Migrate from Claude Cowork to open-source Kortix
You can move off Claude Cowork in three moves: write your agents, skills, and rules as files, bring your own model keys, and reconnect the tools your team already uses. Kortix, the open-source AI Operating System, holds all of it in one git repo you own.
What you leave behind
Claude Cowork runs each task in the cloud. Claude's work runs on Anthropic's servers, in an isolated environment, and your sessions and files are saved to your Claude account (per Anthropic's Cowork documentation). Local files come within reach only through the Claude Desktop app, and only in the folders you connected. As of October 2026, Cowork sits inside a paid Claude plan: Pro, Max, Team, or Enterprise.
The setup you build there lives in Anthropic's product: the global instructions, the folder instructions, the connectors, the scheduled tasks. The model is Claude, and there is no self-host path. When you delete a task, Anthropic removes it from backend storage systems within 30 days.
What you keep
You keep your files. On desktop, Cowork reads and writes files in the folders you connected, and those files stay on your machine either way. In Kortix they sit in a git repo you own, where your agents, skills, company memory, and connectors are files: versioned, diffable, and shared by the whole company.
The platform comes with you too. Kortix is open source: read it, fork it, audit it. Self-host is free.
Move 1: the repo
Create a Kortix project, and write the setup as files. kortix init scaffolds a project: it creates kortix.yaml plus your agents, skills, and runtime config. kortix ship pushes the repo and brings the whole thing live in the cloud. The full command surface is documented in the Kortix on GitHub README.
An agent is agents/<name>.md (behavior) plus an agents.<name> block in kortix.yaml (governance). The text you kept in Cowork's global instructions and folder instructions maps onto these files: an agent's markdown carries how one job is done, and skills carry know-how the whole company shares. Company memory lives in the repo as plain files today. Work comes back one way: a session's work reaches main through a change request a person approves, and merge is default-deny for agents. For the manifest, agents, models, and secrets, Read the docs.
Move 2: the models
Bring your own API 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. On Cowork the model is Claude. Here the choice is part of project setup, alongside agents and secrets, so it is versioned with the rest of the repo.
Move 3: the connectors
Point Kortix at the tools the team already uses. The connector catalog covers 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP. Connector credentials are brokered server-side and never enter the machine. For each tool you set Allow, Ask, or Block, down to the arguments of each call, and approval gates are off until you set them. The manifest declares the machine image, connectors and triggers, and secret names and grants, so what an agent may reach is written in a file you review.
The switch, in one pass
The order is the three moves above: repo first, then models, then connectors. Start a session and an agent works on its own cloud computer, on its own branch, and opens a change request when the work is ready. You review it and merge it to the default branch.
If you want the whole stack on your own hardware, the self-hosting guide has the install. The full comparison sets Kortix against the field, and the Kortix team keeps the head-to-head with Claude Cowork at Kortix.
Where a Cowork setup lands
# kortix.yamlagents: support-triage: file: agents/support-triage.md # what it does, in plain markdown connectors: all # what it may reachWhere a Cowork setup lands: an agent's behavior in a markdown file, its reach in this governance block, both reviewable in git.
FAQ
There is no export step. kortix init scaffolds a project with agents, skills, and runtime config, and kortix ship brings it live in the cloud. The real work is rewriting your instructions as agent and skill files, and you set that pace.
Your Cowork task history and outputs stay in your Claude account under Anthropic’s terms. What you rebuild is the setup itself, and it comes over as files you own: instructions become agents and skills in your repo.
Yes. 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.
The project is a git repo you own: run it on Kortix Cloud, in your VPC, or on your own on-prem network. Secrets are encrypted at rest with a key per project. In Cowork, by contrast, sessions and files are saved to your Claude account on Anthropic’s servers.
Make the switch in three moves
Your agents, skills, and rules as files in one repo you own.