Your cloud, grown in Sheffield.

Most clouds ask you to trust a brand. We would rather show you the kit, name the software, and tell you what is finished and what is not. Forge runs on machines we own, in Sheffield, looked after by the engineer who answers your email.

Forge, layer by layer

  1. Edge and entry pointbunny.net in front, Pangolin at the door. No ports open inward.
  2. Network coreMikroTik routing, VLANs and firewall, all as code.
  3. Compute: three gen nodesProxmox hosts, a Kubernetes control plane and workers on each.
  4. Storage: Ceph, three copiesReplicated across the nodes. A lost machine loses nothing.
  5. AI node and batteriesAn NVIDIA DGX Spark for inference. Batteries, not a diesel generator, carry the cluster through a power cut, carbon-free.

Every part is named and credited on the "What it runs on" page.

Hardware

Our own machines

Brocco, our cluster in Sheffield, runs Proxmox and Kubernetes on servers we own, with Ceph storage, an NVIDIA DGX Spark for AI and a MikroTik network core. Batteries carry it through a power cut and shut it down cleanly if the mains does not come back.

Software

Open source, named

Every part of Forge carries a credit line to the project it is built on. You can look any of them up. Nothing is a black box.

Traffic

Already serving

Live websites are served through Forge today, from our edge over a block of internet addresses leased to us.

Where we are

What is live, and what is not yet

We would rather you knew exactly where things stand. This list is kept current, and we will say so when something moves.

Live today

Kubernetes
Production workloads on Brocco today.
Websites
WordPress and Ghost sites, with the bunny.net CDN in front.
Databases
PostgreSQL and MySQL, managed by us and backed up automatically.
Email sending
Outgoing email for our sites and services through our own Stalwart relay.
Automation
Infrastructure as code, applied by pipelines that check every change.
Observability
Prometheus, Grafana, Elasticsearch and Kibana across the platform.
AI inference
An open model running on our DGX Spark for our own tools.

In progress

Zero-trust
Access granted per person and per service across every machine and service.
Defence in depth
Clients already run in their own isolated spaces. We are adding further layers, so that no single control is the only thing between one client and another.
Dual-site high availability
A second site that holds an encrypted, automatically kept copy and can take over if Sheffield goes dark. Today the off-site copy is made by hand.
Forge sign-in
One "Sign in with Forge" account across the apps you use with us, built on Keycloak. In testing now.
Diagnosis
Alerts diagnosed with help from a model on our DGX Spark.
Further automation
Extending what runs itself, and what puts itself back after a power or network cut.

Why Forge

Four things we are building it for

Local data

Your data is on machines in Sheffield, run by a company in Sheffield. If you want to see them, you can get the tram.

Open source

Built from software anyone can read. No lock-in - if you leave, you take your data with you in standard formats.

Light on power

A cloud should be efficient as well as local. We have cut Brocco's power draw from about 500 W to about 200 W and measure it all the time. We will publish the numbers, rather than put a leaf on a logo.

Self-healing

Storage is spread across several machines, settings live in Git, and the platform is being built to put itself back when something breaks. A chain of batteries bridges power cuts.