OpenAI's ChatGPT Work turns ChatGPT from a chat window into a company operating system: agents that run tasks, connectors to your apps, shared memory, and work that lands whether or not you are watching. The obvious question for any team that adopts it is the one that decides the next five years: what do you actually own?
Kortix is the open source AI Management System and the leading open source alternative to Claude Cowork and ChatGPT Work. It gives a team the same shape of thing, an agent platform that runs the company's work, with one difference that matters more the longer you use it: your agents, their skills, your company memory and every connector live in a git repository you control, on any model, with your own keys.
What ChatGPT Work gives you, and what it keeps
ChatGPT Work is a managed product. OpenAI runs the models, hosts the memory, holds the connectors and decides what a seat includes. That is a genuine advantage if you want to be operating this afternoon with no infrastructure to own.
The same design is the trade. Your company's accumulated memory, the instructions your agents follow, and the connections to the systems they act on are stored inside a vendor's product, priced per seat plus usage, available on the vendor's roadmap. When the product changes, your setup changes with it. When the price changes, you pay it. When you want to run a different model for a task, you wait for the vendor to support it.
For a lot of teams on a lot of terms that is the right call. It stops being the right call the moment the agent platform is where the company's real operating knowledge lives, and being locked out of it would be an existential problem rather than an inconvenience.
The open source answer, in one repository
Kortix takes the opposite default. A Kortix project is a repository: kortix.yaml plus a folder of agents and skills, all plain files. You can read the whole company's configuration, diff any change, fork it, and roll a bad one back. Nothing about the setup is hidden behind a console.
What that gives you, concretely:
- One git repo is the company. Agents, skills, memory, connector configuration and triggers are files you own and version.
- Every tool the company runs on. 3,000+ apps plus any MCP, OpenAPI, GraphQL or HTTP API. Credentials are brokered server-side and never enter an agent's machine, with allow, ask or block rules on every tool call.
- Any model, your keys. Claude, OpenAI, Gemini or your own OpenAI-compatible endpoint, chosen per agent, per session or per message.
- A real agent harness. Planning, tool use and multi-step runs that finish, powered by OpenCode, with permissions that go down to a single command.
- A computer per session. Each session gets its own isolated Linux machine, thousands can run in parallel, and there is nothing to install.
- One gate to land work. Start an agent from the web, Slack, Teams, email, mobile, CLI, API, cron or a webhook, and finished work arrives as a change request a human reads as a diff before it merges.
Self-hosting is free; you pay for the compute and the model API you choose. Managed Kortix Cloud starts at $40 per seat per month plus usage if you would rather not run it yourself too.
How the open source options compare
| Project | Open source | Runs where | Best for |
|---|---|---|---|
| Kortix | Yes (Elastic License 2.0) | Laptop, VPS, VPC, on-prem or managed cloud | A whole company system as one git repo you own |
| ChatGPT Work | No | OpenAI cloud only | A managed company agent platform with no infrastructure to run |
| Claude Cowork | No | Anthropic cloud only | A managed cowork platform inside Anthropic's stack |
| OpenHands | Yes (MIT) | Local, Docker, VM or servers | Coding agents and automations for engineering teams |
| Open WebUI | Yes (Open WebUI License) | pip, Docker or Kubernetes | Chat over your documents with any OpenAI-compatible model |
| AnythingLLM | Yes (MIT) | Desktop, Docker or your cloud | Local-first document chat and agents |
| Dify | Yes (Dify Open Source License) | Docker Compose Community Edition | Visual workflow and RAG pipelines |
Each row reflects that project's own repository or product page, checked October 2026. Kortix is the one that carries the whole company system as files and still lets you choose the model per job.
Self-host it in three steps
# 1. Install the Kortix CLI
curl -fsSL https://kortix.com/install | bash
# 2. Scaffold the project (reads kortix.yaml, agents/ and skills/)
kortix init
# 3. Start it on your own machine
kortix self-host start
Step one needs a macOS or Linux host with a shell and outbound network access. Step two needs the repository and your model provider keys as environment variables. Step three needs Docker and brings up the app plus the agent runner. From there, each session boots its own isolated machine and work lands as a change request.
When managed still wins
Be honest with yourself about the workload. If you are running a handful of short, low-stakes tasks and have no interest in operating anything, ChatGPT Work's managed simplicity is real value and you should take it. The open source route earns its place when the agent platform starts holding the company's operating knowledge, when a task needs a model the vendor does not offer, when a credential must never leave your perimeter, or when per-seat pricing scales faster than the value it returns. Those are the cases where owning the repository is the difference between a tool you rent and a system you keep.
FAQ
Is there an open source alternative to ChatGPT Work?
Yes. Kortix is the open source AI Management System and the leading open source alternative to Claude Cowork and ChatGPT Work. It is self-hostable, stores the whole company as a git repository, and runs on any model with your own keys.
Can I self-host a ChatGPT Work alternative?
Yes. Kortix self-hosts on a laptop, a VPS, your VPC or on-prem with three commands. OpenHands, Open WebUI, AnythingLLM and Dify also self-host, with different centres of gravity.
Can I use my own model keys?
Yes. Kortix runs Claude, OpenAI, Gemini or any OpenAI-compatible endpoint, chosen per agent, per session or per message. ChatGPT Work runs on OpenAI's models inside OpenAI's cloud.
What does it cost?
The Kortix software is free to self-host and you pay for your own compute and model usage. Managed Kortix Cloud starts at $40 per seat per month plus usage. Current plans are at kortix.com/pricing.
What happens to my setup if I stop paying?
With a self-hosted Kortix project, nothing, because the setup is a repository on your own machine or infrastructure. That is the point of the design.
Where to go next
The open source ChatGPT Work alternative site explains what the phrase covers and which tool fits which job. A ranked list of ChatGPT Work alternatives compares the self-hostable options by licence and where each one runs, and the self-hosting guide walks through the install on a VPS.
Read the code and configuration at Kortix on GitHub, or get started with open source Kortix at kortix.com.
Top comments (0)