Getting started with open-source Kortix: three commands to your first session

Getting started with Kortix takes three commands: install the CLI, scaffold a project, and ship it. Kortix is open source, so what you build is a git repository you own, and you can run the whole system on your own hardware or let Kortix host it for you. The whole setup runs from the installer to your first reviewed change request, with a short job for each command.

What lands in your project

Kortix is an open-source AI Management System: your agents, the skills they share, company memory, and connector configuration all live as files in one git repository (Kortix docs). Scaffolding a project creates a kortix.yaml manifest plus starter agents and runtime configuration (Kortix on GitHub). Because every layer is text, you can grep the whole company, diff any change an agent makes, and roll back any part of it.

Those files are the platform's moving parts. Agents are OpenCode agents with scoped reach into tools, one per role or task. Skills are reusable know-how that records how your company does a job, written once and shared into every session. Memory is the company brain, kept in plain files that accumulate what the system learns. You can watch a session live and diff every change an agent or a skill makes before it lands (Kortix about).

A session runs an agent inside an isolated Linux sandbox on a branch named after that session (Kortix docs). The agent can install, run, and break anything it needs, because only what it commits survives. Thousands of these sandboxes run in parallel on a single configuration, each fully isolated (Kortix about). The same about page describes the platform as six verbs you use in order: build, host, manage, monitor, train, and eval.

Install the CLI and scaffold

The installer downloads a prebuilt binary for macOS or Linux; Windows is not supported (Kortix CLI docs).

curl -fsSL https://kortix.com/install | bash
kortix init
kortix ship

kortix init creates a project directory with a general-purpose starter and a kortix.yaml manifest that declares the runtime version. kortix ship lints that manifest, commits your local changes, pushes the branch, and prompts for any missing secret or connection; the first run creates the cloud project if you have not linked one yet (Kortix CLI docs).

If you would rather skip the setup, sign up and create a project at Kortix, then start a session with nothing installed (Kortix on GitHub).

Start a session with a prompt

Hand a session a task and list what it proposes:

kortix sessions new --prompt "Summarize this week's commits and open a change request"
kortix cr ls

The first command boots a sandbox and runs an agent on your prompt; the second lists the change requests that sessions have opened (Kortix on GitHub). Work runs three ways: on demand when you ask, human-assisted when the agent checks in for a decision, or automated on a cron schedule or signed webhook. The same session can start from the web app, Slack, your terminal, or the API, and it returns finished work when the session is done.

Review the change request

An agent opens a change request when its session has commits ready to merge. You read the diff, then merge it or send it back (Kortix docs). Work reaches the default branch only through a change request a human approves, so the project changes one reviewed diff at a time and no agent merges its own work.

The gate is enforced, not advisory. Merge is deny-by-default for an agent. You can switch on approval gates for the actions that matter, and every session keeps a full audit trail (Kortix on GitHub).

From the CLI, kortix cr diff <cr> shows the unified patch and kortix cr merge <cr> lands it on the default branch (Kortix CLI docs).

Self-host it or use managed cloud

Kortix runs on your own infrastructure or on Kortix Cloud. Self-hosting is free and runs on a laptop, a VPS, your own VPC, or your own on-prem network (Kortix on GitHub). Start a production-style local instance with kortix self-host start, then point the CLI at it with kortix hosts use selfhost; kortix hosts use cloud switches back to managed hosting. The installer pulls its images from Docker Hub, so self-hosting still needs a network connection.

Managed hosting is priced per seat: the Team plan is $40 per seat per month plus usage (Kortix pricing). Both paths run the same command surface, so moving between your own host and the cloud does not change how you work.

Models, tools, and the licence

Kortix is model-agnostic. Choose the model per agent, per session, or per message, bring your own API key from any major provider, or use the ChatGPT subscription you already pay for. The harness underneath is powered by OpenCode and configured by a file in your repo, with permissions per tool down to a single shell command. Because the harness is open source too, the runtime is never the thing you are locked into. The same platform connects to 3,000+ apps in a click plus any MCP, OpenAPI, GraphQL, or raw HTTP API, with credentials brokered server-side so the raw key never enters the sandbox, and you can set each tool call to allow, ask, or block (Kortix).

Kortix is open source (Elastic License 2.0): self-host, read and modify the code (LICENSE).

Next reads: What is Kortix?, Introducing Kortix, and Kortix vs Claude Cowork.

Start with Kortix: free to start, free to self-host.