Buy it, or run it
Everything Hanzo does is open source, so the honest question is not whether you could run it — you could — but what that costs. These are real numbers from real compose files.
Run it yourself, and here is the bill
The open-source stacks people reach for, priced from the compute their own compose files reserve.
| Stack | Images it runs | vCPU | GB | Compute |
|---|---|---|---|---|
| redis | redis:7-alpine | 0.5 | 1 | $8.76/mo |
| postgres | postgres:16-alpine | 1 | 2 | $17.52/mo |
| wordpress | mariadb:11 · wordpress:6-php8.2-apache | 1.5 | 1.5 | $19.71/mo |
| ghost | mysql:8 · ghost:5-alpine | 1.5 | 2.5 | $24.09/mo |
| n8n | postgres:16-alpine · n8nio/n8n:1.60.1 | 2 | 4 | $35.04/mo |
| All of them, running | $105.12/mo | |||
Compute only, from each stack’s own compose file: DigitalOcean droplet retail + Hanzo platform margin (estimate). One replica of each, at the footprint the file reserves. Storage, egress and anything with a GPU are not in it — serving a model is a GPU, and no compose file above asks for one.
And the part that is not a line item
The machines are the cheap half. This is what you are holding once they are paid for.
- Upgrades
- Every image above has a release cadence, and a CVE in any of them is yours to notice and roll.
- Backups
- Taking one is a cron. Proving a restore works is a rehearsal, and it is the half that gets skipped.
- Availability
- One replica of each is a single point of failure. A second is roughly twice the compute plus the state problem.
- Identity
- Each of these has its own users and its own idea of an admin. Joining them is a project.
- On-call
- Someone answers when it pages at 3am, and that person is on your payroll.
We have not costed that list, and will not: an invented engineer-hours figure would be worth less than the arithmetic you can do yourself, knowing your own team. What Hanzo replaces is the whole column — one identity, one bill, one thing that pages someone else.