Long-running services on metal we operate ourselves.
Always-on backends, background workers, and reproducible build pipelines on infrastructure we own — predictable performance on metal we run, with private networking and autoscale on the way.
Live — running in production · Preview — built, but not yet proven for customers, and parts may not work yet · Coming soon — on the roadmap
Always-on services with health checks and one-click rollback.
Long-running services are health-checked and self-healing, and any previous build is one click away if a release goes wrong. Logs and metrics are built in, so debugging a running service doesn't mean bolting on a separate observability stack.
Scale on real load signals and down to zero when idle.
Scaling follows real signals — concurrency and CPU — and can fall all the way to zero when a service is idle so you stop paying for nothing. You set the floor and ceiling, so a traffic spike scales out within bounds you chose rather than a surprise bill. (Coming soon.)
Service-to-service traffic over an encrypted private mesh.
Services talk to each other over an encrypted private mesh with private DNS, so internal traffic never touches the public internet. There are no bastion hosts or VPN appliances to run — the network is part of the platform, not a box you operate on the side.
Queues, cron, and long jobs decoupled from the request path.
Queues, scheduled cron jobs, and long-running tasks live off the request path, with durable delivery, retries, and backoff so work isn't lost when something fails mid-flight. A job can outlive the request that scheduled it, which is what you want for anything slower than a page load. (Coming soon.)
Reproducible builds with layer caching and artifact storage.
Builds are deterministic and layer-cached, so an unchanged dependency isn't rebuilt and an unchanged step isn't re-run. Artifacts are shared across services, and the same pipeline runs locally and in the cloud — the build that passes on your machine is the build that ships.
Provisioned Postgres on our own on-box cluster.
A Postgres database is provisioned per account on our own cluster, with pgvector where the server offers it. This is in preview: provisioning is switched on per environment, so availability is not guaranteed yet.
Where do my backends actually run?
On a cloud server we rent and operate ourselves, in one region today. We run the whole stack on it — there is no third-party PaaS underneath your app.
Can services scale to zero when idle?
Not yet. Autoscale is coming soon: it is designed to scale on concurrency and CPU signals, but customer autoscale policies are not active today.
How do my services talk to each other privately?
Private networking is in preview: an encrypted private mesh with private DNS between services, with no bastion hosts or VPN appliances to run.
What databases can I use?
Managed Postgres is in preview: a database provisioned per account on our own cluster, with pgvector where the server offers it.
Are builds reproducible?
Yes — deterministic, layer-cached builds with shared artifacts across services, and the same pipeline runs locally and in the cloud.
Signups are invite-only right now — request access and we'll follow up.