First run in the agent workspace

Open the UI or Tailnet UI link only after the Cloud Console reports ready. The first visit creates the initial assistant profile and confirms how the agent should think and speak.

Step 1: Personalize your agent

Enter a Profile Name, review the Personality / System Instruction, choose a Wake Word, and select the appearance. The system instruction should describe durable behavior and boundaries, not a single request.

Good instructions identify the assistant’s role, preferred response style, important safety rules, and when it should ask for clarification. Do not place passwords or API keys in the instruction.

Step 2: Intelligence providers

If the agent launched with managed credentials, onboarding shows that the managed OpenRouter key is configured. You can continue immediately or add an optional provider.

For BYOK, select a provider, enter its key in the dedicated field, and wait for validation. A key is stored by the agent configuration system, not inserted into chat. Remove a provider that should no longer be available.

See managed models and BYOK for billing differences.

Step 3: Voice customization

Choose a visible voice card to preview and select a voice. The available list depends on the runtime and configured speech engine. A GPU agent with hosted Voice Studio can expose custom voice profiles after you create them.

Advanced private-network settings

On a Tailscale deployment, Advanced can configure a private OpenAI-compatible model server, Browser MCP endpoint, or external Voice Studio address. Enable Voice Studio transcription only when that endpoint serves the expected transcription API.

Use tailnet addresses that the agent VM can reach. Do not enter a device’s loopback address unless the service actually runs on the agent VM.

Learn the navigation

The desktop workspace opens studios from the top navigation:

• Profiles: identity, model, speech, and activation.
• Notes: Markdown notes and attached media.
• Vault: generated images, audio, and video.
• Voice: voice profiles and samples.
• Tools: tool policies, skill files, and custom tools.
• Agents: background subagent activity and task details.

The same destinations appear in the mobile dock or More sheet.

Verify the setup

1. Send a short typed message.
2. Test push-to-talk or wake mode.
3. Confirm the selected model responds.
4. Open Profiles Studio and look for a saved synchronization state.
5. Stop and start the Cloud agent only after confirming normal use, then verify the profile remains.

Continue with chat and voice or profiles and models.