5eves: Managed WordPress Hosting on Kubernetes

The challenge

• Provisioning a real, isolated WordPress site on demand is a very different problem from deploying one site: it has to happen without a human touching the cluster.
• Multi-tenancy means one customer’s site must not be able to reach another’s, while still sharing infrastructure economically.
• Every provisioned site generates credentials that have to be stored and retrieved without ever sitting in plaintext.
• Selling capacity you do not have is how hosting platforms fail; capacity has to be enforced in code.
• Running WordPress behind a proxy introduces HTTPS detection problems that surface as admin redirect loops rather than as clear errors.

What we built

• Built on Kubernetes, with the whole platform defined as versioned manifests rather than hand-configured servers.
• An API service provisions customer sites programmatically against the cluster, so a new site is an API call rather than a support ticket.
• Studio, the AI-assisted theme builder, runs as its own service with persistent storage, so design work survives restarts.
• Shared Redis and MySQL services back the tenant sites, keeping per-site cost down without giving tenants access to each other.
• Every stored site credential: SSH, SFTP and the rest: is encrypted at rest behind a platform key.
• A capacity gate governs how many sites can be sold against real cluster capacity, enforced in configuration rather than tracked by hand.
• Traefik handles ingress, with HTTPS forced explicitly at the application layer because the proxy resets the header WordPress would otherwise rely on.

The result

• A working multi-tenant hosting platform running isolated WordPress sites on Kubernetes.
• Site provisioning is automated end to end through the API.
• AI-assisted theme building available to customers as part of the product.
• Credentials encrypted at rest; capacity enforced rather than estimated.

What this means for you

Automating a task takes a person out of it, so the code has to enforce the rules that person would have followed: tenants kept apart, credentials encrypted, and no more capacity sold than exists.

Faults behind a proxy often look like something else. A reset HTTPS header shows up as an admin redirect loop rather than a clear error, which is why HTTPS is forced at the application layer.

Defining the platform as versioned manifests means it is changed through reviewed files rather than by hand on a server.