Compute platform Launching soon

Cluster

Anycast edge, regional control planes, and workloads that wake on request.

Our compute layer: scheduling across regions, isolated workloads, service discovery, secret delivery, and idle capacity that costs nothing until a request arrives.

Why we built our own

Renting someone else’s platform means inheriting their cold starts, their idea of isolation, their pricing for an app nobody is using at 4am, and their timeline for fixing any of it. For a small team running many small workloads, that arithmetic never worked.

So we built the whole path ourselves — from the address a request lands on to the process that answers it.

What it does

Anycast edge

One address, announced from every edge location. Requests land at the nearest point of presence and are answered there when we can, forwarded inward when we cannot.

Regional control planes

Each region governs itself: placement, health and capacity decisions are made close to the workloads they affect, so a distant region having a bad day is not your bad day.

Placement that thinks

A scheduler decides where each workload belongs from live capacity and health signals, and moves it when the answer changes.

Isolation you can choose

Workloads run as containers or as hardware-isolated microVMs, depending on how much of a boundary the job deserves. Same platform, same deployment path.

Idle costs nothing

Workloads with nothing to do are swept out of memory and brought back on the next request. Capacity follows demand instead of sitting warm and billed.

Secrets that never land in an image

Credentials live in an encrypted store and are delivered into a workload at the moment it runs — not baked into a build, not left in a config file.

Service discovery built in

Workloads find each other by name through the platform’s own resolver. No sidecar to run, no registry to operate.

Supervised processes

Every long-running process is owned by a supervisor that starts it, watches it, restarts it and reports on it — the same way across every tier.

How we build it

  • Each tier is independently operable; one layer failing must not cascade.
  • The default state of an unused workload is asleep, not billed.
  • Isolation is chosen per workload, not assumed platform-wide.

Cluster is preparing for commercial launch. Pricing and account opening are not live yet — the terms that will govern paid use are published in advance.