Launch a Cloud agent

Launching creates a dedicated cloud computer, boot disk, network rules, and Pneum.ai runtime. The Cloud Console preselects a recommended configuration, but you can change placement, compute, networking, and model credentials before anything is created.

Open the launch dialog

Go to Cloud Console → Agents and select the plus button in Agent fleet. The dialog shows a generated name, a Recommended setup ready summary, and an infrastructure estimate.

If launching is disabled, the button’s label explains the gate. Common gates are incomplete billing, an unverified email, a plan limit, or no currently enabled provider.

Name the agent

Use a name that identifies the environment’s purpose, such as Home assistant or Release helper. The refresh icon generates another name. The name is used in the console and contributes to the provider resource name, but the immutable agent ID remains the authoritative identifier.

Use or customize the recommendation

For a standard first agent, keep Customize setup (optional) closed. Pneum.ai selects:

• an enabled cloud provider;
• a CPU runtime;
• a launchable location;
• a recommended instance size;
• a valid boot-disk size;
• internet access with an IP allowlist;
• managed model credentials.

Open customization when you need GPU voice features, a specific location, Tailscale, a larger computer, a larger disk, or BYOK. See launch options for field-level guidance.

Read the estimate

The estimate separates machine, GPU, and boot-disk cost. It also names the pricing source, configuration, and location when available. An estimate is not a reservation or final invoice; egress, taxes, delayed provider adjustments, and runtime changes may differ.

Do not launch if the estimate is unavailable and you are not prepared to accept the provider’s actual price. Retry after the location and machine lists finish loading.

Submit the launch

Select Launch Agent once the form is valid. Avoid closing the browser during the initial request. After the control plane accepts it, the dialog closes and an agent card appears.

The provider operation is asynchronous. A visible card means the request is tracked; it does not mean the workspace is ready yet.

Follow provisioning

Watch the status and any error block on the agent card. A successful first boot progresses from provisioning to running, then loading, then ready. The UI link can appear before the services respond, so use the readiness label rather than repeatedly opening the endpoint.

Public agents receive a stable HTTPS hostname when configured. Tailscale agents report their MagicDNS identity and expose the private UI after tailnet HTTPS is ready.

After the agent is ready

1. Open the UI link.
2. Complete first-run setup.
3. Install only the integrations the agent needs.
4. Review approval policies before enabling write tools.
5. Configure a schedule or stop the VM when it is not needed.

Launching begins infrastructure billing. Closing the browser does not stop or terminate the VM.