Install and run your first task

By the end of this guide, you will be able to open AGI Cockpit, run your first task, review its result, and complete the task.

1. Install AGI Cockpit

Open the AGI Cockpit download page and choose the package for your operating system.

On Windows, install AGI Cockpit from Microsoft Store. On macOS, open the .dmg and move AGI Cockpit to Applications. On Linux, use the distributed AppImage or .deb package.

After launch, choose guest mode if you want to continue without an account. Sign in as an AGI Labo member if you need Autorun or remote access from another device.

2. Prepare an agent

The new-task screen can start Claude Code, Codex CLI, Grok Build, Antigravity CLI, Cursor CLI, Cockpit Agent, or Terminal. If Cockpit cannot find an agent, the screen shows Install or Open Settings.

If the screen shows Install, use it to install the corresponding CLI. If it shows Open Settings, confirm the launch command in Settings. Return to the new-task screen after installation; the agent is ready when it becomes selectable.

When you use Claude Code, Codex, or Grok Build in Native UI, you can start before authenticating. Task details then shows sign-in guidance, and Cockpit retries the first instruction in the same task after authentication succeeds. This guidance and automatic retry do not apply to Terminal UI or the Terminal agent. In those modes, complete the CLI's sign-in flow inside the terminal, then resume or recreate the task.

An AI agent's subscription and authentication are separate from your AGI Labo sign-in. Signing in as an AGI Labo member does not authenticate Claude Code, Codex, or another task agent.

3. Create your first task

1. Open New task at the top of the window.
2. Choose Project workspace or Temporary folder. For an existing project, select only the directory you intend the agent to inspect.
3. Select an AI agent.
4. For supported agents, choose Native UI or Terminal as the UI mode.
5. For this first check, choose Supervised approval mode and enter a short request with a clear completion condition.
6. Create the task.

A read-only request is a safe first check.

Inspect this folder and describe its main files and their roles in no more than five points. Do not change any files.

A temporary folder is deleted automatically when the task is completed. Before starting in an existing project, make sure you understand that the agent can operate on files in that directory.

4. Confirm the successful state

The new task appears in the task list and normally starts as Running. The task details view shows the instruction you sent and the agent's progress or response.

After one response finishes, the task moves to Awaiting confirmation. This does not mean the entire task is complete. It means the task can accept another instruction or decision. If the result is incomplete, send a follow-up instruction from the composer in task details.

When the work is finished, select Complete in task details. Completed tasks move out of the active list and do not accept regular follow-up instructions.

5. If the task does not start

Create a task from an AI agent

If the cockpit command is installed from Cockpit settings, an AI agent can create the same kind of task:

cockpit task create \
  --instruction "Inspect this folder and describe its structure in no more than five points. Do not change any files." \
  --directory /path/to/project

If the command omits the directory, Cockpit starts the task in an operating-system temporary folder.

Related pages

• Task list and task details
• Ask
• Autorun