Billing Overview
Cazpian is licensed as a flat annual subscription per deployment. There are no compute meters, no credits, and no per-query charges. AI is included.
What you pay for
| Item | VPC Dedicated | On-premises |
|---|---|---|
| Software subscription | Flat annual fee, by band | Flat annual fee, by band (+25%) |
| Infrastructure | Your AWS bill, paid by you at AWS prices (compute, storage, Amazon Bedrock) | Your data center |
| Deployment setup | One-time fee | Scoped per environment |
In a VPC Dedicated deployment the entire platform — control plane, identity, data planes, compute pools, and AI — runs inside your own AWS account. Nothing runs outside it.
Bands
Each band sets two caps. Hours are unlimited in every band.
| Band | Users | Concurrent vCPU | Included services days |
|---|---|---|---|
| Team | up to 25 | 192 | 5 |
| Business | up to 100 | 768 | 10 |
| Enterprise | up to 250 | 3,072 | 15 |
| Beyond | custom | custom | custom |
Current list prices are published on the pricing page.
How usage is measured
- User — a provisioned login in the deployment. Service accounts and API keys are not users. Someone who is provisioned but did not sign in this month still counts.
- Concurrent vCPU — the sum of vCPUs provisioned to running compute at a point in time, as configured. Stopped pools count for nothing. CPU utilization of that compute is not a licensing measure, and Spark-advertised cores are not counted; only the physical vCPU allocated to running tasks is.
- Burst headroom — every band includes 15% headroom above its concurrent-vCPU cap.
- Warnings, never blocks — the platform warns at 80% of the cap. Running workloads are never interrupted, and nothing is blocked mid-term.
- Band review at renewal — if the sustained peak stays above the band for 30 or more days in a quarter, the next band applies from the next renewal, not mid-term.
Where to see usage
- Compute pools show their hourly cost at your own AWS rates and the AutoStop savings for each workspace.
- AI Studio shows token usage by agent, model, and user, including cached tokens for models that support prompt caching.
- A quarterly usage report is produced from the platform's metering and shared with Cazpian for the renewal review.
Terms
- Annual subscription, invoiced upfront. A 3-year commitment earns 10% off and a price lock.
- Quarterly releases with a pinned upgrade window; security patches out of band; current and previous release supported.
- Add-ons: 24/7 P1 support, additional professional-services days, and custom work under a statement of work.
A fully managed, multi-tenant service is available on a waitlist basis only. Contact us if that is the option you need.