Launch options reference

Open Customize setup (optional) in the launch dialog to control the deployment. Changing one field can reload dependent choices and the cost estimate.

Cloud Provider

The selector shows providers known to the control plane. Only providers marked ready can be chosen. An unavailable option does not mean your account is broken; that provider may not have passed its regional, image, networking, or billing readiness checks.

Provider choice affects available regions, machine names, pricing sources, and provisioning time. It does not change the core Pneum.ai workspace.

Runtime Profile

Choose CPU for everyday chat, tools, notes, research, and connected services. Choose GPU for hosted Voice Studio or compatible open-weight models such as Gemma.

Basic plans are CPU-only. GPU availability also depends on provider configuration and regional capacity. Model size and speed depend on GPU memory and the selected model. A GPU selection can require a larger minimum disk and may offer fewer launchable zones.

Region and Zone

auto placement (nearest available) lets Pneum.ai choose a launchable region using the available location signal. A manually selected region keeps data and compute in that region but may have a different price or capacity profile.

Within a region, any (auto-select in region) lets the provider choose an available zone. Pin a zone only when another resource or policy requires it. GPU capacity is especially sensitive to zone selection.

Instance Size

The list describes each machine with a tier label, provider machine name, vCPU, and RAM. The default balances the expected runtime with cost. Choose more capacity when sustained tools, browser work, or concurrent background tasks exhaust the recommended size.

A larger machine raises the hourly estimate immediately. Changing machine size after launch is not currently a Cloud Console operation; launch a replacement environment if the original size is wrong.

Boot Disk (GB)

The disk stores the runtime, updates, notes, configuration, and local artifacts. The form enforces a minimum for the selected runtime. Increase it for coding workspaces, retained media, or large local models.

Disk storage bills while the agent is stopped. Termination deletes the disk.

Exposure

Internet creates a public HTTPS endpoint protected by an IP allowlist. Tailscale joins the computer to the tailnet configured in account settings and keeps the workspace private to that network.

For internet exposure, use Add my IP and keep the range as narrow as practical. For private resources or stable access across changing public IPs, configure Tailscale before launch.

LLM Credentials

Managed provisions an OpenRouter key controlled and metered by Pneum.ai. Prompts are not stored in the Cloud Console billing system. You can set a monthly managed-model spending cap.

BYOK does not provision the managed key. Add a supported provider credential inside the agent during onboarding or in Profiles Studio. Provider charges then go directly to that provider.

OpenRouter Attribution sends Pneum.ai attribution headers on managed requests. It does not add conversation content to the header.

Validate the final combination

Before launch, confirm the summary, estimate source, instance, location, image, disk, and any GPU line. If a dependent list is still loading, wait before submitting. See the launch field table for a compact reference.