Skip to main content
When creating a task, make three choices. An Agent does the work you assign. It can also have its own contact identity and take part in group chats. The engine decides how work is executed, and the model provides the reasoning. One engine can use different models.

​
Install and choose an engine

Check what is installed in Library. When creating a task, the picker also separates installed engines from ones that need installing; follow the prompts to install. Current options include Codex, Claude Code, Qwen Code, GitHub Copilot, OpenCode, Hermes Agent, MiMo Code, CodeBuddy Code, Pi and more. They differ in display modes, sign-in methods and model connections. Check Library and the picker for the options available in your version. Start with one engine that already works and finish a small task. Compare other options later as your work requires. You do not need to install every engine at once.

​
Your own subscriptions and models provided by gugu

Manage existing vendor subscriptions or API connections in Settings → Subscriptions. When choosing a model, your own vendor models and models provided by gugu, billed in credits, are shown separately. Models provided by gugu are connected by gugu, so you do not configure a gateway address yourself. Check the picker for compatibility with the current engine and mode. Some engines may still require your own account as a prerequisite. Models and plans on sale change, so the docs do not keep a full list of models and prices. For choosing and billing details, see Models, subscriptions and credits.

​
Continue with the same Agent

A task keeps the requests, process and replies. Go back to the original task to add requests; you do not need to re-explain the whole background. An Agent in a task can also join the project group as a contact. To have it continue something, mention it or reply to its message, and say what result you want. An online status means the device it runs on is connected. Work status comes from the execution process. Handle connection failures, expired sign-ins and insufficient quota as the interface prompts. A successful tool run only means that step finished; still check the final deliverable. To open past work, see Task workspace. To check progress on your phone, see Mobile.