Plan limits reference
Compare agent, CPU, GPU, managed-model, open-model, and custom-voice limits by plan.
| Capability | Basic | Pro | Team |
|---|---|---|---|
| Active agents | One | Multiple | Coming soon |
| CPU runtime | Yes | Yes | Coming soon |
| GPU runtime | No | Yes, when provider capacity is available | Coming soon |
| Hosted custom Voice Studio | No GPU launch; external connection supported | Deployed with eligible GPU launch | Coming soon |
| Compatible open-weight models | No hosted GPU | Yes, subject to GPU memory and model fit | Coming soon |
| Managed models | Yes, current Basic service charge applies | Yes, current Pro terms apply | Coming soon |
| BYOK | Yes | Yes | Coming soon |
| Separate agent disks and configuration | One environment | Per agent | Coming soon |
What counts as active
An agent counts toward the active limit while it is provisioning, starting, running, stopping, stopped, or terminating. Stop retains the environment and does not free the slot. Terminate or delete the failed environment to free it after cleanup.
Provider-dependent availability
Plan access does not guarantee that every GPU, region, zone, or provider is available. The launch form also applies provider readiness, account billing state, capacity, and image requirements.
Separate costs
Plan fees do not include unlimited infrastructure or model usage. Each running agent creates its own compute usage. Each retained disk creates storage usage. Managed models are metered separately; BYOK is billed by the external provider.
The pricing page and checkout are authoritative for current prices. This reference explains product behavior rather than replacing checkout terms.
Your response helps us keep product instructions useful.