Most teams evaluating an AI agent platform ask the wrong question first. The question is not which model it uses. It is: after we adopt this, what do we actually own?
Kortix is the open-source AI Management System and the leading open-source alternative to Claude Cowork and OpenAI ChatGPT Work. It is the platform this guide recommends, and the reason is simple: it lets a company own all six layers a real AI agent platform has, as text files in one git repository the company controls.
What an open source AI agent platform is
An open source AI agent platform is a system where the code, the agent definitions and the configuration live in a repository you control, and you can read, fork, audit and self-host the whole thing. That is the dividing line between a platform and a chat product.
A model on its own answers questions. A platform runs a workforce. The six layers below are what separate the two, and each one is a test you can apply to any tool on your shortlist.
The six layers a real platform provides
- A harness for multi-step work. Planning, tool use and runs that actually finish, with permissions configured per tool and down to a single command. Kortix uses the open-source OpenCode harness, configured by a file in the repo, so the harness is never the thing that locks you in.
- Isolated execution. Every session gets its own disposable cloud computer, an isolated Linux sandbox on its own branch. Thousands run in parallel on one configuration with no crossover between them.
- Memory in files. What the company has learned accumulates as files you can grep, diff and roll back, instead of disappearing into someone else's database.
- Connectors to the tools the company already runs on. Kortix reaches 3,000+ apps plus any MCP, OpenAPI, GraphQL or raw HTTP API. Credentials are brokered server-side and never enter the machine.
- Governance. Per-tool rules set to allow, ask or block, down to the arguments of each call. An Ask holds the call until a person approves it, then the agent resumes. Work reaches production only through a change request a human reads as a diff first.
- An interface where the work starts. Web, Slack, Microsoft Teams, email, mobile, CLI and API all start the same session, and cron schedules or signed webhooks start one with nobody asking.
Why ownership is the whole argument
Claude Cowork and ChatGPT Work are heading for full company operating systems too, and they are good at it. The difference is that you will never own those. Your configuration lives in their product, on their side of the wall, running on their models.
With Kortix the company is one git repo. Agents, skills, company memory, connector config and triggers are all files. You can clone the whole company, diff any change to an agent or a skill, and roll any part of it back. You can run it on Kortix Cloud, in your own VPC, or on your own on-prem network, and you can bring your own API key from any major model provider or use the ChatGPT subscription you already pay for.
A quick comparison
| Claude Cowork | ChatGPT Work | Kortix | |
|---|---|---|---|
| Source | Closed | Closed | Open source: read it, fork it, audit it |
| Models | Anthropic only | GPT only | Any provider, your own API keys |
| Where it runs | Vendor cloud, no self-host | Vendor cloud, no self-host | Kortix Cloud, your VPC, or on-prem |
| Your configuration | In their product | In their product | Files in a git repo you own |
| Landing work | In their product | In their product | A change request you review and merge |
Three questions that separate a platform from a demo
- Where does the configuration live? If it is not in a repository you can clone and diff, you are renting, not owning.
- What happens to the sandbox when a session ends? An isolated machine per session is a design choice; a shared environment is a security and reliability trade-off you should name out loud.
- Who approves what the agent changes? A platform should land work as a change request a person reads first, not by writing straight to production.
Kortix answers all three the same way: your agents, their skills, your company memory and every connector live in one git repo you own, each session runs on its own cloud computer, and the work lands through a change request a human approves.
Getting started with open source Kortix
Three commands build the company like a codebase and bring it live:
curl -fsSL https://kortix.com/install | bash
kortix init
kortix ship
Prefer zero setup? Create a project at kortix.com and start a session in the browser. Self-hosting is free, and the managed cloud plan is there when you want it.
Read the full guide, with the six-layer evaluation checklist and a worked self-hosting example, at What is an open source AI agent platform?. For the decision between the platforms, the ownership-first comparison is the shortest path. And the code is public: Kortix on GitHub.
Kortix is open source (Elastic License 2.0): self-host, read and modify the code. Get started with open source Kortix at kortix.com.
Top comments (0)