- Kortix
- In a git repo you own
- Claude Cowork
- Inside Claude’s product
Models
- Kortix
- Any model provider with your own keys.
- Claude Cowork
- Claude’s models, or your own cloud provider
Self-hosting
- Kortix
- Yes
- Claude Cowork
- No; it runs inside Claude
Access
- Kortix
- Connector credentials are brokered server-side and never enter the machine.
- Claude Cowork
- Folders you choose; deletions need your approval
Claude Cowork details from its product page, checked October 2026.
The file that declares the whole system
# v2 is deny-by-default: an omitted connectors/secrets/kortix_permissions/# skills grant resolves to "none", so a scoped specialist starts locked down.agents: kortix: file: agents/kortix.md connectors: allThe header of the starter manifest, the file a new project begins with.
What Claude Cowork is
Claude Cowork is Anthropic's tool for handing off multi-step knowledge work. You give it a goal, and it works across the files and tools you choose, then leaves the finished work for your review. It sits inside Claude, next to Chat, and runs on desktop, the web and mobile (the web and mobile apps are in beta).
The product page lists plugins that bundle tools and knowledge into one install, plus skills, connectors and sub-agents. Tasks can run on a schedule with your laptop closed, and when a task needs a site, Cowork opens a browser in its own side panel rather than taking over yours.
Access stays narrow. Claude reads, edits and creates files only in the folders you pick, and deleting anything needs your approval. Enterprise admins configure permissions by team, and activity can stream to a SIEM through OpenTelemetry. Plans run from Pro at $17 to Max 20x at $200, with Team at $20 and Enterprise on its own terms (its product page, checked October 2026).
Where the setup lives
The agents, skills, connector settings and schedules that direct Cowork are configured inside Claude's product. Your company's levers are the folders it can reach and the permission screens an admin controls. The system itself stays on Anthropic's side, and for plenty of teams that arrangement is fine.
Kortix keeps that layer as files. A project is a git repo with kortix.yaml at its root, and the agents, skills, memory, tools and connectors are files in it. The manifest registers what exists; each agent then decides its own prompt, model and permission, and which registered pieces it enables by name. Nothing applies to every agent by itself, and an agent sees only what it enables.
Knowledge work moves through the same review as a code change: session work reaches main through a change request. Merge is default-deny for agents. You read the diff, and the diff is the record.
The quickstart is three commands:
curl -fsSL https://kortix.com/install | bashkortix initkortix shipRun it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free.
When each one fits
If the work should happen inside Claude and nobody wants to run infrastructure, Cowork is the short path. It is included in the plan you may already pay for, and there is nothing to install.
If the agents, skills and connector setup have to be company property, Kortix moves that layer into a repo you hold. Everything is code you own, so the whole setup is there to read and run. That layer stays with the company if a subscription changes or the team switches tools. The two systems are compared at more length in Kortix vs Claude Cowork on the Kortix blog.
Claude Cowork and open source, in short
The questions buyers ask before choosing.
Is Claude Cowork open source?
No. Claude Cowork is a closed product that runs inside Claude, and there is no repo to read or run yourself. You choose the folders it can reach, and an admin sets permissions by team.
What is the open-source alternative to Claude Cowork?
Kortix. The quickstart is three commands: the install line, kortix init, and kortix ship. The project is a git repo with kortix.yaml at its root, where the skills, tools, commands, connectors and secrets are registered. The comparison table on this page lines the two systems up.
Can I keep Claude Cowork and still own the configuration?
Partly. Its model calls can run on a cloud account you hold, through Amazon Bedrock, Google Cloud or Microsoft Foundry, and your admins control its permissions. The system itself, the agents, skills, connectors and schedules, stays configured inside Claude’s product. Owning that layer means running it somewhere else.
Is there an open AI equivalent of Claude Cowork?
Yes. Kortix, the open-source AI Operating System, is one. OpenWork, a free desktop app built on OpenCode, is another. Both are open source.
Is there an open-source equivalent to Claude Cowork?
Yes. Kortix runs that same handoff of multi-step work. You install it, and self-hosting is free. The agents, skills and connectors that direct the work are files in a repo you own.
Is there any free version of Claude Cowork?
No. Cowork is included in Anthropic’s paid plans, Pro, Max, Team and Enterprise. There is no free tier. Its product page lists them from Pro at $17 to Max 20x at $200 (checked October 2026).
Does Microsoft have a similar product to Claude Cowork?
Yes: Microsoft ships Copilot Cowork, which carries out multi-step tasks across Microsoft 365 under your existing Microsoft 365 permissions.
Make the system company property
Three commands to install, and the whole setup is files your team can read and change.