Plans and agent limits

Your plan controls product access. Cloud infrastructure and model usage remain separate metered charges.

Basic

Basic is intended for one active CPU agent. That agent includes the assistant workspace, voice and chat, notes, research, connected services, approvals, and supported tools.

An agent counts toward the active limit while it is provisioning, starting, running, stopping, stopped, or terminating. A stopped agent still exists and retains its disk, so it still occupies the single active-agent slot.

Basic managed model usage includes the service charge shown in current pricing. You can instead select BYOK and pay a supported model provider directly.

Pro

Pro supports multiple separated agents and allows CPU or GPU at launch. An eligible GPU launch can host Voice Studio or run compatible open-weight models, including Gemma variants that fit the selected GPU and its available memory.

Every running agent has its own compute charge. Separation is useful for independent work, home, customer, or project contexts, but it is not a way to share one runtime bill across several agents.

Team

Team may appear in product material as coming soon. Do not plan production access around a Team feature until the plan is selectable in checkout and its workspace controls are available.

Subscription versus usage

The invoice can contain three layers:

1. the recurring plan subscription;
2. reconciled cloud infrastructure usage;
3. managed model usage.

BYOK model charges appear at the model provider rather than in Pneum.ai managed-model totals.

Changing plans

Open Cloud Console → Billing → Plan to compare or change an available subscription. Review the confirmation carefully. A downgrade cannot make an existing environment fit a lower limit by itself; stop and terminate unneeded agents before relying on the lower plan.

See plan limits reference for a compact comparison and usage and invoices for billing-page details.