Migrating a team off Claude Cowork: what to move, and in what order

Claude Cowork is excellent for one person on one desktop. The hard part starts when a team tries to move off it: the work is real, but the configuration, the memory and the permissions live in a closed product, not in anything you can copy. Kortix is the open-source AI Management System — the leading open-source alternative to Claude Cowork and ChatGPT Work — and this is the migration order that avoids losing what your team already taught its agents.

Start with what you actually own

Before moving anything, write down where the value sits. In Cowork, each person's agents, skills and context stay on their own machine. In Kortix, agents, skills, memory, triggers and connector config are files in one git repo you own, so they can be reviewed like code. That difference decides the migration: you are not copying a settings database, you are writing the company down. Settle the open-source question behind that first: https://claudecoworkalternative.com/is-claude-cowork-open-source.html

Move the configuration in three layers

The order that works is bottom-up, smallest blast radius first.

1. The machine and the model. Kortix is model-agnostic: pick the model per agent, per session or per message, bring your own API key, and keep the option to self-host. Nothing about your workflow should depend on one lab.
2. The agents and skills. Each is a markdown file in the repo. Port the prompts that already work, then let the team diff and review them the way they review code.
3. The permissions. This is the layer teams skip and regret. In Kortix every connector and every tool call can be set to allow, ask or block — down to the arguments of a single call. Move the Cowork approval habits into explicit rules.

Install and verify before you widen

Kortix installs in three commands: curl -fsSL https://kortix.com/install | bash, then kortix init, then kortix ship. Start with one team and one job, watch the session, and read the first change request before you invite the rest of the company. The step-by-step install notes are here: https://claudecoworkalternative.com/install.html

Run the migration as a change request

The property that makes this safe is that work reaches the default branch only through a change request a person reads as a diff. Migrate one workflow, review the diff, merge, then widen. The same shape applies to the repo itself: every agent, skill and permission rule is reviewed before it lands. The migration guide with the full sequence: https://claudecoworkalternative.com/migration.html

Self-host where the data must stay

If the reason you are leaving a hosted assistant is data residency, Kortix runs on your laptop, a VPS, your VPC or on-prem, or on managed cloud. Connector credentials are brokered server-side and never enter the machine. What self-hosting actually gets you: https://claudecoworkalternative.com/self-hosting.html

Finish with a comparison, not a feeling

A migration is not done until the old surface is compared against the new one on the axes that matter — ownership, model choice, deployment, connectors, review. The comparison we use is here: https://claudecoworkalternative.com/comparison.html and the common questions are answered here: https://claudecoworkalternative.com/faq.html

Kortix is open source (Elastic License 2.0) — self-host, read and modify the code. The full comparison hub is https://claudecoworkalternative.com/ and you can get started at https://kortix.com.