Quickstart

In five minutes you'll have a workspace, a manager and two specialists working on a real task. You need Lumen installed and at least one API key connected.

1. Create a workspace

Click New workspace and drag in a project folder. Lumen indexes the files and Git history so every model starts with the same picture.

2. Add a manager

Click Add model, choose Claude and set its role to Manager. The manager reads your mission, makes a plan and assigns work. You can pick any model as manager; Claude is a good default for planning and review.

3. Add two specialists

Add DeepSeek as Engineer and GPT as QA. Give each the folders it may touch. Lumen draws the connections for you: the manager delegates to both, and the engineer hands off to QA.

YouYou

Claude

DeepSeek

GPT

4. Give a mission

Open the crew's group chat and write what you want done, as you would to a colleague:

Users get logged out when a refresh token expires mid-session.
Find the cause, fix it, and add regression tests.

Claude replies with a plan, then assigns the fix to DeepSeek and the tests to GPT. You'll see each assignment as a gold line in the chat.

5. Approve the result

When the work is done, Claude asks you to review. Open the diff, read the test report and click Approve. That's a full loop.