● managed kubernetesv1.31 · upstream

managed k8sno wrappers,no surprises

Pure upstream Kubernetes. No forks, no proprietary 'improvements', no 3am tears. Just a control plane that doesn't fall over and a regular kubectl.

control plane free · SLA 99.95% · upstream, no forks
hyperion-prod · v1.31 · msk-a
healthy
CP
Control plane (HA)
3 etcd · auto-upgrade
free
node-01
8 vCPU
12 pods · 32 GB
node-02
8 vCPU
9 pods · 32 GB
node-03
8 vCPU
14 pods · 32 GB
$ kubectl get nodes — 3 ready · uptime 47d
upstream, no forkscontrol plane free99.95% SLAauto-upgradeGPU B300 nodeskubectl without wrappersTerraform provideringress · cert · DNS out of the boxingress · dns · csi natively
upstream, no forkscontrol plane free99.95% SLAauto-upgradeGPU B300 nodeskubectl without wrappersTerraform provideringress · cert · DNS out of the boxingress · dns · csi natively

Why another managed K8s

Most managed-Kubernetes offerings collapse into one of two camps: either a two-year-old fork with holes in RBAC, or a 'wrapper' with a murky API and its own CLI. We went a different way.

option A

Roll your own on VMs

  • Full control
  • You sort out etcd quorum yourself
  • You upgrade the version (and break prod)
  • CSI, CNI, ingress — all by hand
  • etcd backups? at your own risk
h3llo · managed

Managed Kubernetes by us

  • Pure upstream — no forks
  • Control plane free, etcd HA out of the box
  • Upgrades in one command, zero downtime
  • CSI, CNI, ingress, cert-manager — native
  • Node and pod autoscale, GPU B300, ARM nodes
  • Full kubectl access, public API, no vendor lock
option B

A 'wrapper' k8s from a local vendor

  • Support in Russian
  • Their own CLI and their own weird API
  • Version trails upstream by 2 releases
  • Paid control plane
  • Vendor lock via proprietary CRDs
● dx

One CLI, two ways to spin up a cluster

Via our CLI or Terraform — pick whichever fits. No web-only, no 'click the button'.

h3llo CLI · 30 secondscopy
# create a 1.31 cluster on 3 nodes in moscow
$ h3llo k8s create hyperion-prod \
    --version 1.31 \
    --region msk-a \
    --nodes 3:s2.standard \
    --addons ingress,cert,metrics

# in about 4 minutes:
cluster hyperion-prod is ready
kubeconfig saved to ~/.kube/config

$ kubectl get nodes
node-01   Ready   Master,Worker   8 vCPU
node-02   Ready   Worker          8 vCPU
node-03   Ready   Worker          8 vCPU
Terraform · IaCprovider
terraform {
  required_providers {
    h3llo = {
      source  = "h3llo/h3llo"
      version = "~> 0.4"
    }
  }
}

resource "h3llo_k8s_cluster" "prod" {
  name    = "hyperion-prod"
  version = "1.31"
  region  = "msk-a"

  node_pool {
    name     = "default"
    type     = "s2.standard"
    min_size = 3
    max_size = 12
  }
}
● builderlive

Build a cluster right here

Tweak the knobs — see what the cluster looks like and what it costs. Then hit 'Create' and 4 minutes later you have a kubeconfig.

hyperion-prod · v1.31 · msk-a
healthy
CP
Control plane (HA)
3 etcd · auto-upgrade · free
free
node-01
4 vCPU
7 pods · 8 GB RAM
node-02
4 vCPU
10 pods · 8 GB RAM
node-03
4 vCPU
6 pods · 8 GB RAM
Ingress / NGINXrunning
cert-managerrunning
Metrics + Grafanaoff
Loki + Promtailoff
Kubernetes version
Region
Node type
Node count: 3
autoscale: 1 → 50, you can change it later
per month
5,940/ mo
control plane — 0 ₽
Add-ons (installed automatically)
Create clusterTerraform exampleready in ~4 minutes
● use cases

What people usually run on it

01 / web
Web apps and APIs
Stateless services behind ingress, autoscaling by CPU/RPS, blue-green deploys. SLA 99.95% — your app included.
02 / data
Data pipelines
Airflow, Spark, dbt inside the cluster. CSI with snapshots, a dedicated node-pool for heavy jobs, no noisy neighbors.
03 / ml
ML inference and training
B300 Blackwell Ultra GPU nodes, GPU time-slicing, a dedicated inference pool. No queueing for resources.
04 / db
Stateful workloads
Postgres operator, Redis, Kafka inside the cluster. Local SSD CSI, anti-affinity, automatic S3 backups.
● pricing

You pay only for the nodes

The control plane is always free, no matter how many clusters you run. You pay for what actually burns resources — the worker nodes.

dev / test
One cluster, light nodes, not for prod. Perfect for sandboxes and staging.
from 1,980 ₽ / mo
  • 1 × s2.standard (4 vCPU, 8 GB)
  • Single-AZ
  • Community support
  • SLA 99% (best-effort)
popular
production
Multi-AZ, autoscale, SLA 99.95%. This is where most customers live.
from 17,520 ₽ / mo
  • 3 × m4.compute (8 vCPU, 32 GB)
  • Multi-AZ + auto-failover
  • 24/7 support, reply < 1h
  • SLA 99.95% with financial guarantee
  • All add-ons included
enterprise
Isolated clusters, dedicated hosts, FSTEC and certification on request.
on request
  • Dedicated bare-metal nodes
  • Isolated control plane
  • SLA 99.99%, RPO/RTO in the SLA
  • Personal SRE
  • FZ-152 / FSTEC / PCI DSS
● resourcesfree

Kubernetes guides and reviews

No fluff, no marketing. Real practice from our SREs and architects — grab the PDF, read it over the weekend, ship it on Monday.

All resources →
● quickstart

4 minutes from sign-up to kubectl

If our CLI is already installed — skip to step 3. If not — here's the full path.

Let's go →
1

Install the CLI

brew install h3llo/tap/cli · or curl get.h3llo.cloud | sh
2

Authenticate

h3llo auth login — the browser opens for you, the token is stored locally.
3

Create a cluster

h3llo k8s create my-cluster --nodes 3:s2.standard · ~4 minutes and you have a kubeconfig.
4

Deploy whatever you want

kubectl apply -f . — it's just Kubernetes. No new concepts.
● faq

What people usually ask

Is this really pure upstream Kubernetes?
Yes. We take releases from kubernetes.io; the only patches we apply are CVEs, and only until the same fix lands upstream. No forks, no custom CRDs, no 'h3llo-only' features.
How much does the control plane cost?
Zero. Other providers sometimes charge for the control plane (a separate tariff or an hourly rate) — we don't. You pay only for worker nodes, disks, and traffic.
How do version upgrades work?
One command: h3llo k8s upgrade <cluster> --to 1.31. Rolling upgrade, zero application downtime (assuming your PDBs are set up correctly). You can skip every other minor version.
Do you have GPU nodes, and which ones?
Yes — B300 Blackwell Ultra and H200. You can attach them to an existing cluster as a separate node-pool with taints so only ML workers land on them.
Can I pull an etcd snapshot and restore it myself?
Yes. etcd backups are available via the CLI and API in a standard format — you can restore them onto our cluster or onto your own hardware. No vendor lock.
What about certification?
Russian Software Registry, FZ-152, FSTEC (on request), PCI DSS. We provide documentation across the board and help with certifying your perimeter.
● kubernetes the way it should be

4 minutes —
and you have a prod cluster

Sign up by email. No card required. 4,000 ₽ already on your balance — enough to spin up a couple of test clusters for a whole month.

Create cluster →Documentation