suna

What Kortix is

Kortix is an open source AI management system described as the leading open source alternative to closed products such as Claude Cowork and ChatGPT Work. The core idea is that a company's agents, the skills those agents share, the company's accumulated memory, and every connector should live in one git repository that the organization owns outright. That means the configuration is versioned, diffable, and shareable across the whole company rather than locked inside a vendor's product. The README contrasts Kortix with the closed alternatives on several axes: the source is open so anyone can read, fork, or audit it; any model provider works with the user's own API tokens; and it can run in the project's cloud, in the user's own virtual private cloud, or on premises. Pricing for the managed cloud is stated as forty dollars per seat per month plus usage, while self hosting is free. The project currently shows 20,132 stars and 3,432 forks, a large following for a tool in this category, and it is written in TypeScript. The homepage is kortix.com, which also hosts the documentation and the managed cloud sign up. By making the setup a repository instead of a SaaS account, Kortix aims to remove the vendor lock in that the README argues is built into lab owned competitors.

How a company becomes a git repo

The README makes the git repository metaphor literal rather than figurative. A project is a git repo plus a kortix.yaml file, and inside it sit the agents, the skills they have built, the way work actually gets done, the facts the company has learned, and the definition of the machines everything runs on. The CLI scaffolds this with three commands: first curl the install script and run it, then kortix init to create the project with its agents, skills, and runtime config, and finally kortix ship to push the repo and bring the whole thing live in the cloud. After that the loop continues with commands such as kortix sessions new to start a session with a prompt, kortix cr ls to review proposed changes, and kortix chat to talk to a session's agent from the terminal. The documentation reference lists the full command surface. For users who want zero setup, the site offers sign up, project creation, and a session start with nothing to install locally. The framing is that a company can be built and improved like a codebase, with the configuration as the source of truth and the cloud as the runtime that executes it. This turns organizational knowledge into plain files a team can review and fork.

Running agents on cloud computers

Each Kortix session gets its own cloud computer, described as a disposable, isolated Linux sandbox running on a branch named after the session. The agent, built on OpenCode, works inside that sandbox and can install, run, and break anything, but only what it commits survives, which contains the blast radius of a mistaken action. Work reaches the main branch only through a change request that a human approves, so the company self improves one reviewed change at a time. The README notes that thousands of sandboxes can run in parallel on the same configuration, each fully isolated and each feeding results back through change requests. The architecture diagram in the README shows the path from project to session to cloud computer to agent to change request and finally to main. Because the sessions are isolated and branch based, a failed experiment stays in its own sandbox instead of polluting shared state. The model choice, API tokens, and infrastructure all remain the user's, which the README presents as the defining difference from competitors that run on their own cloud with their own models. The open source license status is not a standard SPDX identifier in the repository metadata, so users should check the repo for the exact terms before relying on it commercially.

Editorial conclusion

Kortix carries 20,132 stars, is written in TypeScript, and is presented as an open source AI management system whose configuration lives in a git repository the user owns.