What is an open-source ChatGPT Work alternative, and who owns it?

What an open-source ChatGPT Work alternative is

An open-source ChatGPT Work alternative is an agent system you run on infrastructure you control that finishes the same kinds of work ChatGPT Work does: research, drafts, decks, and reports. The open-source part adds one property: you can read the code, run the system yourself, and keep the result. Kortix is the open-source AI Operating System, and how it compares with ChatGPT Work comes down to where the work and the configuration live.

What ChatGPT Work is, and what stays with OpenAI

OpenAI introduced ChatGPT Work on July 9, 2026 as "an agent in ChatGPT that helps you take on more ambitious tasks." It gathers context across your apps and files, produces finished materials such as sheets, slides, docs, and web apps, and can stay with a complex project for hours by breaking it into smaller steps (launch announcement). Plugins connect it to tools such as Slack, Microsoft Teams, Google Drive, and CRMs, and Scheduled Tasks run recurring work (OpenAI's product page, checked October 2026).

What stays with OpenAI is everything that decides how the agent behaves. The model is OpenAI's, the work runs in OpenAI's cloud, and the access, connected tools, and action limits are set by workspace admins in OpenAI's Admin Console, with oversight through OpenAI's Compliance API. You rent that configuration with a paid plan, and you do not get a copy of it to keep.

The ownership test: where your configuration lives

The fastest way to tell a hosted agent from one you own is to ask where the configuration lives. In an open-source system, the agents, their skills, the memory, and the connector definitions are files in a git repo you can clone, diff, and roll back. In a hosted product, those definitions sit inside the vendor's workspace and you reach them through a settings screen.

Kortix passes that test by construction. An agent is agents/<name>.md plus an agents.<name> block in kortix.yaml. Skills are skills/<name>/SKILL.md, memory is memory/, and kortix.yaml declares the machine image, the connectors and triggers, and the secret names and grants. Every key is listed in the manifest reference (Read the docs). Secret values never appear in the repo, and members and roles are live project state rather than a file.

Because those files are portable, ownership extends to hosting and reach. Self-host runs the same repo on a machine you control, and the connectors and automation page shows how a connector and a trigger are declared as a block in the repo.

What to look for in an alternative

Run four checks when you compare open-source options.

• The source is open. Read it, fork it, audit it.
• The configuration is yours. Agents, skills, memory, and connectors are files in a git repo you can clone.
• Hosting is your choice. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free.
• Governance is yours to set. Per-resource permissions for people and agents. Roles, groups, and an audit trail. Session work reaches main through a change request. Merge is default-deny for agents.

Kortix is the leading open-source alternative to Claude Cowork and ChatGPT Work, and it meets all four. The governance and permissions page covers the grants, the approval gates, and the change-request gate behind the last check.