Claude Cowork vs ChatGPT Work: the open-source option you own

The axes that outlast a release cycle

Claude Cowork is Anthropic’s agent for knowledge work, and ChatGPT Work is OpenAI’s agent for finished work across a team’s tools. Both are closed products from a single vendor, and both change constantly, so a feature-by-feature comparison goes stale within weeks. The questions that keep their answers are structural: who owns the source, whose models run the work, where the configuration and memory live, and where the product can run. On those axes the two closed platforms agree, and Kortix answers differently.

Where the two closed platforms are the same

Neither Claude Cowork nor ChatGPT Work has a self-hosted edition. Claude Cowork’s help center says its sessions run in the cloud on Anthropic’s servers, in an isolated environment, and that sessions and files are saved to your Claude account. ChatGPT Work’s page describes an agent available on macOS, Windows, web and mobile through OpenAI’s service. Neither page offers a way to run the software on hardware you control, and neither publishes the source.

Anthropic and OpenAI hold the configuration, the memory and the work history. Cowork keeps projects, instructions and memory with the Claude account, and stores task history on Anthropic’s systems. ChatGPT Work keeps files and past work in a team space inside ChatGPT. You set the goal and review the output, and the account that holds all of it stays with the vendor.

The fork: who owns the work

Kortix is the open-source third option. It is an open-source AI Operating System: your agents, skills, company memory, and connectors in one git repo you own. An agent is agents/<name>.md (behavior) plus an agents.<name> block in kortix.yaml (governance). The configuration is a file you can diff and review.

Kortix runs where you choose. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free, and the self-hosting guide covers the install. Each session has its own isolated machine and branch. Session work reaches main through a change request. Merge is default-deny for agents.

Kortix treats the model as a separate choice. Any model provider with your own keys, and connector credentials are brokered server-side and never enter the machine.

# kortix.yamlagents:  support-triage:    file: agents/support-triage.md   # what it does, in plain markdown    connectors: all                  # what it may reach

The git repo is the fork in the road: the configuration, the memory and the work history are files you read, diff and keep, rather than state that lives with a vendor. Kortix is open source. Read it, fork it, audit it.

How to decide

If you want an agent that turns a goal into a deck or a report inside the vendor’s app, on the vendor’s model and in the vendor’s cloud, Claude Cowork and ChatGPT Work do that on their own terms. Pick the vendor you already pay for. If the work has to run on infrastructure you control, on a model you choose, with the configuration and history in a repo you own, pick Kortix, the leading open-source alternative to Claude Cowork and ChatGPT Work.

For more detail, the open-source AI agent platforms guide maps the field, and the self-hosted AI agent platform page covers running it on your own hardware. For a worked example against one rival, Read the Kortix blog.