Open source AI agent platform
Kortix: The Open Source AI Agent Platform You Own End to End
An open-source AI agent platform is the software layer a company runs its AI agents on and owns outright, and Kortix is the open-source AI Management System built to be exactly that. Your agents, the skills they share, your company memory and every connector live in one git repository you own, and each session runs on its own isolated machine. Run it on a laptop, a VPS, your own VPC or on-prem hardware, or let Kortix host it.
# the company, as one file version: 2 harness: opencode image: kortix/agent-computer:latest models: default: your-own-key connectors: 3000+ apps, MCP, HTTP triggers: - cron: "0 8 * * *" - webhook: signed permissions: allow | ask | block
What you get
Six things the platform hands you, as files you own
One git repo holds the whole company
Agents, skills, memory, connector config and triggers are text files you own. Grep the whole company, diff any change and roll any part of it back.
Any model, your own keys
Kortix is model-agnostic. Pick the model per agent, per session or per message, bring a key from a major provider, or point it at an OpenAI-compatible endpoint.
Every session gets its own Linux machine
Each session boots an isolated sandbox on its own branch, so an agent can install, run and break things without touching production. Only what it commits survives.
3,000+ connectors and any API
Wire up 3,000+ apps in a click, or any MCP, OpenAPI, Postman, GraphQL or raw HTTP endpoint. Connector credentials are brokered server-side and never enter the machine.
A human gate on every change
Finished work lands on main as a change request a person reads as a diff. Agents can also start from web, Slack, Teams, email, mobile, CLI or API, or from a cron schedule with nobody asking.
Self-host or managed cloud
Run Kortix on your own infrastructure (laptop, VPS, VPC, on-prem) or use Kortix Cloud. The same config runs either way.
Install
Install it and run the loop
The install is three commands. The CLI sets up the project, and kortix ship brings the whole thing live.
# 1. Install the CLI
curl -fsSL https://kortix.com/install | bash
# 2. Scaffold a project: creates kortix.yaml plus your agents, skills and runtime config
kortix init
# 3. Ship it: pushes your repo and brings the whole thing live
kortix shipFrom there, kortix sessions new --prompt "..." starts a session, kortix cr ls lists what agents have proposed, and kortix chat talks to a session's agent from your terminal.
Every unit of work then follows one path:
- A project is a git repo with a
kortix.yamlthat declares the machine image, the connectors and the triggers. - A session starts an agent on an isolated sandbox, on its own branch.
- The agent commits and pushes its work.
- The work opens a change request.
- You review the diff and merge it to the default branch.
The comparison
Why an open source platform is the one you own
Claude Cowork and ChatGPT Work are heading for the same place: a full system an entire company runs on. OpenAI and Anthropic run theirs in their own clouds, and your configuration stays inside their products. Kortix is the leading open-source alternative to both, and the difference is ownership. Every agent, all of its data, every skill, every connector, the memory and the whole configuration are files in a git repo you own. You can grep the company, diff any change, roll it back, and move it to another machine. Kortix is open source (Elastic License 2.0), so you can self-host it, read the code and modify it (Kortix on GitHub).
Kortix is the open-source pick: the whole company as one git repo you own, any model with your own keys, and a human gate on every change.
Try KortixClaude Cowork completes multi-step knowledge work across a person's files and tools (Anthropic), and ChatGPT Work turns a goal into finished files (OpenAI). Cowork runs Claude models and Work runs OpenAI's models. Hosting follows the vendor: Cowork in Anthropic's cloud or your own Bedrock, Google Cloud or Microsoft Foundry account, Work in OpenAI's cloud.
| Platform | Kortix | Claude Cowork | ChatGPT Work |
|---|---|---|---|
| Open source | Yes, Elastic License 2.0 | No | No |
| Models | Any provider, your own keys | Claude models | OpenAI models |
| Where it runs | Your laptop, VPS, VPC, on-prem or Kortix Cloud | Anthropic cloud, Bedrock, Google Cloud, Microsoft Foundry | OpenAI cloud, plus desktop local files |
Models are a choice on Kortix and a constraint on the closed platforms. Kortix takes any major provider with your own keys, or an OpenAI-compatible endpoint. The larger difference is who owns the configuration: on Cowork your folders, tools and permissions live in Anthropic's product, and on Work they live in OpenAI's. On Kortix they are files in a git repo you own, versioned and diffable next to the code.
Pick your priority
What a company actually wants, and how Kortix delivers it
The whole company is a repository you own
Agents, skills, memory, connector config and triggers are text files in one git repo. Grep it, diff any change, roll any part of it back, and move it to another machine.
Kortix is open source (Elastic License 2.0) — self-host, read and modify the code.
Pick the model per agent, per session or per message
Bring an API key from a major provider, use the Kortix Gateway, sign in with a ChatGPT subscription you already pay for, or point an agent at any OpenAI-compatible endpoint.
Model choice is a setting in the repo, so changing it never changes how your agents, skills or connectors are defined.
Every tool the company runs on
Wire up 3,000+ apps in a click, or any MCP, OpenAPI, Postman, GraphQL or raw HTTP endpoint. Credentials are brokered server-side, never inside the machine.
Each tool call is ruled allow, ask or block, down to a single shell command.
Runs where you decide
Install on a laptop, a VPS, your own VPC or an on-prem network, or use managed cloud. The same configuration runs either way and the CLI switches hosts.
Self-hosting the open-source platform is free; cloud plans buy hosting and compute. See the self-hosting guide.
Nothing reaches main without a person
Each session runs on its own isolated machine and branch. Work lands as a change request a human reads as a diff, and merge is deny-by-default for an agent.
Many sessions run at once because none can change production without an approving review.
Questions
Questions before you start
Do I have to install anything before I can use Kortix?+
No. You can create a project in the web app and start a session without installing anything. The CLI is there when you want the same platform from your terminal, and it installs with one command.
Can I use the ChatGPT subscription I already pay for?+
Yes. Kortix takes an API key from any major model provider, the Kortix Gateway, or the ChatGPT subscription you already pay for. You choose the model per agent, per session or per message.
Can I run Kortix on my own server or VPC?+
Yes. Kortix runs on a laptop, a VPS, your own VPC or an on-prem network, and the CLI switches between your host and Kortix Cloud with kortix hosts use selfhost and kortix hosts use cloud. The self-hosting guide walks through the sequence.
What stops an agent from making a change I did not approve?+
Two controls. You set permissions per tool call to allow, ask or block, down to a single shell command. And work reaches the default branch only through a change request a person merges. An agent can run and break things on its own isolated machine; only what it commits and a human approves survives.
Start with open source Kortix
Kortix is open source and self-hostable. Run it on your own machine, or start in the cloud and add the CLI later. Compare it with the other open-source options on the best open source AI agent platforms page.
curl -fsSL https://kortix.com/install | bash