For the apps your coding agent builds

Your agent built the app. Let it ship and run it.

Kubrain is a deployment platform your coding agent operates — every deploy, secret, scale, and rollback is one tool call from Claude, Cursor, or the CLI. It starts as simple as a PaaS and grows into real multi-service production without a re-platform. And when the agent is out of its depth, real DevOps engineers pick up the incident with the full context attached.

Driven from Claude, Cursor, or the CLI
One node to HA — no re-platform
AI healing, real engineers behind it
agent session · kubrain-mcp
▸ you
What you'd do otherwise

You've outgrown the one-click PaaS. You don't want to become a DevOps team.

The day your app needs more room than one box, the easy path runs out. The usual next move is a VPS you now own forever, or managed Kubernetes plus a pile of HCL your agent guessed at and can't safely apply. Same task on Kubrain: one sentence your agent already understands, on a platform where the resources are built in and the state comes back as something it can read.

PaaS ceiling → your own servers
≈ a weekend of yak-shaving · and you own it from now on
Kubrain
One operation, three ways — MCP, CLI, or native IaC. Zero tabs, fully reproducible.
Fits your workflow

Made for vibe-coded projects.

You built the app by describing it. Ship it the same way. Point Claude, Cursor, or your own agent at Kubrain and let it handle the infra.

Native Infrastructure as Code you never hand-write — just Infrastructure as a Prompt.

agent session · kubrain-mcp self-documenting infra state
Ask “what’s running?” and get the real answer, live from state — as a declarative blueprint you can edit and kubrain apply back. That’s the native IaC: no stale wiki, no drift, no Terraform to reconcile.
Built-in Skills — your agent already knows the right way.
Kubrain ships opinionated Skills with every tool — the correct, simplest recipe for each task baked in. Your agent doesn't burn tokens rediscovering how to size a cluster, wire a database, or set up TLS from scratch; it calls the Skill and gets it right the first time. Fewer round-trips, less context, best practices by default — so you spend less on the model and less time waiting.
No re-platform, ever

The deploy that doesn't dead-end.

Every simple deploy platform is great until service number two. Add a background worker, a cron job, and a queue and you're stitching together five products that were never meant to talk to each other — or you're migrating the whole thing to somewhere that can hold it.

Kubrain doesn't put you in that position, because underneath it's Kubernetes: services, jobs, storage, and networking under one API from day one. You just never pay the operational tax for it — that part is ours. Start on a single node, split into services when you're ready, go highly available the day it matters. Same cluster, same tool calls, nothing to move.

Dev 1 node
A single-node cluster for prototypes, side projects, and preview environments. Cheapest way to run real Kubernetes.
Most projects
Normal 3+ nodes
A production cluster with room to breathe. Spread your services across nodes, autoscale on load, survive a node failure.
AI auto-healing included
HA multi-zone
High availability: a replicated control plane and workers across zones. For when downtime costs real money.
AI auto-healing included
Grow with one call — scale_cluster(name="prod", workers=8) or resize_cluster(name="prod", ram=16) — no rebuild, no downtime.
AI auto-healing on Normal & HA. An agent watches your cluster around the clock — when something breaks it diagnoses, tries the safe, known fix (restart, roll back, rebalance, scale), and verifies recovery. If it can't fix it confidently, it escalates to the DevOps team with the full context attached, so a human picks up exactly where the agent left off. You wake up to a resolved incident, not a pager.
Vanilla, upstream Kubernetes — always current. No forks, no proprietary API, no lock-in. Your manifests, Helm charts, and kubectl work unchanged, and we track upstream releases closely so you're never stuck on an old version.
Batteries included, best practices by default.
The pieces every cluster needs come preconfigured and wired together the right way — so you don't spend a weekend on YAML you'd rather not think about. Turn any of them on with one tool call.
Ship from anywhere

Deploy from your laptop or your repo.

Push a local folder straight up, or connect a Git repo and let every commit build itself. Kubrain builds the image in the cloud, ships it to your cluster — and if something looks wrong, one call rolls you back to the last good release.

Encrypted secrets, injected at runtime. Store passwords, API keys, and connection strings with set_secret(name="myapp", key="DATABASE_URL", value_file=…) — the value is read straight from a file, so the real secret never enters the agent's context. Encrypted at rest, never printed in logs or transcripts, and mounted into your pods as env vars at deploy time. No plaintext in your repo, no .env files to leak.
See what's running

Monitoring, built in — not bolted on.

Every cluster ships with metrics, logs, and alerts on day one. No Prometheus to install, no Grafana to wire up, no exporters to babysit. Your agent reads it the same way it deploys — one tool call — so it can spot a problem, pull the logs, and fix it without you opening a dashboard.

agent session · kubrain-mcp
▸ you
how's the api service doing?
kubrain ⏎get_metrics(svc="api", window="1h")
CPU 34% · Mem 61% · p95 128ms
req 1.2k/min · 5xx 0.02%healthy
kubrain ⏎tail_logs(svc="api", filter="error", lines=3)
14:02 ERROR db timeout on /orders
14:02 WARN retry 1/3 succeeded
kubrain ⏎set_alert(svc="api", metric="5xx", above="1%", notify="slack")
✔ alert armed · pings #ops on breach
Your data stays yours

Scheduled backups. No lock-in.

Turn on automatic backups of your databases and volumes on the schedule you choose, kept for as long as you set. And whenever you want, download everything — data and the full stack definition — in a standard, portable format. Leave any day, take it all with you.

agent session · kubrain-mcp
▸ you
back up prod nightly, keep 30 days
kubrain ⏎schedule_backup(cluster="prod", cron="0 3 * * *", retain="30d")
✔ nightly backups armed · 03:00 · 30-day retention
▸ you
give me everything to take with me
kubrain ⏎export_all(cluster="prod", include="data,stack")
→ bundling volumes + DB dumps + manifests…
✔ prod-export.tar.gz ready · signed download link
Humans, when you need them

A real DevOps team on tap.

Stuck on a migration, a weird networking bug, or a production incident at 2am? Book time with the engineers who run the platform. Pay by the hour — no retainers, no seats.

Expert DevOps hours pay as you go
€60 / hour
Your agent can hand a problem straight to a human — one tool call, at the severity that fits.
agent session · kubrain-mcp
Pricing

Simple prices

No catalogue of a hundred services to decode. Compute is priced per GB of RAM — CPU scales with it automatically. Volumes and S3 are a flat rate per GB. That's the whole bill, and you can predict it to the euro before you deploy.

Same rates whether you run one pod or a hundred. Expert DevOps time is the only extra — billed by the hour, only when you ask.
Ask your agent what it'll cost — before you deploy.
agent session · kubrain-mcp
▸ you
what will a normal 3-node cluster cost per month?
kubrain ⏎
Compute · 3 × 4 GB RAM · 730h
  12 GB × €0.006 × 730h€52.56
Block volume · 80 GB (first 50 free)
  30 GB × €0.05€1.50
S3 · 60 GB (first 20 free)
  40 GB × €0.005€0.20
Load balancer€3.50
Traffic + DNSfree
≈ €57.76 / monthpredictable
→ fixed unless you change the cluster · no metered surprises

Create your account.

Sign up in seconds, point your agent at it, and ship. Nothing to migrate when the project grows — and if you ever want out, export everything, data and full stack definition, and take it with you.