# How to migrate from Claude Cowork to an open source stack

> Moving off Claude Cowork means taking the work that lives in your Claude account and rebuilding it as files in a git repo you own. Kortix is the open-source AI Operating System, and this is the order that avoids rework: inventory, repo, model keys, connectors, first session.

URL: https://opensourceclaudecowork.com/migrate-from-claude-cowork

_Quickstart_

## The target layout, from the Kortix docs

`project tree`

```
project: a git repo with kortix.yaml at its root
agents/   skills/   memory/
session: one agent in its own sandbox on a new branch
change request: a person merges it into the default branch
```

Source: kortix.com/docs

Everything Claude Cowork keeps in your account becomes files in this repo, and finished work reaches the default branch through a change request.

## What moves when you leave Claude Cowork

Claude Cowork keeps your work with your Claude account. Anthropic’s help center says Cowork sessions run on Anthropic’s servers and that your sessions and files are saved to your Claude account, and it groups related tasks into projects with their own files, instructions and memory ([Claude help center](https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork)). There is no export file that carries all of that into another product, so a migration is a rebuild. The pieces you rebuild are the projects and their instructions, the memory each one holds, and the skills and plugins that shape how the work gets done.

In Kortix those pieces become files. Agents and skills are markdown in a git repo, company memory is files in the same repo, and `kortix.yaml` declares the machine image, connectors and triggers. The docs describe the model: a project is a git repo with `kortix.yaml` at its root, and the agents, skills, memory, tools, triggers and connectors are files in it ([Kortix docs](https://kortix.com/docs)).

## The migration, in order

Work in this order to avoid rework.

1. Inventory the work. List every Cowork project, the instruction set behind it, the memory it holds, and the plugins or skills it relies on.
2. Create the repo. Make one git repo for the company and add `agents/`, `skills/` and `memory/`.
3. Rebuild each working agent as a file. Write the behavior as an agent markdown file and the governance as an `agents.<name>` block in `kortix.yaml`.
4. Move the memory. Copy what each project learned into the repo’s memory files, so the context survives the move.
5. Connect model keys. Any model provider with your own keys. Or the ChatGPT plan you already pay for. Or sign in with your OpenCode Console account for OpenCode Zen and Go.
6. Connect the tools. Agents reach 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP, and connector credentials are brokered server-side and never enter the machine.
7. Run the first session. Each session has its own isolated machine and branch, and session work reaches `main` through a change request you approve.

## What replaces Cowork’s approval step

Cowork shows its plan and waits for your approval before anything significant. Kortix keeps a gate in the same place and makes it structural. An agent commits and pushes to its session branch, opens a change request, and a person merges it. Merge is default-deny for agents, and a session can never merge the change request it opened itself. The agent can hold broad tool access inside its sandbox, and the change request is where you keep control.

## Where the migrated stack runs

You can run the migrated stack in more than one place. Run Kortix on Kortix Cloud, in your VPC, or on your own on-prem network, and self-host is free. The [self-hosting guide](/self-hosting) covers the install on your own box, and [Claude Cowork vs ChatGPT Work](/claude-cowork-vs-chatgpt-work) sets the closed products against the open one. For a worked comparison, [Read the Kortix blog](https://kortix.com/blog/kortix-vs-claude-cowork).

_FAQ_

## Migration questions

### Can I export my Claude Cowork projects and import them into Kortix?

No. Anthropic stores Cowork sessions and files with your Claude account, and there is no portable export that another product can read. You rebuild each project as agent, skill and memory files in the repo.

### Do I have to rebuild everything by hand?

You write the agent and skill files, and an agent can help draft them from what the project already does. The work is one file per agent plus the `kortix.yaml` block, not a new application.

### Will switching change which models I can use?

Kortix accepts any model provider with your own keys, or the ChatGPT plan you already pay for, or you sign in with your OpenCode Console account for OpenCode Zen and Go.

### Does an agent merge its own work after the move?

No. Work reaches main through a change request a person approves. Merge is default-deny for agents, and a session can never merge the change request it opened itself.

## Move your agents into a repo you own

Start with one job, then bring the rest of the work across.

[Try Kortix](https://kortix.com) · [See the self-hosting guide](https://opensourceclaudecowork.com/self-hosting)