Initial setup
Check an AI agent, prepare the Cockpit skill and cockpit command, and reach the point where you can create your first task.
By the end of this guide, an agent and the Cockpit integration will be ready for your first task. If the flow did not open on first launch, use the app menu in the lower-left corner and select Setup → First-run setup.
1. Check an agent
Setup checks Claude Code, Codex CLI, Grok Build, Antigravity CLI, Cursor CLI, Qoder CLI, and Muse Code (macOS/Linux). It shows whether each CLI is installed and, for some agents, whether sign-in is ready. When a detected CLI is not yet usable, the screen states Plan and sign-in required. One available agent is enough to continue.
If Cockpit cannot find the agent you want, select Install. Cockpit Agent and Terminal do not require an external CLI.
2. Configure the Cockpit integration
The integration step installs or refreshes the Cockpit skill for detected external agents and makes the cockpit command available. It adds the command directory to shell configuration on macOS and Linux or to the user PATH on Windows.
On macOS and Linux, Cockpit adds or removes only its managed block. It preserves other shell settings, existing PATH lines, blank lines, and line endings. When the command directory is already present in a user-managed PATH entry, Cockpit does not rewrite the shell file.
This updates each agent's user-level skill directory and the applicable user-level path configuration. HTML Mode is optional and can be installed separately. If you add an agent later, open Skills from the app menu to review its status.
3. Sign in to the agent
An AI agent's plan and authentication are separate from AGI Cockpit's local features and AGI Labo sign-in. Obtain an eligible plan from the provider and sign in to that provider account for Claude Code, Codex, or another selected agent.
If a Claude Code, Codex, or Grok Build Native UI task starts before authentication, task details shows sign-in guidance and retries the first instruction after authentication. Terminal UI and Terminal do not provide this retry flow; complete the CLI sign-in inside the terminal, then resume or recreate the task.
For Muse Code, open Settings → Agents → Muse Code → Authentication. Choose Muse Code login and Sign in to Muse Code to run muse login in a terminal, or choose Meta API key and save your key. API-key mode cannot start tasks until a key is saved. After authentication, reload models and choose Native UI or terminal UI. See Agents and models for defaults and permission limits.
4. Choose how to use Cockpit
For local features only, select Start without signing in on the final screen. Sign in as an AGI Labo member if you want Smart routing, Autorun, or remote access from another device.
5. Continue recent work or skip
The final Continue where you left off step lists up to 200 Claude Code and Codex sessions updated in the last 30 days, grouped by workspace. Select individual sessions or an entire workspace, then choose Import selected. Imported tasks resume in Terminal UI. You can also create a project for a listed workspace before importing, or choose Skip for now.
Importing reads the original session files locally without changing or deleting them, and it does not start an agent. When the New task screen opens, setup is complete. Continue to Your first task.
If setup does not complete
| Message or state | What to check |
|---|---|
| The agent you want is missing | Select Manage agents to install its CLI or verify its launch command in Settings |
| Verifying agents appears | Cockpit is showing saved choices while rechecking them; wait for verification to finish |
| Could not verify agents appears | Review the saved choices and verify the CLI launch command under Agents in Settings |
| Native UI requires sign-in | Complete the guidance in task details and wait for the same instruction to retry |
| Terminal UI or Terminal requires sign-in | Complete that CLI's sign-in flow in the terminal, then resume or recreate the task |
After initial setup, use Agents and models, Accounts and Auto, and System prompts to tune the runtime environment.