This guide runs one short read-only request from task creation through result review and completion. If preparation is not finished, complete Install AGI Cockpit and Initial setup first.

1. Create a new task

Open New task at the top of the window.
Under Workspace, choose an existing project, select Create project, choose an existing temporary folder, or select New temporary folder. If the project contains several folders, select the working folder for this task.
Select an AI agent.
For supported agents, use Mode to choose Native UI or Terminal.
If account selection is available, keep Auto. Keep the built-in system prompt.
Choose Supervised approval mode, enter the following request, and create the task.

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

The project picker shows current projects, other projects, and recent folders. Create project lets you name a project and add one or more existing folders; if you add none, Cockpit creates an empty managed folder for it. The first folder is primary and is selected by default for new tasks. Open Project details to review or change the project's folders before creating the task.

A temporary folder created for an earlier task appears by folder name. Selecting it reuses the folder without clearing its current contents; select New temporary folder to start in a different empty location. When unfinished tasks share a temporary folder, completing one of them does not delete the folder while another unfinished task still uses it.

When you select an existing working folder, Cockpit checks before task creation that the path exists, is a folder, and can be read and accessed. A missing path, a selected file, insufficient permission, or a failed check blocks creation. Recover with Choose another folder, New temporary folder, or Check again.

Smart routing is for AGI Labo members. If you select Smart routing while signed out, Cockpit explains the feature and offers View membership plans and Sign in with a member account. After signing in as a member, turn it on to let Cockpit choose an AI agent, model, reasoning level, and workspace from the currently available candidates based on the instruction. Manually selecting a workspace fixes only the workspace; agent and model routing remains active. Under Routing policy, enter preferences such as agents or models to prioritize. The policy saves automatically and applies to future tasks.

When the workspace remains Auto, Cockpit offers only directories that still exist and can be used for a new top-level task. Temporary or internal directories and workspaces reserved for child tasks are excluded. A workspace that you select manually remains fixed and is not replaced by this filtering.

Smart routing creates ordinary AI-agent tasks in Native UI and uses Auto for the account. It does not route Terminal or Creative Studio work. If selection fails, check the agent CLI, provider, and API key settings, or turn off Smart routing and select the configuration manually.

For an agent that reports usage, the area near the composer shows the allowances for the selected agent and account. A fixed account shows that profile. Auto previews the account chosen by the same selection method used when the task is created and labels it Auto · account name. Do not interpret an unavailable value as 0% remaining; open the details to inspect authentication state, source, and reset time when needed.

Draft instructions and attachments remain available until the task is created or the creation screen is explicitly closed. You can inspect Settings or another screen and return to New task without re-entering them. This draft does not persist after the app quits.

A temporary folder is deleted automatically when the task is completed. In a project, the task starts in the selected working folder. Supported agents can also use other folders in that project, so include only folders the task is allowed to access.

2. Confirm that it is running

Cockpit adds the new task to the task list and selects it immediately. It normally enters Running, and task details show the instruction and the agent's progress or response. In Terminal UI, the terminal may appear first and connects to the same process when the runtime is ready.

If a sign-in notice appears, complete authentication for that agent. Native UI retries the same instruction. In Terminal UI or Terminal, sign in from the terminal and resume the task if needed.

3. Review the result

After one response finishes, the task enters Awaiting confirmation. This does not mean all work is complete; the task is ready for another instruction or decision.

Check whether the response satisfies the request. Send a follow-up from the task details composer if something is missing. For tasks that change files, review the diff and generated artifacts as well as the conversation.

4. Complete the task

When every required result is ready, select Complete in task details. Completed tasks move out of active work and do not accept ordinary follow-up instructions.

A temporary working directory is deleted on completion. Save any required result to a persistent location first.

Create the first task from the CLI

After Cockpit integration is configured, create the same task from a supported AI agent or shell:

cockpit task create \
  --instruction "Inspect this folder in no more than five points. Do not change files." \
  --project Example

cat instruction.md | cockpit task create --stdin --project Example
cockpit task create --instruction-file instruction.md --project Example --directory /path/to/project/folder

Use --stdin or --instruction-file for multiline content or text containing backticks, quotes, $, or code fences. --project accepts an ID or exact name and uses the primary folder unless --directory selects another folder in that project. If both are omitted, the task starts in an operating-system temporary folder.

Your first task

1. Create a new task

2. Confirm that it is running

3. Review the result

4. Complete the task

Create the first task from the CLI

Read next