ChatGPT Work is OpenAI's agent inside ChatGPT. OpenAI launched it on July 9, 2026, and its own product page describes an agent that takes action across your apps and files, runs scheduled tasks, and turns a goal into finished work. It runs on OpenAI's cloud, uses OpenAI's models, and is reached through OpenAI's apps.
That is a strong product. The question this post answers is narrower: what do you run when the agent platform has to live on your infrastructure, use the models you pick, and keep its configuration in a repo your team owns?
For that, the open source answer is Kortix. Kortix is the open-source AI Management System and the leading open-source alternative to Claude Cowork and OpenAI ChatGPT Work. The agents, the skills they share, the company memory, the connector configuration and the triggers are files in one git repository you own. Each session boots its own isolated Linux machine on its own branch, and work reaches your default branch only through a change request a person reviews.
What you actually own with Kortix
The unit of the system is a git repo. An agent is a Markdown file. A skill is a Markdown file. Memory is a folder that accumulates. A single kortix.yaml manifest wires the connectors, the triggers and the machine image together. You can grep the whole company, diff any change to an agent or a skill, and roll any part of it back.
Three more things follow from that. The platform is model-agnostic, so an agent can run on OpenAI, Anthropic, Google or your own OpenAI-compatible endpoint, per agent, per session or per message, with your own API keys. Connector credentials are brokered server-side and never enter the machine, and each tool call can be set to allow, ask or block down to its arguments. And you choose where it runs: a laptop, a VPS, your VPC, an on-prem network, or managed cloud.
Three commands to a running agent
curl -fsSL https://kortix.com/install | bash
kortix init my-app
kortix ship
kortix init scaffolds the project and writes kortix.yaml. kortix ship lints the manifest, commits your branch and brings the project live. From there, start a session and review what it proposes:
kortix sessions new --prompt "Draft this week's report from the repo and open a change request"
kortix cr ls
The session runs on its own isolated machine. The agent can install, run and break anything; only committed work survives, and nothing reaches main until you read the diff and merge it.
A starter repository you can clone
If you want to see the shape of a Kortix project before you commit to one, there is a working starter: open-source-chatgpt-work-alternative-starter. It ships a v2 kortix.yaml, a role agent, a weekly-report skill and a memory folder, all as plain files. Clone it, run kortix ship, and you have a ChatGPT Work style agent team in a repo you control.
Kortix and ChatGPT Work side by side
| Kortix | ChatGPT Work | |
|---|---|---|
| Open source | Yes, read, fork and self-host | No |
| Models | Any provider, your own keys | OpenAI's models |
| Where it runs | Self-host, VPC, on-prem or managed cloud | OpenAI's cloud and apps |
| Configuration | Files in one git repo you own | Inside OpenAI's product |
Checked against OpenAI's own pages in October 2026.
Where to go next
If you are weighing the switch, the companion site has a plain-language walkthrough of why teams pick an open source ChatGPT Work alternative and a longer rundown of ChatGPT Work alternatives for different team sizes. For the product itself, start with Kortix and Read the docs. The code is at Kortix on GitHub.
Top comments (0)