Agents with a job to do.

Create named AI agents with instructions, a chosen engine and a workspace.

Start with a clear responsibility

An agent is a named place for a particular kind of work. Write down its job, the material it should use, the result you expect, and the decisions it should bring back to you. A research agent and a release reviewer need different instructions even when they use the same engine.

Keep the work together

Use its workspace for the files that belong to the task. Keep a brief close to the draft it informed, and state which files count as evidence. A name and instructions make work easier to return to; they do not make every generated answer correct.

Choose an engine for the task

Select an available engine in your installation and check its sign-in status. Capabilities vary across adapters. In the hosted preview, Claude and Codex sign-in and approved command execution have been checked live. Start with a short task before relying on a longer workflow.

Keep exploring

Guide

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

Agents with a job to do.

Start with a clear responsibility

Keep the work together

Choose an engine for the task

Keep exploring

Your first agent, step by step.

Understand where the work runs.

Put a team around a real task.

Build your first AI team.