Choose an engine for the work.

Pick an available engine, check its sign-in state and test the capabilities you need.

Separate the agent from the engine

The agent has a role and a workspace. The engine supplies the model and execution adapter for its turns. Choosing an engine affects sign-in, tools, limits and behavior. A familiar model name alone does not tell you which tools are available.

Start with a qualified path

Claude and Codex web sign-in and approved command execution have been checked live for the hosted preview. Other adapters exist in the public core, but installation, credentials and deployment configuration determine availability. Do not assume every provider listed on the website is hosted-ready.

Evaluate a real task

Use a short, representative task with a clear expected result. Check the response, tool record and approval flow. If the provider is signed out or over its limit, address that condition before interpreting an error as a problem with your instructions.

Keep exploring

Learn

Build your first AI team.

Start with one agent, one clear task, and a result you can review.

Get started Read the setup guide →

Preview · Bring your sandbox provider and AI account

Choose an engine for the work.

Separate the agent from the engine

Start with a qualified path

Evaluate a real task

Keep exploring

A practical guide to working with agents.

Give every agent a place on the team.

Understand where the work runs.

Build your first AI team.